Coordination sheet:
- Google sheet ready - please request access and add nodes that will participate
- Content: hosts, HW specs, kernel, sysctl config, changes
Testbed hosts:
- No changes in Google sheet since the last meeting ? I have received and approved a number of requests to edit.
- List from first meeting: psb02-gva.cern.ch, psum06-100g.aglt2.org, star-dtn1.es.net, lbl-dev-dtn1.es.net, cern773-dtn1.es.net, ps02-b.farm.particle.cz, pygrid-sonar2.lancs.ac.uk, ps-london-bw.perf.ja.net, QMUL, BNL, RNP (1-2nodes)
- Missing coverage in Asia: Singapore, Australia, Taiwan ?
- Sites with shallow buffers: Candidates ? IEPSAS-Kosice maybe
Testbed commisioning:
- Come up with list of changes following an initial check of the testbed
- e.g. check if BBR can be enabled, check if perfSONAR commands work (with related changes)
- Develop a script that can check config (and sysctl parameters) that anyone can run before/after submitting tests (TBD)
- Test coordination - looking for test coordinator for each (or multiple) week(s) - first come first served !
- Understand how we collect, store and analyse measurements (we could use existing perfSONAR/ES infrastructure, but will need to store additional meta-data)
- ESnet high-touch testing - would likely require a CSV file to be sent in advance with start/end, src/dst (maybe ports ?) and traceroute in order to be able to start data collection
- High-touch produces pcap files that would need to be analysed further (we're looking for volunteers; python's pandas/polars + jupyterhub; data will be in apache arrow / feather format)
Methodology:
- Testing: CUBIC, BBRv3, TC and jumbo frames
- no tc: cubic/1500, cubic/9000, bbrv3/1500, bbrv3/9000
- tc (N-times for different bps): cubic/1500, cubic/9000, bbrv3/1500, bbrv3/9000
- i.e. 4*N+4 tests in each round (min 8 tests per round for given set of parameters)
. - parameters: single stream, multi-stream, multi-stream combined (cubic+bbrv3)
. - each round needs to run several times (to account for state of the network)