Management of Partially Safe Buffers

Size: px
Start display at page:

Download "Management of Partially Safe Buffers"

Transcription

1 394 EEE TRANSACTONS ON COMPUTERS, VOL. 44, NO. 3, MARCH 1995 Management of Partay Safe Buffers Sedat Akyrek and Kenneth Saem, Member, EEE Computer Socety AbstPactAhfe RAM s RAM whch has been made as reabe as a dsk. We co1sder the probem of buffer management n partay safe buffers,.e., buters whch contan both safe RAM and voate RAM. Buffer managemnt technques for partay safe buffers expcty consder the safety of memory n decdng whch data to pace n the buffer, where to pace t, and when to copy updaks back to the dsk. We present technques for managhg such buffers and study ther performauce usng trace-drven smuatons. ndex Terms-Buffer management, dsks, operatng systems, performance, reabty, safe RAM.. NTRODUCTON A% RAM s RAM whch has been made as reabe as a dsk. S Safe RAM can be mpemented usng batteres or unnterruptbe power suppes pus redundancy for error correcton and detecton. Computng systems, such as transacton processng systems and operatng systems, often guarantee the durabty of modfed data by propagatng t to dsks. Because t can be updated much more qucky than dsks, safe RAM can greaty reduce response tmes for data updates. Safe RAM can aso mprove performance by reducng the tota number of dsk wrtes that must eventuay be performed. Updated objects can be buffered temporary n safe RAM n case they are updated agan. Shoud ths occur, two (or more) dsk wrtes w have been combned nto one when the object s fnay wrtten to the dsk. The dsadvantage of safe RAM s ts cost. A recent study [] found that non-voate RAM was about fve tmes as expensve as voate RAM, athough ths cost dfferenta s key to drop over tme. Gven that some cost dfferenta exsts, purtdy safe buffers make sense. Partay safe buffers are mpemented usng a mx of safe RAM and voate RAM. We note that when speed rather than voatty s consdered, a smar stuaton arses. Snce faster memores are more expensve, system desgners resort to a memory herarchy,.e., a mx of fast, expensve memory and sower, cheaper memory. A varety of exstng systems aready mpement partay safe buffers or assume ther exstence. These ncude both man memory buffers, e.g., n the POSTGRES storage system [17], and controer buffers such as the partay safe buffer mpemented n the BM 3990 storage controer [ 81. Ths paper addresses two ssues. Frst, we consder the management of partay safe buffers, wth emphass on repacement poces. Common (and effectve) repacement Manuscrpt receved Dec. 19, 1991; revsed Aug. 26, Ths work WBS sumxxted.. bv NSF Grant No. CCR and bv CESDS. S. Akyhk wth the Department of Comp&r Scence, Unversty of Maryand, Coege Park, Maryand K. Saem s wth the Computer Scence Department, Unversty of Wateroo, Wateroo, Ont., N2L 3G1, Canada, e-ma kmsaem@uwateroo.ca. EEECS Log Number C poces, e.g., east-recenty used (LRU), are bnd to the safety or voatty of the buffer. For exampe, the LRU pocy mght pace an updated object n the voate buffer f that s where the east recenty used object happens to resde. n a partay safe buffer, repacement decsons shoud take the safety of the buffer nto account. An exampe of a pocy whch takes safety nto account s one whch makes a read repacements n the voate buffer and a update repacements n the safe buffer. We present four canddate repacement poces and evauate ther performance usng trace-drven smuatons. Our evauaton consders the foowng factors: Sze of the Safe BufSer. Performance can be expected to mprove as the safe buffer sze s ncreased. However, the margna benefts of addtona buffer space w decrease wth the buffer sze. Furthermore, the buffer sze affects dfferent buffer management poces dfferenty. Workoud Characterstcs. Workoad characterstcs, such as read/wrte ratos, burstness of request arrvas, and request ocaty, affect the performance of the buffer. Our study consders a varety of workoads wth dfferng characterstcs. Update Stagng. When a drty object s repaced n a buffer t must be wrtten back to the dsks, causng a deay. To reduce these deays, drty objects resdng n the safe buffer can be wrtten asynchronousy to the dsks. Ths s known as stagng. Stagng can reduce response tmes f dsk utzaton s not too hgh. Stagng aso affects the reatve performance of the repacement poces. The second goa of ths paper s to compare partay safe buffers wth voate buffers. Wrte-through voate buffers, whch must guarantee the safety of every update as t occurs by forwardng t to a dsk, w ceary be outperformed by a partay safe buffer. Our study quantfes ths performance gap. A more nterestng comparson can be made between partay safe buffers and voate buffers n whch unsafe updates can be permtted, at east temporary. We w refer these buffers as copy-buck buffers. Copy-back buffers are common n operatng systems and database management systems. Copy-back buffers share many of the advantages of partay safe buffers. Copy-back buffers remove dsk /O from the crtca path of update requests, resutng n much faster response tmes. Furthermore, by bufferng unsafe updates n the buffer, t may be possbe to combne severa updates to the same bock or page nto one, resutng n mproved wrte throughput. Copy-back buffers aso have a dsadvantage: Appcatons must be wng to toerate some ost work n the event of a faure. For exampe, updated data n a Unx fe system buffer s not fushed to dsk mmedatey. nstead, buffered updates are synchronzed (.e., coped back to the dsks) /95$04.00 Q 1995 EEE

2 AKYjREK AND SALEM: MANAGEMENT OF PARTALLY SAFE BUFFERS 395 perodcay. A faure may cause updates that occur after the most recent synchronzaton pont to be ost. n database management systems, buffer managers that permt unsafe updates are sad to use a TFORCE [7] pocy. Such buffer managers are often used n conjuncton wth a separate mechansm, such as after-mage (REDO) oggng [7] and perodc checkpontng, to guarantee the safety of a updates. The combnaton of oggng, checkpontng, and a -FORCE voate buffer provdes advantages smar to those of partay safe buffers: Wrte response tmes are fast and the safety of updates s guaranteed. However, a partay safe buffer can provde the guarantee more smpy, snce no REDO oggng or checkpontng s necessary. Furthermore, checkpontng mechansms ntroduce a tradeoff between faure recovery tme and overhead durng norma operaton. (More frequent checkponts reduce recovery tme at the expense of addtona overhead.) Partay safe buffers requre no such tradeoff. Our performance study compares partay safe buffers and voate copy-back buffers. Partay safe buffers are the superor aternatve f they can provde compettve request response tmes, snce they do not ose updates to faures. n database management systems, partay safe buffers can be vewed as smpe aternatves to REDO oggng and checkpontng. A. Reated Work A number of exstng or proposed systems aready empoy safe RAM to mprove performance. As we have aready noted, the BM 3990 dsk controer [8] s capabe of bufferng up dates n ts non-voate memory. Both voate and nonvoate buffers are mantaned, and updates appear n both. Drty data n the safe buffer s staged to dsk asynchronousy when possbe. XPRS [18], a POSTGRES-based transacton processng system, buffers frequenty updated data bocks n safe RAM to provde fast recovery. The technques used to manage the safe buffer are not descrbed there. The Phoenx fe system [SJ mantans an entre fe system n safe RAM. n [3], the use of partay safe buffers n transacton processng systems s dscussed. An anaytc queueng mode s used to estmate the sze of the safe buffer requred to avod synchronous dsk wrtes, when asynchronous stagng s used. (Synchronous wrtng s requred when the safe buffer becomes fed wth drty data.) However, that work does not consder buffer management, whch s the focus of the mode n ths paper. Our mode ncorporates the mpact of the safe buffer and the buffer management poces on read response tmes as we as wrte response tmes, and s based on trace-drven smuatons rather than an anaytc mode. Stochastc smuatons descrbed n [12] compare severa aternatve safe RAM confguratons n database systems. Safe extended man memory, sod state dsk, and safe dsk (controer) buffers are consdered. That work uses a dfferent safe RAM mode than ours. n [ 121, read-referenced data that s ocated n the safe extended man memory must be transferred to voate man memory to be read. Furthermore, drect transfer of data between the safe RAM and the dsks s not possbe. The focus of that work s on comparng the atematve safe RAM confguratons, and not on buffer management. Two recent studes have used trace-drven smuatons to evauate safe buffers n operatng systems. One study [] was based on traces from the Sprte fe system, whe the second used traces from Unx fe systems [ 131. The traces used n [ 131 are bock eve traces smar to our own, whe the Sprte traces are captured at a hgher eve. n partcuar, the Sprte traces do not ncude requests for fe system meta-data, such as -nodes [9]. Such references account for a substanta fracton of the requests (partcuary wrte requests) n our traces andnthoseof[13]. Both of these studes used a safe RAM confguraton caed wrte-asde, whch we dd not consder n our study. Under ths confguraton, the safe RAM handes a update requests, but t cannot be used to satsfy read requests. Ths confguraton s smar to that mpemented by the BM 3990 dsk controer and the mode studed n [3]. The Sprte study aso consdered a unfed confguraton smar to the one used n ths study. Under the unfed mode, read requests can be satsfed from ether the safe buffer or the voate buffer. n the Sprte study, the unfed partay safe buffer was managed usng a pocy smar2 to the L.RU GobuUWrte Purge pocy descrbed here. Other poces were not consdered. The Sprte study concuded that the unfed confguraton provded better overa performance. Both of these studes showed that sma amounts of safe RAM can provde sgnfcant reductons n wrte traffc. Ths genera resut corroborates one of our own. Nether of these studes consdered the effects of asynchronous stagng from the safe buffer, presumaby because ther prncpe performance metrc was the reducton n wrte traffc due to the safe buffer. (Stagng s a technque for mprovng response tmes at the expense of ncreased traffc behnd the cache.) Nether study evauated poces for managng both read and wrte re- quests under the unfed partay safe cache confguraton. Ths s the prmary focus of our study. Severa technques have been suggested for mprovng the dsk performance, assumng that updates are temporary buff- ered. Safe RAM s dea for use n combnaton wth these technques, snce buffered updates w not be ost n the event of a faure. Ng [ 111 suggests that safe RAM can be used to emnate the wrte penaty assocated wth dupexed dsks. Buffered wrtes can aso be pggy-backed onto read operatons [ 161, aowng the dsk to be updated wth very tte cost. Ether of these technques can be combned easy wth the buffer management technques descrbed n ths paper. However, we do not consder such extensons here. Fnay, severa proposed memory-resdent transacton processng system desgns [4], [6], [o] rey on safe memory to commt transacton updates qucky. Sma safe memores have aso been used to provde fast recovery n the Sprte fe system [2]. n those systems, safe memory management s ted cosey to the transacton manager or the fe system. n contrast, the technques descrbed n ths paper are not ted to the semantcs of any partcuar data manager. Actuay, ony the safety of cornmned updates s guaranteed. 2 The pocy n [] copes data from the safe cache to the voate cache n some stuatons. The poces descrbed here do not.

