16:59:15 <ahf> #startmeeting Network team meeting, 7 March 2022
16:59:15 <MeetBot> Meeting started Mon Mar  7 16:59:15 2022 UTC.  The chair is ahf. Information about MeetBot at http://wiki.debian.org/MeetBot.
16:59:15 <MeetBot> Useful Commands: #action #agreed #help #info #idea #link #topic.
16:59:17 <ahf> hello hello
16:59:20 <nickm> whoa hi!
16:59:21 <Diziet> Afternoon
16:59:34 <eta> o/
16:59:41 <juga> o/
16:59:41 <ahf> hello hello
16:59:52 <ahf> pad is at https://pad.riseup.net/p/tor-netteam-2022.1-keep
17:00:59 <ahf> giving folks a few to join in
17:01:04 <ahf> o/
17:01:10 <GeKo> o/
17:01:32 <ahf> how are folks doing with their respective boards on https://gitlab.torproject.org/groups/tpo/core/-/boards ?
17:01:59 <nickm> i have 2 huge items and am tryint to move ahead with them :)
17:02:22 <ahf> i am super stuck by grant crap, but i think i will recover once we are done with this week...
17:02:24 <Diziet> I have 4 tickets which are all somehow aspects of the same thing...
17:02:28 <ahf> nice
17:02:47 <dgoulet> good
17:03:37 <ahf> can't see anything fishy
17:03:47 <ahf> dgoulet: anything on releases?
17:03:48 <nickm> eta: Can you move what you're working to into "doing"?
17:04:07 <dgoulet> still working on 047 tickets that need to be in before stable but no ETA for next one
17:04:38 <eta> nickm: done
17:04:43 <nickm> ty!
17:04:45 * eta has arti!90 and not much else
17:04:46 <ahf> ok i do see something bad
17:04:47 <ahf> hm
17:04:49 <eta> uh, arti#90
17:04:55 * ahf moves tor#40395 out of doing
17:05:41 <ahf> ok, now i think is all OK
17:05:49 <ahf> dgoulet: perfect
17:06:17 <ahf> dgoulet: i'll let you do the wiki entry update for CoreTorRelease when you have an idea about stable estimate
17:06:26 <dgoulet> for sure
17:06:36 <ahf> https://gitlab.torproject.org/tpo/core/tor/-/issues?scope=all&utf8=%E2%9C%93&state=opened&label_name[]=Backport
17:06:43 <ahf> looks like our backport situation is alright
17:07:06 <ahf> maybe once we are out of the 0.4.7 stable period we can take a look at these and make the plan for them?
17:08:33 <ahf> ok, thanks to nickm there has actually been made progress to a tpo/core/team ticket this period!
17:08:50 <ahf> core/team#19 is in a much better place now
17:08:56 <nickm> ahhh
17:08:57 <ahf> but i think it still doesn't do all the stuff we wanted
17:09:05 <ahf> so lets keep it open for now
17:09:12 <nickm> sounds good
17:09:21 <ahf> otherwise i don't think any of us had time for any of these things
17:10:07 <ahf> i only see one new ticket from other teams which is something jnewsome found during shadow experiments with onion services with timeouts
17:10:12 <ahf> looks like there is activity on that one still
17:10:28 <ahf> we have no reminders, no announcements
17:10:57 <ahf> mikeperry: yo, so since we didn't do s61 A/V meeting today, should we keep the sync here ?
17:11:02 <mikeperry> yeah
17:11:10 <ahf> (remember all the s61 meeting is next monday at 15 utc in the usual bbb)
17:11:18 <ahf> mikeperry: excellent, you can take over now then
17:11:34 <mikeperry> yep, I moved the meeting since several s61 ppls are out today
17:12:11 <mikeperry> so last week, jnewsome added the ability to simulte mixed networks, with a percentage of Exit bandwidth running 0.4.6
17:12:33 <mikeperry> we ran a sim with 50% Exit bandwidth on 0.4.6, and 50% with congestion control enabled
17:13:06 <mikeperry> so negotiation was exercised, and we got to see how congestion control should behave at that level of exit upgrade
17:13:40 <mikeperry> the results are surprisingly promising.. the old 0.4.6 traffic did not choke out the congestion control circuits, which was one major concern with the upgrade
17:13:45 <mikeperry> so that is great news
17:13:53 <ahf> very nice
17:14:08 <mikeperry> I ran two more sims with 75% 0.4.6 and 90% 0.4.6 over the weekend. I am still downloading those results to look at them
17:14:26 <dgoulet> quite epic indeed
17:14:43 <ahf> that was your big concern after the patch was merged right? so this is really good news in general
17:15:26 <mikeperry> yeah. there was a chance that we would need to wait for a lot of Exits to upgrade before turning it on at all, or there might be no benefit (or worse, degredation)
17:15:36 <ahf> nice
17:16:06 <mikeperry> but at the 50% point, there is still an improvement in perf with congestion control, and no instances that I can see of circuits being choked out by the old alg
17:16:25 <ahf> sweet
17:17:30 <mikeperry> I need to record these results and update the sim plan, but this was a very good result. it should move up our timeline a bit wrt turning the thing on
17:18:20 <ahf> nice
17:18:24 <mikeperry> I also did a patch for sbws for the CIRC_BW fields: https://gitlab.torproject.org/tpo/core/tor/-/issues/40568
17:18:50 <mikeperry> it looks like juga was able to test that with our network-health exit
17:18:58 <juga> yes
17:19:32 <mikeperry> juga: those fields are the client's view of congestion control. so they will only update during upload activity
17:19:46 <juga> upload, not download?
