EJFAT Group Meeting Jun. 16, 2022

From epsciwiki
Jump to navigation Jump to search

The meeting time is 11:00am.

Connection Info:

You can connect using ZoomGov Video conferencing (ID: 161 012 5238). (Click "Expand" to the right for details -->):

Meeting URL
 https://jlab-org.zoomgov.com/j/1610125238?pwd=QnEvcjV6VFFndWZsQW15SmJKU0RJZz09&from=addon

Meeting ID
161 012 5238

Passcode
503371

Want to dial in from a phone?

Dial one of the following numbers:
US: +1 669 254 5252 or +1 646 828 7666 or +1 551 285 1373 or +1 669 216 1590 or 833 568 8864 (Toll Free)

Enter the meeting ID and passcode followed by #

Connecting from a room system?
Dial: bjn.vc or 199.48.152.152 and enter your meeting ID & passcode

Agenda:

  • Previous meeting
  • Status:
    • Using ESnet FPGA f/w build 28 April
      • Specs
      • Jumbo Frames
      • arp, ping, ICMP filtering
      • Port entropy
    • Script based LB Control Plane
    • Support C libraries for LB Host Control Plane - in unit test code review legal review
    • ESnet smartnic open-source GitHub repo - in legal review
    • ESnet private, forkable Jlab P4 and simulations GitHub repo - in legal review
    • ERSAP feed end bottleneck needs investigation; Timmer's blaster may provide relief
    • New machines (6) rec'd, installed w/ Ubuntu 20.04 on EJFAT subnet (VLAN 937 172.19.22.0/24)
    • Spare EJFAT equip loaners:
      • (4) DAQ dev machines indra-s[1-3] 129.57.29/109.23[0-2]
        • alkaid: 24 Xeon Gold 3.4 GHz cores, 100Gbs
        • indra-s1: 24 Xeon Gold 3.0 GHz cores, 100Gbs
        • indra-s2: 32 Xeon Gold 3.2 GHz cores, 100Gbs
        • indra-s3: 32 Xeon Gold 2.3 GHz cores, 100Gbs, 750GB ram disk
      • (4) DAQ Farm machines dafarm6[1-4] currently on 129.57.29.17[1-4] - each 32 Xeon 2.0Ghz cores - 1 Gbs NIC + (4) 10Gbs Spare NICs
      • (4) Unbuilt DAQ Farm machines - each 32 Xeon 2.0Ghz cores - 1 Gbs NIC + (4) 10Gbs Spare NICs
      • PR408549 (6) 100Gbs NICs - ETA 1 July
      • PR408870 PR408938 (2) 100Gbs Arista switches, transceivers, cables, etc - ETA 1 July 5 October
  • Next Steps:
    • EJFAT VLAN Checkout
    • Network Performance:
      • FPGA LB Throughput - max sustained 90Gbs
      • Host NICs
      • Host S/W Reassembly - better algorithms, buffering, asynchronicity, etc.
      • EJFAT UDP Transmission Performance
      • Need better parameters for event reassembly/reconstruction
    • Control Plane
      • Will interact with SLURM / Kubernetes
      • Python based (?)
      • Control Plane daemon for compute host (?)
      • Demonstrate CP based flexibility/elasticity
    • Look at iperf2 for network testing
    • Look at ROCE / NIC
    • DPDK - ESnet reports can stream 100 Gbps using DPDK.
    • SLURM env for EJFAT VLAN (Hess)
    • DAQ/VTP Data Generation Test Harness
    • Vivado Licesnses for new machines (?) (Singh)
    • ACAT 2022 - September/Italy - Abstract / Paper
    • RT2022 - August 01-05 Conference
    • RT 2022 Paper
  • Back Burner / Downstream:
    • Hall-B FT calorimeter and hodoscope streaming readout test
      • May be able to use Abbott's indra-s1 setup
      • May be able to use new VTP f/w with Hall-B VTP's
      • CODA 3.10 + ERSAP for new VTP f/w
      • CODA 2.0 (non-streaming) for old VTP f/w
      • Diagram
      • Hall-B to start taking data June 8
      • Hall B VTPs on .167. subnet
    • HOSS
      • parallelize writing of raw data files
      • distribute raw data across multiple compute nodes for calibration skims
      • 1 Gbs at hi-luminosity
      • Hall-D comms with DAQ 109 subnet require network customization; (EJFAT subnet)
      • Hall-D EJFAT use case
      • Hall-D EJFAT Network Diagram
    • IPV6 testing
  • AOT

Notes:

  • numa tools on ubuntu:
    • sudo apt install hwloc-nox
    • sudo apt install numactl
  • lstopo
  • Open Nic Shell
  • To control the scheduling class, you can use the "chrt" command. To pin to CPUs, use the "taskset" command. Or use the underlying syscalls.
  • kernel dameon threads
    • handle NIC driver interrupts
    • set scheduling class to sched_rt of reassembly process
  • want to set cpu socket of reassembly in common NUMA domain as NIC
  • DPDK will own NIC bypassing kernel, etc.; packetgen_dpdk