3 396 EEE TRANSACTONS ON COMPUTERS, VOL. 44, NO. 3, MARCH 1995 BEFORE REQUESTS SLOW SAFE MEMORY / j MEMORY PARTALLY-SAFE BUFFER Fg. 1. Storage mode. 1 ~ Read Object G n the next two sectons we present severa buffer management technques for partay safe buffers, our smuaton mode, and the traces that drve t. A comparson of the buffer management technques based on the smuaton mode s presented n Secton V. Secton V descrbes how asynchronous stagng can be used to further mprove the performance of the partay safe buffer. Fnay, n Secton V, we compare partay safe buffers wth voate, copy-back buffers.. MANAGNG PARTALLY SAFE BUFFERS We have used a smpe mode to study partay safe buffers. The storage system s assumed to manage a coecton of fxed-sze objects, each wth a unque dentfer. (Objects may correspond to pages or fe bocks n a rea system.) Three types of storage are avaabe to hod them. Sow, safe storage (e.g., dsks) mantans a copy of every object. A faster buffer aso exsts, and s dvded nto safe and voate parts. Copes of some of the objects resde n the buffer. The mode s ustrated n Fg. 1. The system s oad conssts of read and wrte requests. The buffer manager must satsfy these requests by observng the foowng rues: Read Rue: When an object s read, a copy of the object shoud be buffered (ether safe or voate). Update Rue: When an object s updated, an updated copy shoud resde n safe memory (ether fast or sow). Later we w consder voate copy-back buffers, whch do not observe ths rue. We have consdered two bnary dmensons aong whch to cassfy buffer management technques, gvng rse to four casses of buffer managers. The frst dmenson defnes the buffer manager s behavor n the event of a read mss. The second defnes ts behavor n the event that an updated object s aready n the voate buffer. A. Read Mss Pocy Accordng to the read rue, read requests are satsfed mmedatey f the requested object s n (ether part of) the buffer. A read mss occurs f the object s not n ether buffer, n whch case t must be read from the dsk. We have consdered the foowng two poces for handng read msses. SAFE BUFFER AFTER (LRU Voate) AFTER (LRU Goba) Fg. 2. Read mss poces compared. The fgures ustrate the effect of a read mss under the two read mss poces. t s assumed that both the safe and voate buffers axe fu, so that the read mss must cause a repacement. The chans of arrows ndcate recency of use, wth A + B ndcatng that A s more recenty used than B. Under the LRU Voate pocy, object G repaces object E, snce E s the east recenty used object n the voate buffer. Under L.RU Goba, object G repaces object F n the safe buffer nstead, snce F s the east recenty used overa. LRU Voate. Ths pocy specfes that mssed reads cause a repacement of the east recenty used object n the voate buffer. Of course, f free space s avaabe n the voate buffer, repacement s not necessary. LRU Goba. Ths pocy specfes that mssed reads cause a repacement of the east recenty used object n the whoe buffer. Under ths pocy, t s possbe that read repacements w be performed n the safe buffer. Fg. 2 ustrates the behavor of two read mss poces. The LRU Goba pocy s partcuary appeang when the voate porton of the buffer s reatvey sma, because a read requests need not be channeed through the voate buffer. B. Wrte Aocaton Pocy So that the wrte rue can be satsfed wthout wrtng through to the dsk, updated bocks are aways paced n the safe buffer. f necessary, the east recenty used object n the safe buffer s repaced to make room. The wrte aocaton pocy specfes the behavor of the voate buffer n the event of an update. Ths pocy s necessary n case an object s n the voate buffer when t s updated. We have consdered two wrte aocaton poces.

4 391 Fg. 4. The smuaton mode. buffer ht (no dsk /O) Update Object D BUPFERSZESARE TABLE SPEC~EDWA~TOTHEREFERENCESETSZE,WHCH ~TRACE-DEPENDENT.THEREFERENCESETSZESFOREACHTRA~ARE GVENNTABLB. SAFE BUFFER VOLATLE BUFFER SAFE BUFFER VOUTLB BUFFER 1 Parameter 1 Defaut Vaue 1 AFTER (Wrte Purge) AFTER (Wrte Update) Fg. 3. Wrte aocaton poces compamd. The fgures ustrate the effect of an update under the two wrte aocaton poces. The chans of armws ndcate recency of use, as n Fg. 2. Under the W&e Purge pocy, the update to D causes t to be removed from the voate buffer. Object D s paced n the safe buffer, repacng F, the odest safe object. (We have assumed here that the safe buffer s fu.) Under Wrfe Update, object D s updated n the voate buffer and remans there after the update. Note that a subsequent read mss w cause object E to be repaced n the voate buffer n ths case. Under the Wrte Purge pocy, E woud not have to be repaced, snce a free sot s avaabe n the voate buffer. Wrte Purge. Ths pocy specfes that objects are deeted from the voate buffer when they are updated. The space occuped by the deeted object s marked free, and s avaabe to hod a new objectswhen the next read mss occurs. Wrte Update. Ths pocy specfes bat objects n the voate buffer reman there f they are updated. The object s up dated n the voate buffer to refect ts new vaue. Snce up dated objects are aways paced n the safe buffer as we, ths pocy may resut n two copes of an object beng buffered smutaneousy, one safe and the other voate. An object updated n the voate buffer s not consdered to have been!used by the update. Thus, f the east recenty used object n the voate buffer s updated, t w st be repaced when the next read mss occurs (assumng that the Z&U Voate read mss pocy s beng used). Fg. 3 ustrates the two wrte aocaton poces. n [15], a thrd pocy, caed Wrte Aocate, s dscussed. Under ths pocy, updated objects woud be nstaed n the voate buffer f they were not aready there. (The object woud be updated n pace, as under the Wrte Update pocy, f t was aready n the voate buffer.) Ths pocy does not seem approprate for the voate buffer, snce updated objects are automatcay n- staed n the safe buffer anyway. We do not consder the Wrte Aocate pocy further n ths pap&. LTHESMULATON MODEL We have deveoped the smuaton mode ustrated n Fg.4 to study the buffer management technquks descrbed n the prevous secton. The smuator s drven by traces of read and wrte requests. Each trace request (descrbed n more deta beow) ncudes a bock number, a read/wrte fag, and an arrva tmestamp. The arrva process for the buffer s determned by the trace request tmestamps. At each request s arrva tme, the smuator determnes whch bocks must be moved between the buffer and the dsks, accordng to the read and update rues, the buffer management pocy beng mpemented, and the buffer sze. A snge request resuts n as few as zero and as many as two dsk operatons. One dsk operaton may be requred to brng the requested bock nto the buffer. A second operaton may be requred to wrte a repaced, drty bock back to the dsk. f two dsk operatons resut from a snge request they are ntated sequentay, as ustrated by the oop n Fg. 4. The response tme of a request s the sum of the response tmes of the dsk operatons that t generates. Requests that generate no dsk operatons are defned to have a response tme of zero. Dsk servce tmes are assumed to be exponentay dstrbuted, and requests are served n FFO order. We consdered a more eaborate dsk mode n whch seek tmes were cacuated usng the bock numbers of successve requests and addtona assumptons about the dsk s geometry. Whe such a mode woud probaby resut n more accurate absoute servce tmes, we fet that t was unkey to have a strong mpact on the reutve performance of the varous buffer managers. For the sake of smpcty, we eected to use the smper exponenta mode. The smuaton parameters and ther defaut vaues are summarzed n Tabe. The smuator reports a varety of statstcs for each run, ncudng counts of /O operatons, mean

5 398 EEE TRANSACTONS ON COMPUTERS, VOL. 44, NO. 3, MARCH 1995 TABLE REFERENCE TRACE SUMMARY. Trace Trace Number of Number Numbers Reference Set Num. Records/ Name be Records of Reads of Wrtes Sze (RSS) RSS servera post-bufer (70%) (30%) cent A post- buffer (41%) 4889 (59%) serverb pre-buffer (83%) (17%) centb pre-buffer (82%) 3878 (18%) response tmes for read and wrte requests, and utzaton, servce tme, and watng tme at the dsk server. t s mpemented usng the CSM smuaton brary [ 141. A. The Truces The traces were gathered from workstatons runnng a customzed verson of the SunOS 3.2 operatng system kerne. The kerne was modfed to produce a trace record for each bock /O request (read or wrte) to fe systems resdng on the workstaton s dsk(s). Trace records are deposted n a kerne buffer, from whch they can be read by a user-eve process usng a speca pseudo-devce drver. Requests destned for partcuar fe systems can ater be ftered from the traces f desred. Each trace record ncudes a unque dentfer3 for the requested data, the sze of the request, a read or wrte tag, and some addtona peces of nformaton. Each record aso ndcates whether or not the request ht the /O buffer cache on the traced machne. Because of the way the tracng facty s mpemented, each wrte request n our traces represents a request to fush a bock from from the fe system buffer to dsk, rather than the actua modfcaton of the bock n the buffer. n Unx, these bock fush requests are often generated by a speca synchronzaton process whch perodcay fushes drty data from the buffer. Thus, the bock fush request may appear n the trace some tme after the bock was actuay updated n the buffer. Because the synchronzaton process s perodc, t aso means that many of the bock fushes occur n perodc bursts n the reference stream. Rather than attemptng to guess the actua bock update tmes, we used the traces as s, wth the bursty wrte request pattern, to drve our smuatons. Fortunatey, the (smuated) safe buffer tends to smooth out the effects of these bursts anyways. The prmary effect of ths decson s that the wrte response tmes (under a of the buffer management poces) are hgher than they woud otherwse be when the safe buffer s very sma. Where ths effect s apparent n our smuaton resuts n the next secton, we have been carefu to pont t out. Two types of traces were used n our experments. A prebuffer trace ncudes a /O requests, ncudng those that ht the traced system s buffer cache. Such a trace s representatve of the request streams seen by a man memory buffer, such as a fe system s bock buffer. The request response tmes we 3 An dentfer conssts of major and mnor devce numbers and a bock number. 4 Every 30 seconds. measure usng these traces represent tmes that an appcaton program observes as t makes requests to a fe system wth a partay safe buffer. A post-buffer trace ncudes ony those /O read requests that mssed the buffer cache on the traced machne, pus a wrte requests. These traces are representatve of the request streams that mght be seen by a storage controer wth a partay safe buffer (such as the BM 3990), snce they ncude ony those requests that fe through the buffer on the traced machne.5 The request response tmes we measure usng these traces represent tmes that the fe system observes as t makes requests to an underyng storage devce equpped wth a partay safe buffer. Requests n the traces vary n sze from 1K bytes to 8K bytes, wth the vast majorty of the requests beng for 8K bytes. For the purposes of these smuatons, a requests were treated as 8K requests,.e., when a sma bock s requested, the arge bock that t s a part of s requested nstead. Thus, the 8K byte bocks, each of whch has a unque dentfer, are the database objects of our mode. Each trace covers a perod of about 12 hours, startng n md-mornng, on a weekday. Traces were taken from two workstatons. One of these serves as a network fe server. The other s a cent workstaton wth ts own prvate dsk. Each of the four traces was recorded on a dfferent day. The traces from the server record references to a snge arge fe system housng prmary shared data and executabe fes. The cent traces trace references to the oca fesystem on the cent workstaton. Ths fe system hods prmary user fes. We expermented wth a pre-buffer and a post-bu#er trace from each of the workstatons. Some of the trace characterstcs are are summarzed n Tabe. The reference set sze refers to the tota number of objects referenced at east once n the trace. Ths number s much smaer than the tota number of objects stored on the dsk, snce many objects are not referenced at a. The ast coumn of the tabe gves a crude ndcaton of the ocaty present n the request stream. Cumuatve read request nter-arrva tme dstrbutons for each of the traces are shown n Fg. 5. nter-arrva tmes were measured wth 20 msecond resouton, as ths was the resouton of the system cocks on the traced machnes. n genera, the post-b&r traces on both the cent and the server have a hgher percentage of updates. The cent traces exhbt ess reference ocaty than those of the server. Fnay, 5 The cent and server workstatons from whch the traces were recorded each have buffer sxes of approxmatey 2 megabytes.

