Gant Spacing Signaling at the ONU Glen Kame, Boadcom Duane Remein, Huawei May 2018 IEEE 802.3ca Task Foce, ittsbugh, A 1
Total Bust Size In 802.3ca, the OLT GATE message conveys only the payload length of the bust, not the total length with the optical ovehead. To be able to schedule consecutive busts, the DBA at the OLT should be able to calculate the total bust time, including the FEC and optical oveheads. To allow the deteministic calculation, the beginning of the fist envelope in a bust is always aligned to the beginning of the fist FEC codewod in this bust. With such alignment, the OLT can calculate the exact total bust size. Knowing the total bust size, the OLT DBA can schedule busts back to back. Stat Calculated Bust Size End Stat End Calculated Bust Size E1 FEC CW a E2 FEC CW a E3 FEC CW a E4 FEC CW a At the beginning of a bust, FEC codewod is aligned to fist Envelope Stat Heade At the beginning of a bust, FEC codewod is aligned to fist Envelope Stat Heade May 2018 IEEE 802.3ca Task Foce, ittsbugh, A 2
Total Bust Size Calculation Given: L total sum of ganted envelope lengths (in EQ): NN LL = eeeeee_llllllllllll ii ii=00 Calculations: 1. Numbe of 257b blocks taken by L EQs: = LL/44 S 1 RepeatCount T ON S 1 FEC unpotected aea T settling S 3 RepeatCount S 2 RepeatCount T CDR SBD S 2 S 3 FEC payload (56 256b/257b tanscoded/scambled blocks) FEC potected aea (N FEC codewods, last codewod may be shotened) FEC CW 1 FEC CW 2 FEC CW N FEC paity (10 257b blocks) 0 1 54 55 0 9 FEC unpotected aea T Length EBD T T OFF 2. Numbe of FEC codewods (including possible shotened last codewod): N = /5555 3. Total numbe of 257b blocks taken by all FEC codewods: M = + N 10 4. Bust size (in 257b blocks): S = S1_Count + S2_Count + S3_Count + M +T_Length + [Toff] To convet S into the numbe of EQ times (2.56ns): S EQ = S 257 / 66 To convet S into ns: S ns = S 8224 / 825 May 2018 IEEE 802.3ca Task Foce, ittsbugh, A 3
Ovelapping Gants to the same ONU Bad things happen if the OLT gants two ovelapping busts (i.e., Bust #2 StatTime < Bust #1 EndTime) End E1 FEC CW a E2 FEC CW a At the beginning of a bust, FEC codewod is aligned to fist Envelope Stat Heade Stat E3 FEC CW a E4 FEC CW a Stat At the beginning of a bust, FEC codewod is aligned to fist Envelope Stat Heade End End E1 a E2 IEI a IEI a E3 a E4 a FEC CW FEC CW FEC CW FEC CW FEC CW At the beginning of a bust, FEC codewod is aligned to fist Envelope Stat Heade Stat of E3 envelope is not aligned to FEC CW stat. As a esult, the second gant spans 3 FEC codewods instead of expected 2 Stat of E3 envelope (second bust) is not aligned to FEC CW stat. As a esult, in this example, the second gant spans 3 FEC codewods instead of the expected 2. This causes the end of the combined bust expand past the expected end time by one full paity egion (10x257b blocks o ~100 ns) May 2018 IEEE 802.3ca Task Foce, ittsbugh, A 4
To concatenate the gants o not? Gant concatenation capability was somewhat impotant in 10G-EON because evey LLID was ganted independently, even if they wee in the same ONU. This caused a vey significant incease in optical ovehead. In 802.3ca, the gant is given to an entie ONU and is expected to include multiple LLIDs. Thee ae fewe busts needed pe unit of time, and coespondingly, much smalle levels of optical ovehead. Cuent vesion of MRS in D1.0 pevents two sepaate gants fom being joined into a single bust (except fo some ae cone cases). OLT bette not schedule the upsteam tansmission this way, but the DBA/schedule is out-of-scope Instead, pe D1.0, the ONU is equied to ignoe/discad the gants that may not get aligned to stat of FEC codewod due to being concatenated with the pevious gant. May 2018 IEEE 802.3ca Task Foce, ittsbugh, A 5
