1. 3rd April 2001A.Polini and C.Youngman1 GTT status Items reviewed: –Results of GTT tests with 3 MVD-ADC crates. Aims Hardware and software setup used Credit control implementation Latency and rate results - test 1 Latency and rate results - test 2 System acceptability –GTT time schedule and requirements on ZEUS DAQ.

2. 3rd April 2001A.Polini and C.Youngman2 Results of tests with 3 MVD-ADC crates Aims –Implement the credit controlled GTT described in ZEUS Note 99- 74 with data input from 3 MVD-ADC crates. –Measure latency and rate characteristics in the absence of algorithm data processing in week available. –Make a statement about the acceptability of the system w.r.t. the ZEUS DAQ. Hardware and software setup used –parts of the MVD test system as of 23/3/01 3 complete and fully debugged ADC/readout crates. Clock&Control system incl. Pulse generator ext. trigger. –7 Dual 1GHz Intel PCs –1 Intel 12 port giga/fast ethernet switch. PPC VME (fast) and PC(giga) CPUs connected to separate ports –MVD run control and monitoring software.

3. 3rd April 2001A.Polini and C.Youngman3 GTT credit control implementation keypoints –code derived from a non hardware simulation allows RC configurations to be tested allows message and IP connectivity to be debugged –uses a common function library for buffering ordering of events with GFLT# before sending to GSLT/EVB gathering complete events at GTT and EVB –network messages sent consists of: an short control message (<128B) in XDR tagged union. –endian independent –contents (GFLT#,etc.) remove the need to look into the data and, optionally, a data message. –all programs are currently single threaded.

4. 3rd April 2001A.Polini and C.Youngman4 111 2 222 3 1 1 1 3 4 21 2 224 3 GFLT ACCEPT GSLT DECISION EVB GSLT MVD-ADC STT-SLTCTD-SLT GTT TOEVBTOGSLTFROMGSLT 11 3 2 687540/61 12 GTT process, network and trigger connections - credit based.

5. 3rd April 2001A.Polini and C.Youngman5 Control message definitions Credit –0. credit number allocation (SETUP) –1. credit/socket resolution (SETUP) –2. credit list (SETUP) and credit notification (ACTIVE) Trigger –3. SLT data to GTT (ACTIVE) –4. GTT end algorithm credit notification (ACTIVE) –5. GTT algorithm result to GSLT (ACTIVE) –6. GSLT decision to MVD and specific GTT (ACTIVE) EVB –7. GTT result and MVD cluster data (ACTIVE) –8. MVD strip data (ACTIVE)

6. 3rd April 2001A.Polini and C.Youngman6 111 2 222 3 1 1 1 3 4 21 2 224 3 GFLT ACCEPT GSLT DECISION EVBGSLT MVD-ADC STT-SLTCTD-SLT GTT TOEVBTOGSLTFROMGSLT 11 3 2 687540/61 12 GTT hardware realization

7. 3rd April 2001A.Polini and C.Youngman7 Latency and rate results - test 1 Fix SLT & EVB data size/crate and vary N GTT –measurement parameters SLT size/crate 2.6, 2.7 and 1.9 kB EVB size/crate 4.6, 4.6 and 3.3 kB vary N GTT through: 1,2,4,8,12,18,24 and 30 (ie. 5/PC). Pulse generator rate 600 Hz. GSLT accept prescale factor 10 –results are independent of N GTT Sustained rate 500 Hz mean with 60 Hz p2p fluctuation. Latency at TOGSLT 1.7ms with low level but long tail. Latency at TOEVB 2.9ms - did not look at tail. –conclusions credit turn round of 1.7 ms –message transit time ~0.9ms (cf. Table 4 TCP PPC->PC performance tests = 3/4245 + 2/12306) N GTT independence surprising especially for 1 GTT, understood ? rate and fluctuation not understood, credit bursting ?. latency tails not understood, more detailed tests required.

8. 3rd April 2001A.Polini and C.Youngman8 GTT latency at TOGSLT

9. 3rd April 2001A.Polini and C.Youngman9 Latency and rate results - test 2 Fix N GTT and vary SLT & EVB data size/crate –measurement parameters pulse generator rate600 Hz GSLT accept prescale factor 10 N GTT fixed at 6, ie. 1/PC –results –conclusions bigger SLT/EVB data sizes have lower rates and higher latencies ! too early to say how stable the measured values are.

10. 3rd April 2001A.Polini and C.Youngman10 Acceptability of GTT performance Non algorithm processing results –3 MVD-ADC crate system the measured rates (>500Hz) for SLT and EVB data sizes of <2.5kB/crate and <4.5kB/crate are close to the ZEUS DAQ requirement of ~550Hz. the measured SLT and EVB latencies appear to be stable and acceptable if the tails can be understood and reduced further. –CTD-SLT connection The measured latency of the CTD-SLT at the GSLT was measured in the ZEUS DAQ system in 1999. The CTD-SLT data is available a few ms after the GFLT accept thus network contention at the GTT is not anticipated. Conclusions –The measurements indicate that the GTT will work.

11. 3rd April 2001A.Polini and C.Youngman11 Time schedule and ZEUS DAQ requirements MVD related –week 17 (23/4) ? GFLT trigger interface to C&C EVB related –week 18 (30/4) ? tests of PC/PCI-TP interface needed for sending SLT result receiving GSLT decision –week 19 (7/5) ? tests of MVD event building via ethernet –by week 19 (7/5) DDL definitions of data banks GSLT related –by week 19 (7/5) definition of GTT decision –week 19 (7/5) ? tests GSLT connection DAQ chain –week 20 (14/5) and thereafter tests with full DAQ system including the new EVB subsystems.