6 AKYREK AND SALEM: MANAGEMENT OF PARTALLY SAFE BUFFERS % 70% 50% 30% 10% c / Km Read Request nter-arrva Tme (mseconds) Fg. 5. Cumuatve trace request nter-arrva tme dstrbutons. read requests n the cent traces have burster nter-arrva tmes than those n the server traces. Ths can can be seen from the fatness of the cent trace dstrbutons (especay centb) n Fg. 5. These curves ndcate that amost a nter-arrva tmes are ether very short (ess than 100 mseconds) or very ong (greater than one second). The probabty of nter-arrva tmes greater than one second s at east twce as great n the centb as n ether server trace. V. PERFORMANCE OF THE BUFFER MANAGEMENT POLCES We performed severa experments to determne the performance of the buffer manager under varous combnatons of the read mss and wrte aocaton poces. Of prmary nterest n our frst experment s the effect of the safe buffer sze on performance. n ths experment, the szes of the safe and voate buffers were vared, whe keepng the tota buffer sze (safe pus voate) fxed at 10% of the reference set sze for each trace. Ths aows us to dstngush the effects of changes n safe buffer sze from the effects of changes n the tota buffer sze. Reference set szes for each trace are gven n Tabe. Fgs. 6 and 7 show the mean wrte and read response tmes for each of the four buffer management poces. Fg. 6 shows that a reatvey sma safe buffer, 3% or 4% of the reference set sze, s suffcent to reduce the mean wrte response tme to just a few mseconds for the server traces. (The hgh wrte response tmes for very sma safe buffer szes s prmary an artfact of the wrte bursts n our traces.) The reducton was not as sgnfcant for the cent traces for reasons whch we w dscuss shorty. For comparson, Tabe gves the mean read and wrte response tmes acheved by an a-voate buffer of the same tota sze. Wrte response tmes for the a-voate buffer are hgh because a updates are wrtten through the buffer to the dsk to ensure ther safety, and because of the wrte bursts. Read response tmes are hgher for the cent traces because read requests are burster n those traces. The buffer management poces do not have a very strong effect on wrte response tmes. The LRCJ Goba pocy does resut n somewhat ower wrte response tmes than LRU Voate for the cent traces when the safe buffer s arge. Ths s because of cean data paced n the safe buffer by the LRU Goba pocy. (We dscuss ths further n Secton 1V.C.) However, both buffer management and buffer szes do have an mpact on read response tmes, as Fg. 7 ustrates. Furthermore, the behavor we observe s trace dependent. n the foowng sectons, we dscuss the mpact of the buffer management poces on read response tmes and the reasons for the trace dependences. A. ESfect of the Read Mss Pocy The read mss pocy s effect s best ustrated by the steep rse n read response tme under the LRU Voate pocy n Fgs. 7(a) and 7(c). As the safe buffer grows and the voate buffer shrnks, the LRU Voate pocy s unabe to take advantage of the arger safe buffer. The LRU Goba pocy, on the other hand, smpy performs more read repacements n the safe buffer as t grows. TABLE RESPONSE TMESCOMPARSON OF VOLATLEANDPARTALLY SAFE BUFFERS.PARTALLYSABUFFERS ARE MANAGED~SNGTHE LRUVOUTLUWR~ UPDATE POLCY.TMESARERE~RTEDMMLL~ECONDS. 0% Safe Buffer Sze 3% Safe Buffer Sze 5% Safe Buffer Sze 10% Vo. Buffer Sze 7% Vo. Buffe Sze 5% Vo. Buffer Sze Read Wrte Read Wrte Read Wrte Response Response Response Response Response Response Trace Tme Tme Tme Tme Tme Tme servera centa serverb centb

7 400 EEE TRANSACTONS ON COMPUTERS, VOL. 44, NO. 3, MARCH Wrte Response Tm (ms.) Wrte Response Tme (ms.), /, r 1% 7% Safe Buffer Sre 9% 22 3% Voate Buffer Sze j 4 1% 3% 5% 7% Safe Buffer Sze 9% 7% 5% 3% Voate Buffer Sze Wrte Response Tme (ms.) Wnte Response Tm (mr.) TRACE: serverb TRACE: ded 4! / 16.00, L : r / 0.00 t A / -J ;z 3% 1% 22 7% Safe Buffer Sze 3% Voate Buffer Sze % 9% t _ LRU Goba/ -=--_. WrtePurge LRUGobaV _ Wrt.dpdMC ; 5% 7% Safe Buffer Sze 5% 3% Voate Buffer Sze Fg. 6. Mean wrte response tmes. Because of the nfexbty of the LRU Voate pocy, a safe buffer that s too arge s wasted. Ths suggests that, n practce, proper seecton of the safe buffer sze w be more crtca under the LRU Voate pocy than under LRU Goba. Too tte safe buffer space w resut n poor wrte response tmes, but excess space w be wasted. n contrast, the LAU Goba pocy can take advantage of addtona safe buffer space to reduce read response tmes. Ths +-+-s ustrated more ceary n Fg. 8, whch shows read response tme as the safe buffer sze s ncreased (for trace ServerA ). The voate buffer sze s fxed at 5% of the reference set sze. B. Effect of the Wrte Aocaton Pocy When the safe buffer s reatvey sma, the Wrte Purge pocy resuts n a hgher read mss rato than the Wrte Update pocy, eadng to hgher read response tmes. The effect can be seen most ceary n Fgs. 7(b) and 7(c), n whch the performance of the two Wrte Purge poces degrades sharpy when the safe buffer sze fas beow 2% of the reference set sze. The Wrte Purge pocy performs poory when the safe buffer s sma because of a phenomenon we ca update theft. When an object resdng n the voate buffer s updated, the Wrte Purge pocy causes t to be deeted from the voate buffer and paced

8 AKYREK AND SALEM: MANAGEMENT OF PARTALLY SAFE BUFFERS 401 Read Response Tme Cm\.) 15.00; 4.xj 13. w/ 2.OLu TRACE: servera A / : Read Response Tme (ms.) _- LRUGobaU /.. _ wrteupdate,; a- _,-+ t / * LRU Goba/,,.- Wt&Ug~ TRACE: centa & ~ ~ &+ 7% Safe Buffer Sze 3, 3% Voate Buffer Sze 14.00~... 1 / / 3% 7% Safe Buffer Sze 7% 2 3% Voate Buffer Sze Read Response Tme (ms.) 6.20 ; 6.00 L 5.80 k TRACE: serverb / 5.20 r LRU Goba/ Read Response Tme em.) c,_--,_-- / /,, LRU Goba/ wrteupdate,, LRU Goba/, WrtePurge,,, 1, /,, TRACE: centb 5.00 L t 4.20 b 4.00 c 3.80 k 1% 7% Safe Buffer Szt 9% 2 R 3% Voate Buffer Sze Fg. 7. Mean read response tmes % 7% Safe Buffer Sk 7% :, 3% Voate Buffer Sze nstead n the safe buffer. f the safe buffer s sma, ths recenty read object may be competey emnated from the buffer much more qucky than t woud have been had t remaned n the voate buffer. Ths probem can be emnated by makng the safe buffer arger. Another possbty s to move data to the voate buffer when t s repaced n the safe buffer. Ths atter technque s used n the Sprte studes descrbed n [ 11. C. Trace Dependences One of the most strkng features of Fg. 7 s the dfference between the cent and server traces. For the traces from the server, the L.RU Goba pocy s sometmes superor to LRU Voate. However, LRU Goba performs poory on the cent traces regardess of the safe buffer sze. For both the cent and the server, there was tte dfference n performance between post-buffer and pre-buffer traces. Further examnaton of our resuts ndcates that the buffer management poces do exhbt genera patterns of behavor that are ndependent of the traces. However, certan trace characterstcs brng out, or magnfy, dfferent aspects of the behavors. To ustrate, consder Fg. 9, whch s dentca to Fg. 7, except that synchronous read transfer ratos are shown n-

9 402 EEE TRANSACTONS ON COMPUTERS, VOL. 44, NO. 3, MARCH 1995 Read Response Tm (ms.) 5.00 c *..~ J! 5% 10% 15% 20% Safe Buffer Sze 5% 5% 5% 5% VoLBuffer Sze Fg. 8. Read response tmes vs. safe fxed at 5% of the reference set sze. sze. The voate buffer sze s stead of the read response tmes. The synchronous read transfer rato descrbes the average number of dsk operatons performed per read request. Ths ncudes dsk reads to brng requested objects nto the buffers, as we as the wrtes needed to copy repaced drty objects back to the dsk. The effects of the LRU Voate and Wrte Purge poces, whch we have aready dscussed, are ceary evdent n both the cent and server traces. The LRU Voate pocy consstenty drves up the transfer rato as the voate buffer shrnks. Furthermore, the Wrte Purge pocy drves the rato up when the safe buffer s sma. However, ths behavor does not transate to consstent response tme behavor, as comparson of Fgs. 7 and 9 shows. The reason for the poor read response tmes of the LRU Goba poces on the post-buffer cent trace can be seen from Fg. 9(b). The read transfer ratos for the LRU Goba poces are eevated because of the very arge percentage of wrte requests n the trace (see Tabe ). When updates are frequent, read msses become more expensve because repaced objects are more key to be drty. (Snce updates are channeed through the safe buffer, t s key that objects repaced from there w be drty.) n effect, the LRU Goba pocy transfers some of the effort of copyng drty objects back to the dsks from wrte requests to read requests. When wrte requests are frequent, the mpact on read response tmes s severe. As we w show n the next secton, ths probem can be reduced sgnfcanty by usng asynchronous stagng. The LRU Goba poces have poor read response tmes on the pre-bu@r cent trace for a dfferent reason. Fg. 10 shows the mean response tme (for dsk operatons) as a functon of buffer sze for two of the traces. Response tmes for the prebuffer cent trace are severa tmes hgher than those of other traces (represented n Fg. 10(a) by the post-bu$er server trace). Furthermore, use of the LRU Goba poces ncreases the a- ready-hgh response tme st further. The hgh response tmes are caused by bursts of read requests n the pre-buffer cent trace, whch resut n ong queues at the dsk. Ths probem s exacerbated by the LRU Goba pocy, snce t must sometmes wrte drty objects back to the dsk to make room for newy read objects. Because of the ong queues, these wrtes have a sgnfcant mpact on dsk response tme. (The probem s magnfed somewhat because wrtes, ke reads, often arrve n bursts n our traces.) The resut s hgher response tmes for read requests, as Fg. 7(b) shows. The ong servce tmes aso account for the reatvey hgh wrte response tmes that we observed for ths trace (Fg. 6 and Tabe ). n summary, two trace characterstcs have been found to have an mpact on the reatve performance of the buffer management poces. Very bursty read request arrvas tend to drve up response tmes, regardess of the buffer management pocy. However, the probem s exacerbated by the LRU Goba pocy. Snce buffers tend to smooth out arrva bursts, ths s more key to be a factor n a man-memory buffer (prebuffer traces) than n a controer traces). n addton, a very hgh percentage of wrtes n the request stream (such as mght be observed at a controer buffer) may be detrmenta to read response tmes when the LRU Goba read mss pocy s used. However, ths probem can be aevated by update stagng, as we w show n the next secton. V. ASYNCHRONOU STAGNG Read and wrte response tmes can be reduced by asynchronousy stagng (copyng) drty pages from the safe buffer to the dsk. When a drty page s staged, t s not removed from the buffer. ts buffer state s smpy synchronzed wth ts state on dsk. Stagng operatons are ntated by the buffer manager, aowng many /O operatons to be removed from the crtca paths of wrte (and possby read) requests. Asynchronous stagng from the safe buffer has been suggested n [3] and has been mpemented n some systems, ncudng the BM 3990 storage manager. n the foowng experments, we consder the mpact of asynchronous stagng on response tmes and show how t affects the performance of the buffer management poces. Stagng can be mpemented n a varety of ways. n our smuator, the safe buffer s checked perodcay to determne whether t contans any drty objects. f so, a request to fush the odest drty object n the safe buffer s generated. After checkng for (and possby stagng) a drty object, the buffer manager wats C seconds before checkng agan. (The smua- ton parameter C s caed the mnmum stagng nterva.) When C s set to zero, the buffer manager ssues requests to copy objects back to the dsk as soon as they are updated. Larger vaues of C reduce the dsk s utzaton by reducng the number of wrte operatons that are performed. However, f C s too arge, stagng may become neffectve. Stagng reduces wrte response tmes sgnfcanty. Tabe V shows how the mean wrte response tme s affected when stagng s used. The resuts show that by stagng no faster than one bock per second, a safe buffer of sze 1% performs at east as we as a safe buffer three tmes as arge that s not staged. When updated pages are staged mmedatey to the