Cuent Solution Cuent vesion of MRS/MC in D1.0 pevents two sepaate gants fom joining into a single bust. This functionality was achieved in a somewhat convoluted way Evey time MRS Input SD doesn t have an active envelope, it asks MC fo one using MRS_CTRL.indication(X) pimitive, whee agument X epesents the space (in EQs) left in the cuent FEC codewod. MC Envelope Activation SD would pocess such indications and occasionally issue a equest to stat an envelope. If the new envelope is a continuation of existing bust, it can stat immediately when MRS asked fo it, egadless of the value of X. But if the next envelope stats a new bust (i.e., a diffeent StatTime than in the pevious envelope), then MC issues the equest only if this indication coincides with the beginning of FEC codewod (i.e., if X = FEC_CW_SIZE). This ensues that stat of evey bust payload is aligned with stat of a FEC codewod. May 2018 IEEE 802.3ca Task Foce, ittsbugh, A 6
Why to change it Cuent solution equied MRS Input SD and MC Envelope Activation SD to be FEC awae and to tack the space emaining in cuent FEC codewod. MRS Input SD was not able to detect and signal CS when a bust ended until afte a long gap elapsed since the end of the last envelope. The CS had to detect the end of bust on its own by adding a deep look-ahead buffe, that contibuted to the intenal ONU s delay. All of this can be done in a moe steamlined way by allowing MC to signal MRS when the bust has ended. May 2018 IEEE 802.3ca Task Foce, ittsbugh, A 7
How to make this wok? The changes ae quite simple: MC Envelope Commitment pocess Responsible fo queueing envelope desciptos fo late activation. Change: afte enqueueing all envelope desciptos fo one bust, add one special Teminating Envelope Descipto (TED) LLID = 0 Length = GRANT_MARGIN (GRANT_MARGIN = aity + EBD + Ton + S1 + S2 + S3) MC Envelope Activation ocess Responsible fo sending a new envelope descipto to the MRS when MRS asks fo it and if the time is ight. Change: SD became simple, since it doesn t need to tack the FEC codewod offset in the MRS to find out whethe a new bust can o cannot stat at a given time. MRS Input ocess A key pocess esponsible fo soucing EQs fom pope MACs based on envelope desciptos fom the MC. Change: Added an additional state to pocess TEDs. Details ae on slide #13. The pocess doesn t need to be FEC-awae. May 2018 IEEE 802.3ca Task Foce, ittsbugh, A 8
MC Envelope Commitment SD Special Teminating Envelope Descipto (TED) is inseted in pending envelope queue afte committing all the envelopes fo a given bust (all envelopes with the same stat time) In the OLT, a TED neve gets inseted, so a downsteam bust neve teminates. The Envelope Activation SD and all of MRS SDs can emain identical in the OLT and ONUs. UNREGISTERED ChIndex 0 ChIndex = 0 Registeed BEGIN!Registeed WAIT_FOR_ENV_DESCRITORS EnvIndex 0 CLEAR_ENV_DESCRITORS EnvList[ChIndex].Clea() ChIndex++ STORE_ENV_DESCRITOR Env.LLID EnvLlid[EnvIndex] Env.StatTime EnvStatTime Env.Length EnvLength[EnvIndex] EnvList[ChIndex].Append(Env) EnvIndex++ EnvIndex < EnvCount MCC:MA_CONTROL.equest( ChIndex, EnvStatTime, EnvCount, EnvLlid[EnvCount], EnvLength[EnvCount]) * Registeed * EnvCount > 0 BURST_COMLETE Env.LLID 0x00-00 Env.StatTime EnvStatTime Env.Length GRANT_MARGIN EnvList[ChIndex].Append(Env) UCT May 2018 IEEE 802.3ca Task Foce, ittsbugh, A 9
Changes to Envelope Activation SD Don t need to be awae of FEC CW alignment No need to eceive FecOffset fom MRS Don t need to keep tack of StatTime of pevious envelope Simple tansition into ACTIVATE_ENVELOE state. May 2018 IEEE 802.3ca Task Foce, ittsbugh, A 10
