Description : Level 2 Workshop goals and ground rules From : FNALD0::LINNEMANN Posted : 13-APR-1998 14:57:43, Expires : 28-APR-1998 14:57:43 Newsgroups: d0.upgrade From: FNALD0::LINNEMANN X-Tcp-From: linnemann@d0msu7.fnal.gov Subject: Level 2 Workshop goals and ground rules Date: 13 Apr 98 08:57:43 GMT Expires: 28 Apr 98 08:57:43 GMT Path: d0tng.fnal.gov!d0bnl3.fnal.gov!d0msu7.fnal.gov!vaxnews!LINNEMANN Message-ID: <00028908_VAXNEWS@d0msu7.fnal.gov> X-Newsreader: VaxNews 3.22 In preparation for the L2 workshop at UIC on 24/25 April, I will be sending several communications. This first gives the ground rules and goals. Ground Rule =========== Alpha and MBT base functionality frozen. Workshop Goals =============== 1) Freeze what FIC means (and thus MBT and SLIC inputs). 2) Consider possible commonalities between STT and rest of L2: Alpha and MBT cards proceed independent of STT considerations FIC may be influenced by STT requirements Muon and STT may both influence SLIC design if sharing is advantageous I will soon send out for comments background information and the various proposals on both these points, based on the discussions, meetings, and email exchanges to date. Jim -------------------------------------------------------------------------------- Description : Proposals for L2 Inputs From : FNALD0::LINNEMANN Posted : 15-APR-1998 21:39:39, Expires : 30-APR-1998 21:39:39 Newsgroups: d0.upgrade From: FNALD0::LINNEMANN X-Tcp-From: linnemann@d0msu7.fnal.gov Subject: Proposals for L2 Inputs Date: 15 Apr 98 15:39:39 GMT Expires: 30 Apr 98 15:39:39 GMT Path: d0tng.fnal.gov!d0bnl3.fnal.gov!d0msu7.fnal.gov!vaxnews!LINNEMANN Message-ID: <00028920_VAXNEWS@d0msu7.fnal.gov> X-Newsreader: VaxNews 3.22 L2 Input Proposals ================== I) baseline L2 proposal as outlined in Jim's "Standard Crate" talk 160 MHz Cypress inputs to SLIC and MBT FIC (G-link in, Cypress out) with 16-event buffer only for G-link data = L1CFT/PS data, assumed 16-bit format, following L2 protocols G-link raw PS data an escalation, requiring event tagging (clock or SCL) G-link raw SVX data handled on STC card SFO card with SCL in and Cypress out to SLIC's II) Saclay proposals (separate message to follow later) i) backplane inputs ("FIC") to ALL MBT and SLIC via J2 connector not just if G-link input ii) above, with FIFO's on FIC as well replace SFO card: MBT sends SCL info to SLIC's via backplane bus III) Workshop Strawman (variant of French proposals: Baden/Johnson) i) FIC inputs to ALL MBT and SLIC via J2 connector some protocol details below iii) Cypress 212 MHz Fiber standard throughout L2 drop speed of L1 CFT/ L1 PS outputs to Cypress speeds convert muon to 212 MHz Fiber iv) SCL receiver on each SLIC Here are more details of the Workshop Strawman: =============================================== L2 inputs: ---------- SCL 53MHz, to be defined whether copper or fiber raw data 16-bit G-link fiber, 16b frames at 53MHz (STT, possibly PS) ALL others 8-bit Cypress fiber, 8b frames at 53MHz/2 (212 MHz serial) Including: L1 muon requires 160-212 MHz conversion or driver change L1CFT/PS CFT: 20b padded to 32b/track slower transfers to STT (move STC buffers) L1 Cal, L1HWFW Data padded to multiples of 16B in event; event # in L2 header/trailer Cypress Fiber protocol a subset of Muon Cypress copper protocol: (same) end of event symbol: required; this is a control value, NOT a data value other symbols: ignored L2 preprocessor components: --------------------------- FIC (Fiber Input Card) Attaches to the back of the crate Type A: Receives 4 16-bit Glink (1.3 GHz fiber) FIC provides event tag and as much as practical of L2 header/trailer Type B: Receives 16 Cypress hotlink (212 MHz fiber) Type C: Receives 16 Cypress hotlink (160 MHz copper) - current muon spec L2 would prefer not to make type C A future ECB can be used to resolve the issue FIC performs serial-to-parallel conversion A mimimal FIFO exists for each input channel to couple clocks. Data transfered through J2 at 64-bits x 53 MHz similar to VRB spec. see http://www-cdf.fnal.gov/upgrades/daq_trig/daq.html Control lines on J2 might include clock 3-bit FIC type ID for each of 8 channels (from FIC): End_Event Data_Present Data from type A is sent as 16-bits from each of 4 inputs. Data from 16-input FIC types are mux'ed by 2 to 64-bits. SLIC (Second Level Input Computer) Receives data through J2 from a FIC of either type. 16-deep input fifos for each input channel. FPGA routing/filtering of data from fifo to DSPs 4 TI C6x integer DSPs per card. 2 output fifos and Cypress hotlink drivers (212 MHz fiber) VME accessible output memory. 