10 AKYREK AND SALEM: MANAGEMENT OF PARTALLY SAFE BUFFERS 403 Syncbmnous Read Transfer Rato Synchronous Read Transfer Rato ! TRACE: servera c L % 6% 8% Safe Buffer Sze 6% 4% 2% Voate Buffer Sze Syncbmous Read Transfer Rato 0.18 r 1 TRACE: serverb 0.18 TRACE: deme 0.15 t L 2% 8% Safe Buffer Sze 8% 2 :2 2% Voate Buffer Sk 0.12 Fg. 9. Synchronous read transfer ratos. The tota buffer sze s fxed at 10% of the reference set sze. dsk (C = 0), update tme s reduced to near zero for a but the update-ntensve trace centa. Even for that trace, a 3%-&e buffer couped wth mmedate stagng reduces the wrte response tme to zero. n quatatve terms, the data n Tabe V confrm the predctons of anaytc mode of [3]: Sma, staged safe buffers can reduce wrte response tmes to near zero. t mght be expected that asynchronous stagng woud ncrease read response tmes because of contenton for the dsk. Fg. 11 shows mean read response tmes for each of the four buffer management agorthms usng stagng wth C = 0. Ths fgure shoud be compared wth Fg. 7, whch shows read response tmes wthout update stagng. n abso- ute terms, the stagng dd not have a strong mpact on read response tmes. The dsk s utzaton was ow n a of our experments, and the smuated dsk had tte dffcuty handng the addtona traffc caused by stagng operatons, even when C = 0. n the next secton we w consder stagng to a more heavy utzed dsk. Fg. 11 aso shows that stagng mproves the performance of the LRU GZobu pocy reatve to the others on the cent workstaton traces. These were the traces on whch the LRU Goba pocy performed poory wthout stagng (see Fg. 7). Stagng s partcuary benefca under the LRU Goba pocy because t ceans drty objects n the safe buffer. When the

11 404 EEETRANSACTONSONCOMPUTERS, VOL. 44, NO. 3, MARCH1995 Dsk Rcspnse Tme mw TABLBV MEANWRTERESPONSETMESUSNGA~YNCHRONOUSSTAGNG.RESPONSE TMES FOR THE PARTALLY SAFE BUFFERS WERE DETERMNED USNG THE LRU GWBAUWRRE UPDATE BUFFER MANAGEMEW POLCY. HOWEVER, ALL OF THE POLCES PRODUCED SMLAR WRTE RL?SPoNSE ThES. AL r TMEs ARE REPORTED N M~SECONDS. 7 3%-safe 3%-safe %-voate %-voate Trace no stagng c=o servera cent A serverb centb TABLE V UNSAFEUPDATECOMPARSON-READRES~ONSETMES.THEVOLKZE BJF%RWASMANAGEDUS~GLRUREPLACEMENTANDASYNCHRO- NZATON~NTER~ALOF~~SECONDS.THEDATAFORTHEPARTL~LLYSA~E Bm WERE OBTANE)USNG THELNJVOL~TLEWR~EUPDAPDA~~ POLCY, W~~HANDW~~HO~TA~YNCHRONOUSSTAGNG.ALLRME~ARENML- LsmNDs. 80 /, * /,,,,,,, /,,-z-:~],,..!,,,, TRACE: centb LXU Goba/ wrteupdate 1-1 4% 8% Safe Buffer Sre 6% 2; 2% Voate Buffer Sze Fg. 10. Average dsk response tmes. The tota buffer sze s fxed at 10% of the refexnce set sze. L&U Goba pocy eects to repace an object n the safe buffer, the object s ess key to be drty. V. UNSAFE UPDATES For some appcatons, t s not necessary that a updates be paced nto safe memory mmedatey. Such appcatons ether toerate some ost updates n the event of a faure, or preserve the durabty of ther updates usng some other mechansm, such as oggng. We have performed experments to compare a partay safe buffer (wth does not ose updates) to a voate, copy-back buffer, whch may. We assume that the copy-back buffer performs perodc synchronzaton operatons to mt the amount of data that mght be ost because of a faure. The synchronzaton operatons are ntated at fxed ntervas caed synchronzaton ntervas. Each synchronzaton operaton ntates a batch of dsk wrte operatons to copy a drty objects n voate memory to the dsks. Tabes V and V compare read and wrte response tmes from a voate copy-back buffer wth those of a partay safe buffer. Both buffers provde comparabe read response tmes and very ow wrte response tmes. Wrte response tmes for the voate buffer are somewhat better than those of the partay safe buffer wthout stagng. When bocks are staged from the safe buffer, wrte response tmes are comparabe. The data n Tabes V and V show that voate copyback buffers and partay safe buffers can provde comparabe buffer performance,.e., they can be vewed as aternatves. However, partay safe buffers have severa ad- vantages. No updates are ost from a partay safe buffer n the event of a faure. Ths property aso emnates the need to mpement perodc synchronzaton. n transacton processng systems, whch use an addtona mechansm (such as REDO oggng and checkpontng) to guarantee

12 AKYjREK AND SALEM: MANAGEMENT OF PARTALLY SAFE BUFFERS 405 Read Respanse Tme m.) t TRACE savera c c c O RACE: centa c 9.00 c!, E :T E 1% Safe Buffer Sze 3% Voate Buffer Sze Read Response Tme Cm.) Read Response Tme (ms.) : TRACE: serverb 31.00~ : c ,, / / Lm 3% 7% Safe Buffer Sze 2 7% :z 3% Voate Buffer Sk Fg. 11. Mean read response tmes usng asynchronou stagng. the update safety, the need for the addtona mechansm s emnated.6 We expermented wth other synchronzaton ntervas and found that the copy-back buffer performed best wth shorter ntervas. (Ten seconds was the shortest nterva we tred.) Smary, the partay safe buffer performed best wth short stagng ntervas. Snce the utzaton of the dsks s ow for a 6 A og may be used for other purposes as we, such as to record the commt decsons for each transacton. However, such a og can be much more compact than a REDO og [ 181. of the traces, the extra dsk operatons generated at the synchronzaton ponts do not contrbute to sgnfcant ncreases n response tmes for concurrent read or wrte operatons. For both the partay safe buffer and the voate buffer, wrte response tmes are near zero. Ths ndcates that synchronous /O s rarey requred to satsfy wrte requests,.e., that the bocks repaced by newy wrtten bocks tend to be cean. n a more heavy oaded system, ths may not be true. To test ths hypothess, we ncreased the request oad offered to the smuated system by mutpyng the request nterarrva tmes (determned from the traces) by a scang factor.

13 EEE TRANSACTONS ON COMPUTERS, VOL. 44, NO. 3, MARCH 1995 TABLE V UNSAFE UPDATE COMPARSON - WFUE RESPONSE TMES THE VOLATLE BUFFER WAS MANAGED USNG LRU REPLACEMENT AND A SYNCHRON- ZATON NTERVAL OF 10 SECONDS. THE DATA FOR THE PARTLQLY SAFE BUFFER WERE OBTANED USNG THE LRU VOLATLE/WRTE UPDATE POLCY, WTH AND WTHOUT ASYNCHRONOUS STAGNG. NOTE THAT THE HGH WRTE RESPONSE TMES FOR THE WRTE-THROUGH AU-VOLATLE BUFFER ARE AN ARTFACT OF THE BURSTS OF WRTE REQUESTS N THE REFERENCE TRACES. ALL TMES ARE N MLLSECONDS. 3%-&e 7%-vo. stagng 3%-&e 7%-vo. stagng (C = 1) ( (C = 0) ] For exampe, by choosng a scang factor of 0.5, we dvde the actua traced nterarrva tmes n haf before suppyng them to the smuator. Dsk utzaton ncreases from about 8% (for trace ServerA ) wthout scang to over 60% wth a scang factor of The scaed traces provde a somewhat more artfca workoad than the unscaed ones. However, scang s a smpe way to get a rough dea of performance under hgher utzatons n the absence actua traces of heaver workoads. Fg. 12 shows the mean wrte response tme as a functon of the scang factor (the offered oad) for the two server traces. (Smar behavor was observed for the cent traces.) To produce these traces, the synchronzaton nterva was fxed at 10 seconds for the voate buffer, and the stagng nterva set to C = 0 (mmedate stagng) n the partay safe buffer. n practce, the seecton of an optma stagng or synchronzaton nterva becomes a compex probem at hgher oads. (For ow oads, short ntervas are amost aways best.) The fgures show that as the offered oad ncreases, the voate, copy-back buffer can mantan ow wrte response tmes at hgher oads than the partay safe buffer. The reason for ths s that the partay safe buffer channes a update requests through ts safe buffer (3% of the reference set sze), whereas updates can be buffered anywhere n the voate buffer (10% of the reference set sze). At hgh oads, the smaer safe buffer tends to f up wth unstaged updates. When ths occurs, update requests are key to encounter a deay whe a repacement s performed n the safe buffer. Of course, the mproved performance of the voate buffer at hgh oads comes at the expense of addtona data oss n the event of a faure, snce many unsafe updates resde n the voate buffer. Mean Wrte Response Tme (ms.) : LRU Voate/Wrte Update, C = 0 ;/,; - :. -,(/ TRACE: servera,.,,., / -,*;! f f, 1, f. X, /- Voate, Copy-Back Buffe (ower oad) (hgher oad) nter-arrva Tme Scang Factor Mean Wrte Response Tme (ms.) LRU Goba/Wrte Update, C = 0 LRU Voate/Wrte Update, C = 0 TRACE: serverb ( :, - :,,! Voate, Copy-Back Buffer 1,11 / -,,,,,, (ower oad) (hgher oad) nter-arrva Tme Scang Factor Fg. 12. Mean wrte response tmes vs. nter-arrva tme scang factor. V. DSCUSSON AND CONCLUSON We have consdered the probem of managng partay safe Even at ow utzatons, requests experence sgnfcant average watng t$es at the dsk server because of the bursty arrva rate. Utzaton does ncrease at the same rate as the scang factor because many requests have nterarrva tmes of zero. buffers. Our experments support a number of concusons about ther use: Ony a sma safe buffer s necessary. For our traces, a few megabytes was aways suffcent wthout stagng. f stagng s used, even smaer safe buffers w st provde good performance.