17:20:15 <mikeperry> I had a minor brain fail juggling all of these things for the alpha and forgot to mention that in the spec patch
17:20:26 <juga> :)
17:21:07 <mikeperry> yeah.. if it is a huge problem for sbws to upload instead of download, then hrmm.. I will have to think about what to do
17:21:18 <juga> ok
17:21:36 <mikeperry> right now there's no easy way to see the congestion control state of an exit from the client (which is the download direction -- the exit is controlling it)
17:22:47 <mikeperry> there is also the question as to if upload is a good enough measurement for sbws.. in theory it *should* be exactly the same as a download.. but in practice maybe there's implementation quirks that make it different :/
17:23:38 <juga> if you think it's worth trying upload, we can try it
17:25:02 <mikeperry> the advantage of upload is also we can get hints if any of the bw auths do not have enough bandwidth to measure circs.. in that case, they will cause Exits to start sending XOFF
17:25:07 <mikeperry> which we can watch for in the logs
17:25:15 <mikeperry> so maybe upload is the way to go
17:25:29 <juga> ic, i can try it in a branch
17:25:29 <mikeperry> just makes me a little nervous that it is different than what we have always been doing, ofc
17:26:37 <mikeperry> maybe I should also add an XOFF field to STREAM_BW.. sbws listens to that yeah?
17:26:45 <mikeperry> anyway we should chat after this meeting
17:26:54 <juga> ok, let's chat later
17:27:52 <ahf> :-)
17:28:40 <mikeperry> dgoulet is still working on the overlod patch, I take it? there was the additional piece of making sure the descriptor updated if overload continued.. so we get fresh timestamps indicating fresh overload
17:28:52 <dgoulet> right, should be one very soon
17:29:34 <mikeperry> GeKo: anything for network-health?
17:29:43 <GeKo> just two small things
17:30:03 <mikeperry> we have one exit on 0.4.7 with CC enabled that we can test with.. do we know if torralf or others have updated to it yet?
17:30:18 <GeKo> i started creating a laundry list of things to check when we want to analyze some performance issues
17:30:24 <GeKo> e.g. due to ddos
17:30:43 <GeKo> i put that on the wiki
17:30:53 <GeKo> so we don't always start from scratch
17:30:56 <GeKo> https://gitlab.torproject.org/tpo/network-health/team/-/wikis/Tor-network-health-work#investigating-network-performance-issues
17:31:08 <GeKo> if things are missing, let's add them there
17:31:35 <GeKo> the other thing i that i organized our net health analysis work into a separate project
17:31:57 <GeKo> it collects all analysis issues related to netw ork health in a central place
17:32:06 <GeKo> https://gitlab.torproject.org/tpo/network-health/analysis
17:32:09 <mikeperry> ah, yeah. wrt ddos there were some analysis tickets we had to find relays that were consistently showing up in slow paths in onionperf
17:32:21 <GeKo> so in case you are wondering where issue X you filed might have landed
17:32:49 <GeKo> yeah, something like that would be useful
17:33:19 <GeKo> if there are tickets missing in that analysis project feel free to add them
17:33:25 <GeKo> that's all from me
17:33:47 <mikeperry> ok I can try to dig thru my history to find what I am talking about. I'll chat with you and juga after this
17:34:18 <GeKo> sounds good
17:34:23 <GeKo> i am afk for a bit, though
17:34:27 <GeKo> but sure
17:35:34 <ahf> anything else we need to dive into?
17:35:41 <mikeperry> I will likely also be somewhat distracted by grant things this week, with ahf. so most likely I'll just be doing followups on things, and organizing the sim plan this week. maaaybe I'll have time to start unit tests, since our params are looking solid
17:36:20 <mikeperry> that's it for s61 I think
17:36:30 <ahf> yeah, crossing fingers we will return to more normal days later this week
17:36:37 <ahf> mikeperry: awesome
17:36:46 <ahf> anybody who have anything else we need to chat about for this meeting?
17:37:02 * dgoulet is good
17:37:26 * GeKo too
17:37:51 <ahf> let's call it then
17:37:54 <ahf> thanks all for joining!
17:37:56 <ahf> #endmeeting