5th DSP stores input data on Unbiased flag for VME readout. SCL mezzanine and filtering on SLIC MBT (Magic Bus Transceiver) Receives data through J2 from a FIC of either type. 16-deep input fifos for each input channel. 128-bit MBus connection. Serial Command Link connection. 2 output fifos and cypress hotlink drivers (212 MHz fiber) ----------------------------------------------------------------------- Description : proposals for using L2 components to implement L2 STT From : FNALD0::LINNEMANN Posted : 15-APR-1998 21:40:26, Expires : 30-APR-1998 21:40:26 Newsgroups: d0.upgrade From: FNALD0::LINNEMANN X-Tcp-From: linnemann@d0msu7.fnal.gov Subject: proposals for using L2 components to implement L2 STT Date: 15 Apr 98 21:40:26 GMT Expires: 30 Apr 98 21:40:26 GMT Path: d0tng.fnal.gov!d0br06.fnal.gov!d0msu7.fnal.gov!vaxnews!LINNEMANN Message-ID: <00028921_VAXNEWS@d0msu7.fnal.gov> X-Newsreader: VaxNews 3.22 How L2 components might be used in STT ====================================== a) STT baseline * L1 CFT input on G-link via J3 (as in VTM card) * Fiber road card fans out to STC's on J2 road bus * STC FPGAs perform cluster and road match * clusters to fitter card on 40 MHz J3 output (cluster) bus fitter card track output: Cypress hotlink (160 MHz) to L2CFT for L2 data VME accessible output memory for L3 readout of full data set Storage of input data on Unbiased Event Qualifier for VME readout. VBD for readout; fiber road card as controller a') STC (Silicon Trigger Card, as influenced by Workshop Strawman) STC receives SMT data through J2 from a FIC. 16-deep input fifos for each input channel (location moved) L1CFT information arrives 212 MHz fiber instead of G-link output Cypress hotlink now 212 MHz fiber b) "Fortner Proposal"--Clustering in SLIC FIC type A + raw data event tagging on FIC (or SLIC) + fanout of L1 CFT information to SLIC complexity rises if limited to 4 G-link inputs per SLIC SLIC does clustering; Fitting done in Alpha Bit3 + Alpha Administrator for download/monitoring c) "Workshop Strawman Proposal" Clustering in STC as a' Transfer of clusters via Cypress Fiber 212 MHz to FIC B/SLIC 1 crate = 9 STC cards + 2 SLIC's per STT crate + fiber road card Fitting in SLIC some issues: transfer times in and out of STC passive fanout enough for L1 CFT information distribution connectivity of SLIC: assignment of DSPs to tracks timing and number of DSP's per card download/control/monitoring/readout: Fiber Road card vs Bit3/Alpha Admin Other possible areas of commonality between STT and rest of L2 similarity of road card and SCL Fanout? if raw input needed for preshower use FIC A use STC with Fiber output (from a') above) as/instead of FIC A? ------------------------------------------------------------------------------ Description : Steps to a decision on L2 Inputs; comments welcome From : FNALD0::LINNEMANN Posted : 15-APR-1998 21:41:03, Expires : 30-APR-1998 21:41:03 Newsgroups: d0.upgrade From: FNALD0::LINNEMANN X-Tcp-From: linnemann@d0msu7.fnal.gov Subject: Steps to a decision on L2 Inputs; comments welcome Date: 15 Apr 98 15:41:03 GMT Expires: 30 Apr 98 15:41:03 GMT Path: d0tng.fnal.gov!d0cd03.fnal.gov!d0msu7.fnal.gov!vaxnews!LINNEMANN Message-ID: <00028922_VAXNEWS@d0msu7.fnal.gov> X-Newsreader: VaxNews 3.22 Steps to Input/ FIC decision and SFO decision ============================================= Here I list some possible decisions we might take, leading roughly from the present situation toward the most aggressive alternatives. I also list what I perceive as advantages and disadvantages of these decisions. I invite corrections and additions. Part of our discussions at the workshop will involve weighing these advantages and disadvantages. You will note that the advantages/disadvantages are not just performance: the lists include impacts on cost, complexity, and particularly schedule. We really need to DECIDE what we are going to build. So a potentially better configuration which requires complex negotiation or longer design time is disadvantageous compared to a less than optimal solution which does meet requirements, and we can start building now. Input Format (Current) ---------------------- DATA Throughput 160MHz Cypress Copper Muon SLIC only 16MB/s 160MHz Cypress Fiber L2 (protocol subset of Muon) Global, L2 Cal 16MB/s "1.3 GHz G-link": 24 bit frames X 53 MHz = 1.3 MHz) 20 bit data G-link L1 CFT 1060MB/s 16 bit data G-link L1 CPS/FPS 848MB/s 16 bit data G-link raw data (STT inputs, POSSIBLY L2 PS inputs) 848MB/s special: lacks event number tagging