14 AKYjREK AND SALEM: MANAGEMENT OF PARTALLY SAFE BUFFERS 407 f the LRU Voate technque s used, excess space n the safe buffer w be wasted. The LRU Goba pocy s more fexbe and can take advantage of the addtona safe buffer space to reduce read response tmes. When the read reference pattern s very bursty, the LRU Goba pocy may resut n poor read response tmes. The same s true for very update-ntensve workoads. The LRU Goba pocy exacerbates the probem of read request bursts when t eects to repace drty objects n the safe buffer, creatng addtona work for the dsks at the wrong tme. These effects can be reduced by asynchronousy stagng drty objects to the dsks. The Wrte Update pocy s preferred to the Wrte Purge pocy because of the poor performance of Wrte Purge when the safe buffer s very sma. The sma buffer case s mportant because sma buffers are usuay suffcent to emnate most or a of the wrte response tme. For arger safe buffers, the dstncton s not sgnfcant. Asynchronous stagng of drty objects from the safe buffer reduces aready-ow wrte response tmes even further. Stagng aso mproves read-response tmes when the LRU Goba pocy s used. For workoads such as ours, t s best to perform stagng operatons wthout deay. f the dsks are more heavy utzed, t may be approprate to use ess bandwdth for stagng. Aternatvey, stagng operatons coud be performed at ower prorty than synchronous (request-ntated) dsk operatons. For ghty oaded dsks, the performance of a partay safe buffer s comparabe to that of an a-voate copyback buffer. By usng the partay safe buffer, the need for perodc synchronzaton of the buffer s emnated, and no updates w be ost n the event of a faure. One extenson of ths work s ts appcaton to transacton processng systems, or other systems that use oggng to guarantee the durabty of updates. f wrte response tmes can be made suffcenty sma by ntroducng safe buffers, then one of the prncpa motvatons for oggng w have been removed. (One verson of ths dea s currenty beng used n the POSTGRES storage system [ 171.) Asde from the emnaton of the compextes of oggng, an advantage of the safe-buffer approach s that faure recovery s very fast, snce there s no need to reconstruct the state of the database from the og. t may be possbe to further enhance the performance of partay safe buffers by takng advantage of buffered updates to reduce the cost of dsk update operatons. Pggy-backed updates, as suggested n [ 181, can be used, or the buffer manager can attempt to deay updates unt the dsk s de. We expect that such enhancements woud be most benefca when the dsk s heavy oaded, whch was not the case n our study. REFERENCES [] M. Baker, S. Asam, E. Dcprt, and J. Ousterhout, Non-voate Memory for fast reabe fe systems, Proc. nt Co@ Archtectura Support Programmng Lunguuges and Operatng Systems, pp , Oct [2] M. Baker and M. Suvan, The recovery box: Usng fast recovery to provde hgh avaabty n the Unx envronment, Proc. Usenx Technca Conf., pp , June [3] C. Copeand, R. Krshnamurty, and M. Smth, The case for safe RAM, Proc. 15th VLDB Co&, pp , Amsterdam, [4] M.H. Ech, A cassfcaton and comparson of man memory database recovery technques, Proc. nt Conf: Data Eng., EEE, pp , Feb [.5] J. Gat, Phoenx: A safe n-memory fe system, Comm. ACM, vo. 33, no. 1, pp , Jan [6] H. Garca-Mona and K. Saem, Hgh performance transacton processng wth memory-resdent data, Proc. nt Symp. Hgh Performance Computer Systems, Pars, Dec. 1987, North-Hoand, [7] T. Haener and A. Ret&, Prncpes of transacton-orented database recoveq, ACM Computng Surveys, vo. 15, no. 4, pp , Dec [8] BM 3990 Storage Contro ntroducton, BM, manua no. GA , [9] S.J. Leffer, M.K. McKusck, M.J. Kares, and J.S. Quarterman, Desgn and mpementaton of the 4.3BSD Unx Operatng System, Addson- Wesey, [o] T.J. Lehman and M.J. Carey, A recovery agorthm for a hghperformance memory-resdent database system, Proc. ACM SGMOD nt Conf: Management ofdata, pp , San Francsco, [] S.W. Ng, mprovng dsk performance va atency reducton, EEE Trans. Computers, vo. 40, no. 1, pp , Jan [12] E. Rahm, Performance evauaton of extended storage archtectures for transacton processng, Proc. ACM SGMOD nt Con$ Munagement of Data, pp , [13] C. Ruemmer and J. Wkes, Unx dsk access patterns, Usenx Conf: Proc., pp , Jan [14] H. Schwetman, CSM Reference Manua (Rev. 14); MCC Technca Report No. ACT-ST , MCC, Austn, Texas, Mar [15] A.J. Smth, Dsk cache - Mss rato anayss and desgn consderatons, ACM Tran Computer Systems, vo. 3, no. 3, pp , Aug [16] J.A. Sowoxth,and C.U. Orj, Wrte-ony dsk caches, Proc. ACM SGMOD nt Conj Management of Data, pp , May [17] M. Stonebraker, Desgn of the POSTGRES storage sysytem, Proc. 13th VLDB Conf., pp , Brghton, UK, [18] M. Stonebraker, R. Katz, D. Patterson, and J. Ousterhout, Desgn of XPRS, Proc. 14th VLDB Conf., pp , Los Angees, Sedat Akyrek receved the BS degree n computer engneerng from the Mdde East Technca Unversty, Ankara, Turkey, n 1988 and the MS degree n computer scence from the Unversty of Maryand, Coege Park, n 1991, and s currenty a PhD canddate n the Department of Computer Scence at the Unversty of Maryand, Coege Park. Akytrck s research nterests ncude dsk and /O systems, operatng systems, databases, and dstrbuted systems. He s currenty workng on adaptve data storage technques for dsk systems. Kenneth Saem receved the BS degree n eectrca engneerng and apped mathematcs from Carnege Meon Unversty n 1983 and the PhD degree n computer scence from Prnceton Unversty n Dr. Saem s an assstant professor n the Department of Computer Scence at the Unversty of Maryand, Coege Park. Hs research nterests ncude database and operatng systems, and transacton processng. He s a member of the ACM and the EEE Computer Socety.

Simon Sheu Computer Science National Tsing Hua Universtity Taiwan, ROC

Simon Sheu Computer Science National Tsing Hua Universtity Taiwan, ROC Mounr A. Tantaou School of Electrcal Engneerng and Computer Scence Unversty of Central Florda Orlando, FL 3286-407-823-393 tantaou@cs.ucf.edu Interacton wth Broadcast Vdeo Ken A. Hua School of Electrcal

More information

Hybrid Transcoding for QoS Adaptive Video-on-Demand Services

Hybrid Transcoding for QoS Adaptive Video-on-Demand Services 732 IEEE Transactons on Consumer Electroncs, Vol. 50, No. 2, MAY 2004 Hybrd Transcodng for QoS Adaptve Vdeo-on-Demand Servces Ilhoon Shn and Kern Koh Abstract Transcodng s a core technque that s used n

More information

tj tj D... '4,... ::=~--lj c;;j _ ASPA: Automatic speech-pause analyzer* t> ,. "",. : : :::: :1'NTmAC' I

tj tj D... '4,... ::=~--lj c;;j _ ASPA: Automatic speech-pause analyzer* t> ,. ,. : : :::: :1'NTmAC' I ASPA: Automatc speech-pause analyzer* D. GERVERt and G. DNELEY Unversty of Durham, Durham, England ASPA: The Programs Snce the actual detals of nterface samplng, dsk storage routnes, etc., wll depend upon

More information

A Scalable HDD Video Recording Solution Using A Real-time File System

A Scalable HDD Video Recording Solution Using A Real-time File System H. L et al.: A Scalable HDD Vdeo Recordng Soluton Usng A Real-tme Fle System A Scalable HDD Vdeo Recordng Soluton Usng A Real-tme Fle System Hong L, Stephen R. Cumpson Member, IEEE, Robert Jochemsen, Jan

More information

Instructions for Contributors to the International Journal of Microwave and Wireless Technologies

Instructions for Contributors to the International Journal of Microwave and Wireless Technologies Instructons for Contrbutors to the Internatonal Journal of Mcrowave and Wreless Technologes Frst A. Author 1, Second Author 1,2, Thrd Author 2 1 Cambrdge Unversty Press, Ednburgh Buldng, Shaftesbury Road,

More information

Following a musical performance from a partially specified score.

Following a musical performance from a partially specified score. Followng a muscal performance from a partally specfed score. Bryan Pardo and Wllam P. Brmngham Artfcal Intellgence Laboratory Electrcal Engneerng and Computer Scence Dept. and School of Musc The Unversty

More information

Improving Reliability and Energy Efficiency of Disk Systems via Utilization Control

Improving Reliability and Energy Efficiency of Disk Systems via Utilization Control Ths paper appeared n the Proceedngs of the 2th IEEE Symposum on Computers and Communcatons (ISCC'08, Marrakech, Morocco, July 2008. Improvng Relablty and Energy Effcency of Dsk Systems va Utlzaton Control

More information

Optimized PMU placement by combining topological approach and system dynamics aspects

Optimized PMU placement by combining topological approach and system dynamics aspects Optmzed PU placement by combnng topologcal approach and system dynamcs aspects Jonas Prommetta, Jakob Schndler, Johann Jaeger Insttute of Electrcal Energy Systems Fredrch-Alexander-Unversty Erlangen-Nuremberg

More information

System of Automatic Chinese Webpage Summarization Based on The Random Walk Algorithm of Dynamic Programming

System of Automatic Chinese Webpage Summarization Based on The Random Walk Algorithm of Dynamic Programming Send Orders for Reprnts to reprnts@benthamscence.ae The Open Cybernetcs & Systemcs Journal, 205, 9, 35-322 35 Open Access System of Automatc Chnese Webpage Summarzaton Based on The Random Walk Algorthm

More information

A Comparative Analysis of Disk Scheduling Policies

A Comparative Analysis of Disk Scheduling Policies A Comparatve Analyss of Dsk Schedulng Polces Toby J. Teorey and Tad B. Pnkerton Unversty of Wsconsn* Fve well-known schedulng polces for movable head dsks are compared usng the performance crtera of expected

More information

Simple VBR Harmonic Broadcasting (SVHB)

Simple VBR Harmonic Broadcasting (SVHB) mple VBR Harmonc Broadcastng (VHB) Hsang-Fu Yu ab, Hung-hang Yang a, Y-Mng hen c, -Mng Tseng a, and hen-y Kuo a a Dep. of omputer cence & Informaton Engneerng, atonal entral Unversty, Tawan b omputer enter,

More information

A STUDY OF TRUMPET ENVELOPES

A STUDY OF TRUMPET ENVELOPES A STUDY OF TRUMPET ENVELOPES Roger B. Dannenberg, Hank Pellern, and Istvan Dereny School of Computer Scence, Carnege Mellon Unversty Pttsburgh, PA 15213 USA rbd@cs.cmu.edu, hank.pellern@andrew.cmu.edu,

More information

The UCD community has made this article openly available. Please share how this access benefits you. Your story matters!

The UCD community has made this article openly available. Please share how this access benefits you. Your story matters! Provded by the author(s) and Unversty College Dubln Lbrary n accordance wth publsher polces., Please cte the publshed verson when avalable. tle Dynamc Complexty Scalng for Real-me H.264/AVC Vdeo Encodng

More information

Error Concealment Aware Rate Shaping for Wireless Video Transport 1

Error Concealment Aware Rate Shaping for Wireless Video Transport 1 Error Concealment Aware Rate Shapng for Wreless Vdeo Transport 1 Trsta Pe-chun Chen and Tsuhan Chen 2 Abstract Streamng of vdeo, whch s both source- and channel- coded, over wreless networks faces the

More information

Failure Rate Analysis of Power Circuit Breaker in High Voltage Substation

Failure Rate Analysis of Power Circuit Breaker in High Voltage Substation T. Suwanasr, M. T. Hlang and C. Suwanasr / GMSAR Internatonal Journal 8 (2014) 1-6 Falure Rate Analyss of Power Crcut Breaker n Hgh Voltage Substaton Thanapong Suwanasr, May Thandar Hlang and Cattareeya

More information

RIAM Local Centre Woodwind, Brass & Percussion Syllabus

RIAM Local Centre Woodwind, Brass & Percussion Syllabus 8 RIAM Local Centre Woodwnd, Brass & Percusson Syllabus 2015-2018 AURAL REQUIREMENTS AND THEORETICAL QUESTIONS REVISED FOR ALL PRACTICAL SUBJECTS AURAL TESTS From Elementary to Grade V ths area s worth

More information

TRADE-OFF ANALYSIS TOOL FOR INTERACTIVE NONLINEAR MULTIOBJECTIVE OPTIMIZATION Petri Eskelinen 1, Kaisa Miettinen 2

TRADE-OFF ANALYSIS TOOL FOR INTERACTIVE NONLINEAR MULTIOBJECTIVE OPTIMIZATION Petri Eskelinen 1, Kaisa Miettinen 2 Internatonal Conference 20th EURO Mn Conference Contnuous Optmaton and Knowledge-Based Technologes (EurOPT-2008) May 20 23, 2008, Nernga, LITHUANIA ISBN 978-9955-28-283-9 L. Saalausas, G.W. Weber and E.