New Envelope Activation SD BEGIN The same SD is used in the OLT and ONUs WAIT_FOR_CHANNEL UCT INIT Back2BackEnv[ch] false Added tansition fom ACTIVATE_ENVELOE TO INIT to explicitly signal the end of bust, so that the next bust can only stat when its StatTime == LocalTime MRS_CTRL[ch].indication() CHECK_ENDING_ENVELOES IsEmpty( EnvList[ch] ) GET_NEXT_ENVELOE NextEnv EnvList[ch].eekHead() IsValid(NextEnv.LLID) CHECK_START_TIME WaitTime NextEnv.StatTime - LocalTime (non-essential changes ae not shown) UCT DISCARD_ENVELOE EnvList[ch].RemoveHead() WaitTime > 0 WaitTime = 0 + Back2BackEnv[ch] ACTIVATE_ENVELOE MRS_CTRL[ch].equest(NextEnv.LLID, LocalTime<5:0>, NextEnv.Length) EnvList[ch].RemoveHead() Back2BackEnv[ch] tue NextEnv.LLID = 0x00-00 May 2018 IEEE 802.3ca Task Foce, ittsbugh, A 11
MRS Input SD Added state TX_INACTIVE to pocess TEDs New behavio: When TED is eceived, MRS Input ocess stats geneating Inte-Bust Idles (_EQ) towad the 25GMII and it does not ask MC fo a new envelope until the envelope length woth of _EQ is geneated. Anothe (non-essential) change is to ename FEC-elated constants (MRS as a laye does not need to be FEC awae, but instead should be capable of ate adjustment.) ARITY_LACEHLDR RATE_ADJ_EQ FEC_CW_SIZE ADJ_BLOCK_SIZE FEC_ARITY_SIZE RATE_ADJ_SIZE NEXT_COL wcol++ wcol 0 IN_CLK ADJUST_RATE InEQ RATE_ADJ_EQ BEGIN UCT TxActive[wCol] * BlkLeft[wCol] RATE_ADJ_SIZE CHECK_HEADER MRS_CTRL[wCol].equest(link_id,epam,env_length) * env_length > 0 CHECK_ENV_SIZE InEQ GetFilleEQ(wCol) NEXT_ROW wrow++ Envam++ EnvLeft[wCol] > 0 TX_ENV_DATA InEQ GetMacBlock(LinkId[wCol]) InEQ = REAMBLE_EQ RELACE_REAMBLE InEQ EnvContHeade(wCol) REQUEST_NEXT_ENV MRS_CTRL[wCol].indication() RESET_CODEWORD BlkLeft[wCol] ADJ_BLOCK_SIZE UCT UCT START_NEW_ENVELOE LinkId[wCol] link_id EnvLeft[wCol] env_length InEQ EnvStatHeade(wCol, epam) TxActive[wCol] tue LinkId[wCol] 0x00-00 TX_INACTIVE TxActive[wCol] false InEQ GetFilleEQ(wCol) UDATE_ENV_SIZE EnvLeft[wCol]-- INIT EnvLeft[] {0} BlkLeft[] {ADJ_BLOCK_SIZE} TxActive[] {false} wrow 0 wcol 0 UCT EnvLeft[wCol] > 0 WRITE_EQ_TO_TX_FIFO ENV_TX[wCol][wRow] InEQ BlkLeft[wCol]-- May 2018 UCT BlkLeft[wCol] > 0 * TxActive[wCol] IEEE 802.3ca Task Foce, ittsbugh, A 12
Functions New EQ GetFilleEQ( wcol ) { if( TxActive[wCol] ) etun IEI_EQ; //Inte-Envelope Idle etun _EQ; //Inte-Bust Idle } Modified EQ GetMacBlock(link_id) { EQ eq; if( link_id == 0x00-00 ) etun _EQ; // Inte-bust Idle } fo( octet_index = 0; octet_index < 8, octet_index++ ) { // as in D1.0 } etun eq; May 2018 IEEE 802.3ca Task Foce, ittsbugh, A 13
Summay (1/2) Bust spacing is a function of DBA/schedule, which is outof-scope fo 802.3ca. We can have infomative text explaining that bust spacing should be at least GRANT_MARGIN. But the fomal enfocement is done by the ONU MC/MRS pocesses (in scope). Simple changes to MC to inset Teminating Envelope Desciptos (TED) at the end of each eceived gant. Simple changes to MRS to pocess TEDs and to geneate Inte-Bust-Idle codes between busts Intoduction of Inte-Bust-Idles also temendously simplified the CS state diagams (see emein_3ca_3_0518.pdf) May 2018 IEEE 802.3ca Task Foce, ittsbugh, A 14
Summay (2/2) The diffeences in the OLT and ONU behavio ae confined to the MC laye. MRS and CS state diagams still emain identical between the OLT and ONUs. Only in the OLT MC Envelope Commitment SD at the OLT neve geneates TED As a esult, OLT MRS neve insets _EQ As a esult, OLT CS neve tuns the lase off. May 2018 IEEE 802.3ca Task Foce, ittsbugh, A 15
Motion #x Accept state diagams and function definitions as pesented on slides 9-13 of Kame_3ca_2_0518.pdf. May 2018 IEEE 802.3ca Task Foce, ittsbugh, A 16
Thank You May 2018 IEEE 802.3ca Task Foce, ittsbugh, A 17