header, not integer multiple of 16 Byte) in addition, there is 1.3 GHZ AMCC (53 MHz X 24 bit?) copper (or fiber?) for SCL (thus on SCL mezzanine) copper in the L1 muon (and from L1 CFT); not seen by L2 Format Decisions ---------------- L2 use Fiber Cypress 160 MHz rather than copper 160 MHz + easier to make work than copper + better ground isolation??? (if L2 in fixed counting house) - L2 muon already using copper - careful handling needed ( 1dB loss from fingerprint) - $30/channel more cost Raise L2 Cypress Fiber 160 MHz to 212 MHz + more bandwidth than 160 MHz (+ 1/3) + avoid extra clock: multiple of 53 MHz (SCL mezzanine, and G-link) - less noise margin Drop 20-bit G-link mode + remove one type of input - transmission 1.6X slower (20 bit to 32 bit) probably only for L2 STT (elsewhere, translated to 32b anyway) Lower L1 CFT/PS output from 1.3 GHz Glink to 212 MHz Cypress Fiber + no need for "classic FIC" except for raw data + system uniformity - L1 CFT arrives later at L2STT (more/earlier buffering? readout harder?) - L1 CFT needs output buffer - cost? own G-links already? Force Muon to fiber + grounding, fewer input types - add functionality to L2 protocol, or lose it in muon - money for cables, converter cards, or change (5?) output cards Layout Decisions ---------------- Move inputs to backplane transition card ("FIC") on J2 + more real estate on SLIC/MBT + protect cables + replace/ test SLIC/MBT cards without disturbing cabling + money and engineering from Saclay (allows financial contribution) - MBT, SLIC must negotiate backplane protocol with Saclay - expense: 2 cards wherever 1 card before (FIC not just for G-link inputs) - more groups required to commission/maintain Force common standard on backplane for MBT, SLIC + flexibility (if have multiple input types) - negotiation time risk Analog through backplane + decouples SLIC/MBT from input type + most flexibility - engineering risk, especially at high speeds Digital through backplane + recoup some of VTM engineering especially for raw data - SLIC/MBT must handle multiple input protocols - negotiation time for protocol details, functionality Force common standard on backplane also for STC (FPGA card) [only works for digital through backplane; requires 212 Cypress Fiber] + flexibility: FPGA card more a "component", e.g. for receiving raw data or could use input section as FIC if raw data needed for L2PS - extra layout: VTM uses J3 now G-link FIC with FIFO and Cypress output ("Classic FIC") + isolates receiving card from G-link input + build only where needed - complex if receiving raw data: event tag - extra FIFOs in system to monitor and control Implement G-link FIC as freestanding card + allows > 4 G-link inputs to SLIC - extra slot somewhere for every 4 G-link inputs FIFO on FIC side of transition card + standard building block + FIC for G-link less special (no FIFO just to shift speed) - negotiation time: more control required across backplane SCL Fanout to SLIC ------------------ L1 SCL information: SCL Initialize, L1 Qualifiers, (part of) L1 Accept turn number [All solutions could recoup SCL handling engineering from other cards] Cypress output from special card ("Classic SFO") + uses simple interface (Cypress) + naturally FIFO'd for processing of event - extra card to build just for SLIC crates - if needed to steer processing, must be treated differently from other in's Backplane Bus from MBT + eliminate SFO card + naturally distinguishes SCL input from standard inputs - need customized backplane wiring J2 pins getting scarce if digital; noise worry if analog? - more functions on MBT - must define bus protocol Road Card instead of/as SFO card + eliminate SFO card design + possibly shared function with L1 CFT fanout? - conflicts with MBus on J3 if road fanout in STT crate on J3 SCL receiver on SLIC + eliminate SFO card - $500 per SLIC for mezzanine - more real estate and functionality on SLIC card - bend SCL protocol even further Event tagging ------------- [requires new interface with FIFO, no matter which scheme] local clock control + same scheme as muon (and L1 CFT) - new interface not needed by other cards in crate - need VME interface to manage, resynchronize? or self-synch every turn? inject L1 SCL information into FIFO's as header + need filtered L1 SCL information to process event anyway + automatically arrives before new event and after old one [critical issue ONLY for G-link SVX II chip raw data input and L2 STT using SLIC for clustering ("Fortner" scheme) STC card scheme uses Road Card to fan out SCL information or L2 PS needs raw data