More information

Why Take Notes? Use the Whiteboard Capture System

Why Take Notes? Use the Whiteboard Capture System Why Take Notes? Use the Whteboard Capture System L-we He Zhengyou Zhang and Zcheng Lu September, 2002 Techncal Report MSR-TR-2002-89 Mcrosoft Research Mcrosoft Corporaton One Mcrosoft Way Redmond, WA 98052

More information

Statistics AGAIN? Descriptives

Statistics AGAIN? Descriptives Cal State Northrdge Ψ427 Andrew Answorth PhD Statstcs AGAIN? What do we want to do wth statstcs? Organze and Descrbe patterns n data Takng ncomprehensble data and convertng t to: Tables that summarze the

More information

Accepted Manuscript. An improved artificial bee colony algorithm for flexible job-shop scheduling problem with fuzzy processing time

Accepted Manuscript. An improved artificial bee colony algorithm for flexible job-shop scheduling problem with fuzzy processing time Accepted Manuscrpt An mproved artfcal bee colony algorthm for flexble ob-shop schedulng problem wth fuzzy processng tme Ka Zhou Gao, Ponnuthura Nagaratnam Suganthan, Quan Ke Pan, Tay Jn Chua, Chn Soon

More information

Analysis of Subscription Demand for Pay-TV

Analysis of Subscription Demand for Pay-TV Analyss of Subscrpton Demand for Pay-TV Manabu Shshkura Researcher Insttute for Informaton and Communcatons Polcy 2-1-2 Kasumgasek, Chyoda-ku Tokyo 110-8926 Japan m-shshkura@soumu.go.jp Tel: 03-5253-5496

More information

T541 Flat Panel Monitor User Guide ENGLISH

T541 Flat Panel Monitor User Guide ENGLISH T541 Flat Panel Montor User Gude ENGLISH Frst Edton (June / 2002) Note : For mportant nformaton, refer to the Montor Safety and Warranty manual that comes wth ths montor. Ths publcaton could contan techncal

More information

Study on the location of building evacuation indicators based on eye tracking

Study on the location of building evacuation indicators based on eye tracking Study on the locaton of buldng evacuaton ndcators based on eye trackng Yue L Tsnghua Unversty yue-l5@malstsnghuaeducn Png hang Tsnghua Unversty zhangp@malstsnghuaeducn Hu hang Tsnghua Unversty, zhhu@tsnghuaeducn

More information

LOW-COMPLEXITY VIDEO ENCODER FOR SMART EYES BASED ON UNDERDETERMINED BLIND SIGNAL SEPARATION

LOW-COMPLEXITY VIDEO ENCODER FOR SMART EYES BASED ON UNDERDETERMINED BLIND SIGNAL SEPARATION LOW-COMPLEXITY VIDEO ENCODER FOR SMART EYES BASED ON UNDERDETERMINED BLIND SIGNAL SEPARATION Jng Lu, Fe Qao *, Zhjan Ou and Huazhong Yang Department of Electronc Engneerng, Tsnghua Unversty ABSTRACT Ths

More information

AMP-LATCH* Ultra Novo mm [.025 in.] Ribbon Cable 02 MAR 12 Rev C

AMP-LATCH* Ultra Novo mm [.025 in.] Ribbon Cable 02 MAR 12 Rev C AMP-LATCH* Ultra Novo Applcaton Specfcaton Receptacle Connectors for 114-40056 0.64 mm [.025 n.] Rbbon Cable 02 MAR 12 All numercal values are n metrc unts [wth U.S. customary unts n brackets]. Dmensons

More information

QUICK START GUIDE v0.98

QUICK START GUIDE v0.98 QUICK START GUIDE v0.98 QUICK HELP Q A 1 STEP BY STEP 3 GLOSSARY 2 A B C 1 INSTALLATION 1. Make sure that the hardware nstallaton s performed by a certfed vendor 2. Install OTOTRAK app from Apple s App

More information

Scalable QoS-Aware Disk-Scheduling

Scalable QoS-Aware Disk-Scheduling Scalable QoS-Aware Dsk-Schedulng Wald G. Aref Khaled El-Bassyoun Ibrahm Kamel Mohamed F. Mokbel Department of Computer Scences, urdue Unversty, West Lafayette, IN 47907-1398 anasonc Informaton and Networkng

More information

Correcting Image Placement Errors Using Registration Control (RegC ) Technology In The Photomask Periphery

Correcting Image Placement Errors Using Registration Control (RegC ) Technology In The Photomask Periphery Correctng Image Placement Errors Usng Regstraton Control (RegC ) Technology In The Photomask Perphery Av Cohen 1, Falk Lange 2 Guy Ben-Zv 1, Erez Gratzer 1, Dmtrev Vladmr 1 1. Carl Zess SMS Ltd., Karmel

More information

User s manual. Digital control relay SVA

User s manual. Digital control relay SVA User s manual Dgtal control relay DISIBEINT ELECTRONIC S.L, has been present n the feld of the manufacture of components for the ndustral automaton for more than years, and mantans n constant evoluton

More information

Modeling Form for On-line Following of Musical Performances

Modeling Form for On-line Following of Musical Performances Modelng Form for On-lne Followng of Muscal Performances Bryan Pardo 1 and Wllam Brmngham 2 1 Computer Scence Department, Northwestern Unversty, 1890 Maple Ave, Evanston, IL 60201 2 Department of Math and

More information

A Quantization-Friendly Separable Convolution for MobileNets

A Quantization-Friendly Separable Convolution for MobileNets arxv:1803.08607v1 [cs.cv] 22 Mar 2018 A Quantzaton-Frendly Separable for MobleNets Abstract Tao Sheng tsheng@qt.qualcomm.com Xaopeng Zhang parker.zhang@gmal.com As deep learnng (DL) s beng rapdly pushed

More information

Production of Natural Penicillins by Strains of Penicillium chrysogenutn

Production of Natural Penicillins by Strains of Penicillium chrysogenutn Producton of Natural Pencllns by Strans of Pencllum chrysogenutn a J. FUSK and ЬЕ. WELWRDOVÁ ^Department of Mcrobology and Bochemstry, Slovak Techncal Unversty, Bratslava b Botka, Slovenská Ľupča Receved

More information

Critical Path Reduction of Distributed Arithmetic Based FIR Filter

Critical Path Reduction of Distributed Arithmetic Based FIR Filter Crtcal Path Reducton of strbuted rthmetc Based FIR Flter Sunta Badave epartment of Electrcal and Electroncs Engneerng.I.T, urangabad aharashtra, Inda njal Bhalchandra epartment of Electroncs and Telecommuncaton

More information

Cost-Aware Fronthaul Rate Allocation to Maximize Benefit of Multi-User Reception in C-RAN

Cost-Aware Fronthaul Rate Allocation to Maximize Benefit of Multi-User Reception in C-RAN Cost-Aware Fronthaul Rate Allocaton to Maxmze Beneft of Mult-User Recepton n C-RAN Dora Bovz, Chung Shue Chen, Sheng Yang To cte ths verson: Dora Bovz, Chung Shue Chen, Sheng Yang. Cost-Aware Fronthaul

More information

) No. 07 L 6781 ) ) )

) No. 07 L 6781 ) ) ) c '"r 0 STATE OF LLNOS ) ) ss: COUNTY OF C 0 O.K ) N THE CRCUT COURT OF COOK COUNTY LLNOS COUNTY OEPARTMENT. LAW OVSON JOHN DOE # ) Pantff ) VS. THE CHCAGO PROVNCE OF THE SOCETY OF JESUS. Defendant. )

More information

Small Area Co-Modeling of Point Estimates and Their Variances for Domains in the Current Employment Statistics Survey

Small Area Co-Modeling of Point Estimates and Their Variances for Domains in the Current Employment Statistics Survey Small Area Co-Modelng of Pont Estmates and Ther Varances for Domans n the Current Employment Statstcs Survey Jule Gershunskaya, Terrance D. Savtsky U.S. Bureau of Labor Statstcs FCSM, March 2018 Dsclamer:

More information

INSTRUCTION MANUAL FOR THE INSTALLATION, USE AND MAINTENANCE OF THE REGULATOR GENIUS POWER COMBI

INSTRUCTION MANUAL FOR THE INSTALLATION, USE AND MAINTENANCE OF THE REGULATOR GENIUS POWER COMBI NSTRUCTON MANUAL FOR THE NSTALLATON, USE AND MANTENANCE OF THE REGULATOR GENUS POWER COMB (TRANSLATON OF THE ORGNAL NSTRUCTON MANUAL N TALAN) PRELMNARY VERSON WARRANTY The devce s guaranteed 24 months

More information

THE IMPORTANCE OF ARM-SWING DURING FORWARD DIVE AND REVERSE DIVE ON SPRINGBOARD

THE IMPORTANCE OF ARM-SWING DURING FORWARD DIVE AND REVERSE DIVE ON SPRINGBOARD THE MPORTANCE OF ARM-SWNG DURNG FORWARD DVE AND REVERSE DVE ON SPRNGBOARD Ken Yokoyama Laboratory of Bomechancs Faculty ofeducaton Kanazawa Unversty Kanazawa, Japan J unjro Nagano Department of Physcal

More information

Decision Support by Interval SMART/SWING Incorporating. Imprecision into SMART and SWING Methods

Decision Support by Interval SMART/SWING Incorporating. Imprecision into SMART and SWING Methods Decson Support by Interval SMART/SWING Incorporatng Imprecson nto SMART and SWING Methods Abstract: Interval judgments are a way of handlng preferental and nformatonal mprecson n multcrtera decson analyss.

More information

Color Monitor. L200p. English. User s Guide

Color Monitor. L200p. English. User s Guide Color Montor L200p User s Gude Englsh Frst Edton (February / 2003) Note : For mportant nformaton, refer to the Montor Safety and Warranty manual that comes wth ths montor. Contents ENGLISH Safety (Read

More information

Quantization of Three-Bit Logic for LDPC Decoding

Quantization of Three-Bit Logic for LDPC Decoding Proceedngs of the World Congress on Engneerng and Computer Scence 2011 Vol II, October 19-21, 2011, San Francsco, USA Quantzaton of Three-Bt Logc for LDPC Decodng Raymond Moberly and Mchael E. O'Sullvan

More information

Integration of Internet of Thing Technology in Digital Energy Network with Dispersed Generation

Integration of Internet of Thing Technology in Digital Energy Network with Dispersed Generation Amercan Scentfc Research Journal for Engneerng, Technology, and Scences (ASRJETS) ISS (Prnt) 2313-4410, ISS (Onlne) 2313-4402 Global Socety of Scentfc Research and Researchers http://asrjetsjournal.org/

More information

Anchor Box Optimization for Object Detection

Anchor Box Optimization for Object Detection Anchor Box Optmzaton for Object Detecton Yuany Zhong 1, Janfeng Wang 2, Jan Peng 1, and Le Zhang 2 1 Unversty of Illnos at Urbana-Champagn 2 Mcrosoft Research 1 {yuanyz2, janpeng}@llnos.edu, 2 {janfw,

More information

Conettix D6600/D6100IPv6 Communications Receiver/Gateway Quick Start

Conettix D6600/D6100IPv6 Communications Receiver/Gateway Quick Start Conettx / Communcatons Recever/Gateway Quck Start.0 Parts Lst able : Conettx System Components Qty. Descrpton Conettx Communcatons Recever/Gateway AC power cord Battery cable P660 I/O cable P660 Rack mount

More information

Academic Standards and Calendar Committee Report # : Proposed Academic Calendars , and

Academic Standards and Calendar Committee Report # : Proposed Academic Calendars , and Unversty of Rhode sland DgtalCommons@UR Blls 2004 Academc Standards and Calendar Commttee Report #2003041: Proposed Academc Calendars 200708, 200809 and 200910 Unversty of Rhode sland Follow ths and addtonal

More information

AN INTERACTIVE APPROACH FOR MULTI-CRITERIA SORTING PROBLEMS

AN INTERACTIVE APPROACH FOR MULTI-CRITERIA SORTING PROBLEMS AN INTERACTIVE APPROACH FOR MULTI-CRITERIA SORTING PROBLEMS A THESIS SUBMITTED TO THE GRADUATE SCHOOL OF NATURAL AND APPLIED SCIENCES OF MIDDLE EAST TECHNICAL UNIVERSITY BY BURAK KESER IN PARTIAL FULFILLMENT

More information

AIAA Optimal Sampling Techniques for Zone- Based Probabilistic Fatigue Life Prediction

AIAA Optimal Sampling Techniques for Zone- Based Probabilistic Fatigue Life Prediction AIAA 2002-383 Optmal Samplng Technques or Zone- Based Probablstc Fatgue Le Predcton M. P. Enrght Southwest Research Insttute San Antono, TX H. R. Mllwater Unversty o Texas at San Antono San Antono, TX

More information

Operation Guide 3197

Operation Guide 3197 MO1004-EA Getting Acquainted Congratuations upon your seection of this CASIO watch. To get the most out of your purchase, be sure to read this manua carefuy. Keep the watch exposed to bright ight The eectricity

More information

3 Part differentiation, 20 parameters, 3 histograms Up to patient results (including histograms) can be stored

3 Part differentiation, 20 parameters, 3 histograms Up to patient results (including histograms) can be stored st Techncal Specfcatons Desgned n France Wth a rch past and a professonal experence bult-up over 35 years, SFRI s a French nvatve company commtted to developng preon In Vtro Dst solutons. SFRI has bult

More information

current activity shows on the top right corner in green. The steps appear in yellow

current activity shows on the top right corner in green. The steps appear in yellow Browzwear Tutorals Tutoral ntroducton Ths tutoral leads you through the best practces of color ways operatons usng an llustrated step by step approach. Each slde shows the actual applcaton at the stage

More information

Operation Guide

Operation Guide MO0503-EA Getting Acquainted Congratuations upon your seection of this CASIO watch. To get the most out of your purchase, be sure to read this manua carefuy and keep it on hand for ater reference when

More information

Craig Webre, Sheriff Personnel Division/Law Enforcement Complex 1300 Lynn Street Thibodaux, Louisiana 70301

Craig Webre, Sheriff Personnel Division/Law Enforcement Complex 1300 Lynn Street Thibodaux, Louisiana 70301 DATE OF APPLCATON: Craig Webre, Sheriff Personnel Division/Law Enforcement Complex 1300 Lynn Street Thibodaux, Louisiana 70301 N GENERAL EMAL ADDRESS: For Local Calls - (985) 532-4380 (985) 446-2255 (985)

More information

SKEW DETECTION AND COMPENSATION FOR INTERNET AUDIO APPLICATIONS. Orion Hodson, Colin Perkins, and Vicky Hardman

SKEW DETECTION AND COMPENSATION FOR INTERNET AUDIO APPLICATIONS. Orion Hodson, Colin Perkins, and Vicky Hardman SKEW DETECTION AND COMPENSATION FOR INTERNET AUDIO APPLICATIONS Oron Hodson, Coln Perkns, and Vcky Hardman Department of Computer Scence Unversty College London Gower Street, London, WC1E 6BT, UK. ABSTRACT

More information

SONG STRUCTURE IDENTIFICATION OF JAVANESE GAMELAN MUSIC BASED ON ANALYSIS OF PERIODICITY DISTRIBUTION

SONG STRUCTURE IDENTIFICATION OF JAVANESE GAMELAN MUSIC BASED ON ANALYSIS OF PERIODICITY DISTRIBUTION SOG STRUCTURE IDETIFICATIO OF JAVAESE GAMELA MUSIC BASED O AALYSIS OF PERIODICITY DISTRIBUTIO D. P. WULADARI, Y. K. SUPRAPTO, 3 M. H. PUROMO,,3 Insttut Teknolog Sepuluh opember, Department of Electrcal

More information

NCH Software VideoPad Video Editor

NCH Software VideoPad Video Editor NCH Software VideoPad Video Editor This user guide has been created for use with VideoPad Video Editor Version 4.xx NCH Software Technica Support If you have difficuties using VideoPad Video Editor pease

More information

Product Information. Manual change system HWS

Product Information. Manual change system HWS Product Informaton HWS HWS Flexble. Compact. Productve. HWS manual change system Manual tool change system wth ntegrated ar feed-through and optonal electrc feed-through Feld of applcaton Excellently sutable

More information

Lost on the Web: Does Web Distribution Stimulate or Depress Television Viewing?

Lost on the Web: Does Web Distribution Stimulate or Depress Television Viewing? Lost on the Web: Does Web Dstrbuton Stmulate or Depress Televson Vewng? Joel Waldfogel The Wharton School Unversty of Pennsylvana August 10, 2007 Prelmnary comments welcome Abstract In the past few years,

More information

Technical Information

Technical Information CHEMCUT Techncal Informaton CORPORATION Introducton The Chemcut CC8000 etcher has many new features desgned to reduce the cost of manufacturng and, just as mportantly, the cost of ownershp. Keepng the

More information

Environmental Reviews. Cause-effect analysis for sustainable development policy

Environmental Reviews. Cause-effect analysis for sustainable development policy Envronmental Revews Cause-effect analyss for sustanable development polcy Journal: Envronmental Revews Manuscrpt ID er-2016-0109.r2 Manuscrpt Type: Revew Date Submtted by the Author: 24-Feb-2017 Complete

More information

Operation Guide 3271

Operation Guide 3271 MO1106-EA Operation Guide 3271 Getting Acquainted ongratuations upon your seection of this ASIO watch. To get the most out of your purchase, be sure to read this manua carefuy. Be sure to keep a user documentation

More information

Detecting Errors in Blood-Gas Measurement by Analysiswith Two Instruments

Detecting Errors in Blood-Gas Measurement by Analysiswith Two Instruments CLIN. CHEM. 33/4, 512-517 (1987) Detectng Errors n Blood-Gas Measurement by Analysswth Two Instruments LouIs F. Metzger, Wllam B. Stauffer, Ann V. Kruplnskl, Rchard P. MIIlman,3 George S. Cembrowskl,2

More information

Reduce Distillation Column Cost by Hybrid Particle Swarm and Ant

Reduce Distillation Column Cost by Hybrid Particle Swarm and Ant From the SelectedWorks of Dr. Sandp Kumar Lahr Summer July 20, 2016 Reduce Dstllaton Column Cost by Hybrd Partcle Swarm and Ant Dr. Sandp k lahr chnmaya lenka Avalable at: https://works.bepress.com/sandp_lahr/33/

More information

Product Information. Manual change system HWS

Product Information. Manual change system HWS Product Informaton HWS HWS Flexble. Compact. Productve. HWS manual change system Manual tool change system wth ntegrated ar feed-through and optonal electrc feed-through Feld of applcaton Excellently sutable

More information

Operation Guide 5200

Operation Guide 5200 MO1103-EA Getting Acquainted ongratuations upon your seection of this ASIO watch. To get the most out of your purchase, be sure to read this manua carefuy. Be sure to keep a user documentation handy for

More information

Simple Solution for Designing the Piecewise Linear Scalar Companding Quantizer for Gaussian Source

Simple Solution for Designing the Piecewise Linear Scalar Companding Quantizer for Gaussian Source 94 J. NIKOIĆ, Z. PERIĆ,. VEIMIROVIĆ, SIMPE SOUTION FOR DESIGNING THE PIECEWISE INEAR SCAAR Smle Soluton for Desgnng the Pecewse near Scalar Comandng Quantzer for Gaussan Source Jelena NIKOIĆ, Zoran PERIĆ,

More information

IN DESCRIBING the tape transport of

IN DESCRIBING the tape transport of Apparatus For Magnetc Storage on Three-Inch Wde Tapes R. B. LAWRANCE R. E. WILKINS R. A. PENDLETON IN DESCRIBING the tape transport of the DATAmatc 1, t s perhaps well to begn by revewng the nfluental

More information

MODELING AND ANALYZING THE VOCAL TRACT UNDER NORMAL AND STRESSFUL TALKING CONDITIONS

MODELING AND ANALYZING THE VOCAL TRACT UNDER NORMAL AND STRESSFUL TALKING CONDITIONS MODELING AND ANALYZING THE VOCAL TRACT UNDER NORMAL AND STRESSFUL TALING CONDITIONS Ismal Shahn and Naeh Botros 2 Electrcal/Electroncs and Comuter Engneerng Deartment Unversty of Sharjah, P. O. Box 27272,

More information

EDT/Collect for DigitalMicrograph

EDT/Collect for DigitalMicrograph May 2016 (Provisiona) EDT/Coect for DigitaMicrograph Data Coection for Eectron Diffraction Tomography EDT/Coect Manua 1.0 HREM Research Inc. Introduction The EDT/Coect software has been deveoped by HREM

More information

Product Information. Universal swivel units SRU-plus

Product Information. Universal swivel units SRU-plus Product Informaton Unversal swvel unts SRU-plus SRU-plus Unversal swvel unts Robust. Fast. Hgh Performance. SRU-plus unversal rotary actuator Unversal unt for pneumatc swvel and turnng movements. Feld

More information

Sealed Circular LC Connector System Plug

Sealed Circular LC Connector System Plug Sealed Crcular LC Connector System Plug Instructon Sheet Kt 1828618- [ ], Receptacle Kt 1828619- [ ], 408-10079 and EMI Receptacle Kt 1985193- [ ] 07 APR 11 Plug Kt 1828618 -[ ] Cable Fttng Receptacle

More information

Novel Quantization Strategies for Linear Prediction with Guarantees

Novel Quantization Strategies for Linear Prediction with Guarantees Smon S. Du* Ychong Xu* Yuan L Hongyang Zhang Aart Sngh Pulkt Grover Carnege Mellon Unversty, Pttsburgh, PA 15213, USA *: Contrbute equally. SSDU@CS.CMU.EDU YICHONGX@CS.CMU.EDU LIYUANCHRISTY@GMAIL.COM HONGYANZ@CS.CMU.EDU

More information

Operation Guide 4717

Operation Guide 4717 MO0812-EB Getting Acquainted Congratuations upon your seection of this CASIO watch. To get the most out of your purchase, be sure to read this manua carefuy. This watch does not have a Time Zone that corresponds

More information

9! VERY LARGE IN THEIR CONCERNS. AND THEREFORE, UH, i

9! VERY LARGE IN THEIR CONCERNS. AND THEREFORE, UH, i 340 WELL, alack PAJAMAS WAS A SOMEWHAT METAPHORCAL 2 TERM. MANY VETNAMESE PEASANTS TENDED TO WEAR 3 BLACK PAJAMAS, BUT WHAT AM REFERRNG TO S THAT 4 OUTSDE OF THE NORTH VETNAMESE UNTS AND ~OME OF 5 THE

More information

TRANSCENSION DMX OPERATOR 2 USER MANUAL

TRANSCENSION DMX OPERATOR 2 USER MANUAL TRANSCENSION DMX OPERATOR 2 USER MANUAL I. PRODUCT DESCRIPTIONS Thank you for using our company the 192 CH DMX OPERATOR. To optimize the performance of this product, pease read these operating instructions

More information

Social Interactions and Stigmatized Behavior: Donating Blood Plasma in Rural China

Social Interactions and Stigmatized Behavior: Donating Blood Plasma in Rural China Socal Interactons and Stgmatzed Behavor: Donatng Blood Plasma n Rural Chna X Chen Yale Unversty and IZA Davd E. Sahn Cornell Unversty and IZA Xaobo Zhang Pekng Unversty and IFPRI March 2018 Abstract Despte

More information

Modular Plug Connectors (Standard and Small Conductor)

Modular Plug Connectors (Standard and Small Conductor) Modular Plug Connectors (Standard and Small Conductor) Applcaton Specfcaton 114-6016 04 APR 11 All numercal values are n metrc unts [wth U.S. customary unts n brackets]. Dmensons are n mllmeters [and nches].

More information

S Micro--Strip Tool in. S Combination Strip Tool ( ) S Cable Holder Assembly (Used only

S Micro--Strip Tool in. S Combination Strip Tool ( ) S Cable Holder Assembly (Used only Instructon Sheet LghtCrmp* Plus LC 408-10103 (for Jacketed Cable) Connectors 18 AUG 09 Rear Protectve Cap Termnaton CoverG Boot Connector Assembly Crmp Eyelet Duplex Clp G Connector kt s shpped wth these

More information

Drum Transcription in the presence of pitched instruments using Prior Subspace Analysis

Drum Transcription in the presence of pitched instruments using Prior Subspace Analysis ISSC 2003, Limerick. Juy -2 Drum Transcription in the presence of pitched instruments using Prior Subspace Anaysis Derry FitzGerad φ, Bob Lawor*, and Eugene Coye φ φ Music Technoogy Centre, Dubin Institute

More information

Product Information. Miniature rotary unit ERD

Product Information. Miniature rotary unit ERD Product Informaton ERD ERD Fast. Compact. Flexble. ERD torque motor Powerful torque motor wth absolute encoder and electrc and pneumatc rotary feed-through Feld of applcaton For all applcatons wth exceptonal

More information

Operation Guide 3270/3293

Operation Guide 3270/3293 MO1109-EA Getting Acquainted Congratuations upon your seection of this CASIO watch. To get the most out of your purchase, be sure to read this manua carefuy. Be sure to keep a user documentation handy

More information

Remarks on The Logistic Lattice in Random Number Generation. Neal R. Wagner

Remarks on The Logistic Lattice in Random Number Generation. Neal R. Wagner Remarks on The Logistic Lattice in Random Number Generation Nea R. Wagner 1. Introduction Pease refer to the quoted artice before reading these remarks. I have aways been fond of this particuar random

More information

Operation Guide 2804

Operation Guide 2804 MO007-EA Getting Acquainted Congratuations upon your seection of this CASIO watch. To get the most out of your purchase, be sure to carefuy read this manua and keep it on hand for ater reference when necessary.

More information

25th DOE/NRC NUCLEAR AIR CLEANING AND TREATMENT CONFERENCE

25th DOE/NRC NUCLEAR AIR CLEANING AND TREATMENT CONFERENCE DEEP BED CHARCOAL FILTER RETENTION SCREEN IN-PLACE REPLACEMENT AND REPAIR Wiiam Burns and Rajendra Paude Commonweath Edison Company LaSae County Station Raymond Rosten and Wiiam Knous Duke Engineering

More information

Step 3: Select a Class

Step 3: Select a Class R Step 1: Roll Ablty Scores a. ndcate dce-rollng method (p. 13):. Roll 3d6 sx tmes, n order. 11. Roll 3d6 twce per ablty, select ether. 111. Roll 3d6 sx tmes and assgn to abltes as desred. V. Roll 3d6

More information

V (D) i (gm) Except for 56-7,63-8 Flute and Oboe are the same. Orchestration will only list Fl for space purposes

V (D) i (gm) Except for 56-7,63-8 Flute and Oboe are the same. Orchestration will only list Fl for space purposes Measure # 1 2 3 5 6 7 8 9 10 11 12 13 1 15 16 17 18 Intro Motve 1 A Motve 1 B - "Wakarathe" Dynamcs mp terraced dymancs to mm.7 p Moderately fast and lght (quarter=96) 3 G Mnor motve 1 con't mp melody

More information

Using wordless picture books in schools and libraries. Ideas for using wordless picture books in reading, writing and speaking activities

Using wordless picture books in schools and libraries. Ideas for using wordless picture books in reading, writing and speaking activities CfE eves Eary to Fourth (Ages 3-16) Using wordess picture books in schoos and ibraries Ideas for using wordess picture books in reading, writing and speaking activities Resource created by Scottish Book

More information

CASH TRANSFER PROGRAMS WITH INCOME MULTIPLIERS: PROCAMPO IN MEXICO

CASH TRANSFER PROGRAMS WITH INCOME MULTIPLIERS: PROCAMPO IN MEXICO FCND DP No. 99 FCND DISCUSSION PAPER NO. 99 CASH TRANSFER PROGRAMS WITH INCOME MULTIPLIERS: PROCAMPO IN MEXICO Elsabeth Sadoulet, Alan de Janvry, and Benjamn Davs Food Consumpton and Nutrton Dvson Internatonal

More information

The Traffic Image Is Dehazed Based on the Multi Scale Retinex Algorithm and Implementation in FPGA Cui Zhe1, a, Chao Li2, b *, Jiaqi Meng3, c

The Traffic Image Is Dehazed Based on the Multi Scale Retinex Algorithm and Implementation in FPGA Cui Zhe1, a, Chao Li2, b *, Jiaqi Meng3, c 3rd Internatonal Conference on Mechatroncs and Industral Informatcs (ICMII 2015) The Traffc Image Is Dehazed Based on the Mult Scale Retnex Algorthm and Implementaton n FPGA Cu Zhe1, a, Chao L2, b *, Jaq

More information

Fast Intra-Prediction Mode Decision in H.264/AVC Based on Macroblock Properties

Fast Intra-Prediction Mode Decision in H.264/AVC Based on Macroblock Properties Fast Intra-Predcton Mode Decson n H.264/AVC Based on Macroblock Propertes Abstract Intra-predcton s a wdely used tecnque n ntra codng. H.264/AVC adopts rate-dstorton optmzaton (RDO) tecnque to obtan te

More information

Automated composer recognition for multi-voice piano compositions using rhythmic features, n-grams and modified cortical algorithms

Automated composer recognition for multi-voice piano compositions using rhythmic features, n-grams and modified cortical algorithms Complex Intell. Syst. (2018) 4:55 65 https://do.org/10.1007/s40747-017-0052-x ORIGINAL ARTICLE Automated composer recognton for mult-voce pano compostons usng rhythmc features, n-grams and modfed cortcal

More information

FPGA Implementation of Cellular Automata Based Stream Cipher: YUGAM-128

FPGA Implementation of Cellular Automata Based Stream Cipher: YUGAM-128 ISSN (Prnt) : 2320 3765 ISSN (Onlne): 2278 8875 Internatonal Journal of Advanced Research n Electrcal, Electroncs and Instrumentaton Engneerng An ISO 3297: 2007 Certfed Organzaton Vol. 3, Specal Issue

More information

arxiv: v1 [cs.cl] 12 Sep 2018

arxiv: v1 [cs.cl] 12 Sep 2018 Powered by TCPDF (www.tcpdf.org) Neural Melody Composton from Lyrcs Hangbo Bao, Shaohan Huang 2, Furu We 2, Le Cu 2, Yu Wu 3, Chuanq Tan 3, Songhao Pao, Mng Zhou 2 School of Computer Scence, Harbn Insttute

More information

Running a shared reading project. A scheme of activities to help older children share picture books with younger ones

Running a shared reading project. A scheme of activities to help older children share picture books with younger ones CFE Leves Eary Senior phase (Ages 3-16) Running a shared reading project A scheme of activities to hep oder chidren share picture books with younger ones Resource created by Scottish Book Trust Contents

More information

Operation Guide 3150

Operation Guide 3150 MO0805-EA Getting Acquainted ongratuations upon your seection of this ASIO watch. To get the most out of your purchase, be sure to read this manua carefuy. Keep the watch exposed to bright ight The eectricity

More information

Loewe bild 7.65 OLED. Set-up options. Loewe bild 7 cover Incl. Back cover. Loewe bild 7 cover kit Incl. Back cover and Speaker cover

Loewe bild 7.65 OLED. Set-up options. Loewe bild 7 cover Incl. Back cover. Loewe bild 7 cover kit Incl. Back cover and Speaker cover Product nformaton Loewe bld 7.65 Page of March 07 Loewe bld 7.65 OLED EU energy effcency class: B Screen dagonal (n cm) / Screen dagonal (n nch): 64 / 65 Power consumpton ON (n W): 80 Annual energy consumpton

More information

User Manual. AV Router. High quality VGA RGBHV matrix that distributes signals directly. Controlled via computer.

User Manual. AV Router. High quality VGA RGBHV matrix that distributes signals directly. Controlled via computer. User Manual AV Router Hgh qualty VGA RGBHV matrx that dstrbutes sgnals drectly. Controlled va computer. Notce: : The nmaton contaned n ths document s subject to change wthout notce. SmartAVI makes no warranty

More information

LONG term evolution (LTE) has now been operated in

LONG term evolution (LTE) has now been operated in IEEE/ACM TRANSACTIONS ON NETWORKING 1 A Pricing-Aware Resource Scheduing Framework for LTE Networks You-Chiun Wang and Tzung-Yu Tsai Abstract Long term evoution (LTE) is a standard widey used in ceuar

More information

Product Bulletin 40C 40C-10R 40C-20R 40C-114R. Product Description For Solvent, Eco-Solvent, UV and Latex Inkjet and Screen Printing 3-mil vinyl films

Product Bulletin 40C 40C-10R 40C-20R 40C-114R. Product Description For Solvent, Eco-Solvent, UV and Latex Inkjet and Screen Printing 3-mil vinyl films Product Bulletn 40C Revson D, Effectve February 2016 (Replaces C, Apr. 15) 40C-10R 40C-20R 40C-114R Product Descrpton For Solvent, Eco-Solvent, UV and Latex Inkjet and Screen Prntng 3-ml vnyl flms Quck

More information

Topology of Musical Data

Topology of Musical Data Topoogy of Musica Data Wiiam A. Sethares Department of Eectrica and Computer Engineering, University of Wisconsin, Madison, USA, sethares@ece.wisc.edu November 27, 2010 Abstract Techniques for discovering

More information

Operation Guide

Operation Guide MO1603-EA 2016 ASIO OMPUTER O., LT. Operation Guide 5484 5485 Getting Acquainted ongratuations upon your seection of this ASIO watch. To get the most out of your purchase, be sure to read this manua carefuy.

More information