17:00:15 <ahf> #startmeeting Network team meeting, 20th september 2021
17:00:15 <MeetBot> Meeting started Mon Sep 20 17:00:15 2021 UTC.  The chair is ahf. Information about MeetBot at http://wiki.debian.org/MeetBot.
17:00:15 <MeetBot> Useful Commands: #action #agreed #help #info #idea #link #topic.
17:00:18 <ahf> hello hello
17:00:22 <GeKo> o/
17:00:25 <ahf> our pad is at https://pad.riseup.net/p/tor-netteam-2021.1-keep
17:00:36 <nickm> hi!
17:01:34 <jnewsome> O/
17:01:44 <ahf> (   ) /
17:01:45 <mikeperry> o/
17:01:47 <dgoulet> hi
17:01:51 <ahf> ok!
17:02:00 <hiro> o/
17:02:01 <ahf> https://gitlab.torproject.org/groups/tpo/core/-/boards
17:02:06 <ahf> how's everybodies' boards looking?
17:02:24 <nickm> ahf: pretty good, I think
17:02:40 <ahf> ya
17:03:33 <ahf> i don't see anything that needs attention
17:04:09 <ahf> we did the release last week with a minor bump on the way where we hadn't gotten the dirauth's to bump their recommended versions, but it was solved and we released 2 days late
17:04:38 <ahf> i don't see anything that affects our 0.4.5 and 0.4.6 series in the issue tracker right now
17:04:46 <ahf> we have some potential backporting to do once 0.4.7 have baked a bit
17:06:35 <ahf> i take that as a silent nod
17:06:48 <ahf> we have two discussion points this week:
17:07:00 <ahf> oh, gaba is afk this week and i am afk on wednesday as an announcement
17:07:06 <ahf> 2021-09-20 [nickm] Plans for this week's arti meeting?
17:07:14 <ahf> nickm: you wanna go here
17:07:17 <ahf> ?
17:07:54 <nickm> sure
17:08:17 <nickm> I'd like a little advance thinking about what to talk about this week.  Just updates and next steps , or any fancier topics we should cover?
17:09:12 <ahf> i don't remember how these were created, but the end goal with the meetings is that it's a space where we can also coordinate with other developers who uses arti, right? but we don't have any of those yet of course
17:09:55 <nickm> yeah
17:10:13 <nickm> also it's a goal to make them interesting so as to encourage interested devs to attend
17:10:52 <ahf> so it should be less of a status meeting (to avoid duplicated content between the weekly status emails) ? and more about arti related topics
17:11:50 <nickm> I think that could be a good idea; though I don't want to make permanent changes without gaba
17:12:03 <ahf> *nod*
17:12:15 <ahf> what topics would be interesting for this week and have value?
17:13:18 <nickm> Hm. I can talk about nearly anything, but my working focus is on guards.  I guess I could give an overview of some piece of code or I don't know what
17:13:31 <nickm> Anyone else have any ideas here?
17:14:59 <ahf> hm, nobody?
17:15:49 <ahf> the last meeting had 2 externals in them iirc. are they something we need to consider if they make sense to do? i am away on wednesday, so i cannot participate so i am not saying what the topic should/could be here
17:16:38 <nickm> ok.  I'll try to think of something
17:16:46 <ahf> ok
17:16:47 <nickm> maybe we could each pick a little example program to work on or something
17:17:50 <ahf> please check how many people who joins in and then we can maybe follow up on the convo next week about these meetings in general
17:17:53 <ahf> ok!
17:17:54 <nickm> Makes sense
17:17:59 <ahf> 2021-09-20 [nickm] Is our 'status update' process below working for us?
17:18:10 <nickm> I think my next question is moot: most people have in fact done status updates now :)
17:18:13 <ahf> i go through them each time, i don't feel like i learn anything i didn't know already from them
17:18:14 <ahf> but
17:18:27 <ahf> i also don't tell people if they don't update unelss it has been REALLY long time since an update
17:18:29 <nickm> When I skimmed it before the majority was out-of-date
17:18:34 <ahf> ah!
17:18:41 <ahf> i have an alarm for 15 min. to meeting time to write it
17:19:11 <dgoulet> I personally don't use the updates at all. With Gitlab and IRC and the fact that I talk to each of you regurlarly, I end up not needing it
17:19:20 <ahf> i think they are less needed in the current network team format than when we were in different timezones and didn't have the thursday meeting
17:19:23 <dgoulet> but I assume people not that close to our dev would think it is useful?
17:19:26 <dgoulet> would use it*
17:19:27 <ahf> and people are really good at also having 1:1 convo's in #tor-dev now
17:19:34 <nickm> ok, makes sense
17:19:42 <ahf> dgoulet: i don't think they are read outside this meeting context but i might be wrong
17:20:15 <ahf> ok added a small line halfway through then we can see if anybody reads it outside the meeting. maybe
17:20:22 <ahf> i did that with an email at some point and got zero responses
17:20:24 <nickm> :)
17:20:40 <ahf> ok!
17:20:45 <ahf> i think it's s61 time with mikeperry
17:20:52 <mikeperry> ok
17:20:53 <ahf> i don't see other topics or announcements today
17:21:48 <mikeperry> so dgoulet and I have been doing the inital tracing and tuning of flow control over onions
17:22:20 <mikeperry> I believe the branch is at the point where it can be run in shadow, but there is some things we can still learn from onion testing and tracing
17:23:00 <mikeperry> I updated the experiments section in my branch of the spec to describe this proces a bit better, and document more parameter values for tuning, etc: https://gitlab.torproject.org/mikeperry/torspec/-/blob/flow_control_updates_v2/proposals/324-rtt-congestion-control.txt#L914
17:23:37 <mikeperry> I had a look at nick's negotiation branch, but I need to dig deeper still
17:24:04 <mikeperry> but I also want to get things prepped for shadow, and finish off the rest of the checklist items for flow control asap: https://gitlab.torproject.org/tpo/core/tor/-/issues/40450
17:24:50 <mikeperry> jnewsome: how goes the shadow scaling?
17:25:45 <jnewsome> Running out of ram
17:25:50 <mikeperry> hiro: also, one thing I notced from your graphs of shadow vs live is that there were far less datapoints coming out of shadow
17:26:04 <jnewsome> Started doing some ram profiling on Friday
17:26:16 <mikeperry> hiro: that's why jnewsome pointed out there's more perf clients than just one: https://gitlab.torproject.org/tpo/network-health/metrics/website/-/issues/40011#note_2751267
17:27:34 <ahf> it's not something where we can throw more RAM at it?
17:27:42 <jnewsome> Starting to wonder if maybe we should plow ahead without 100% client scaling in the meantime
17:27:44 <ahf> i suspect RAM is cheaper than dev time still. or is it a leak?
17:27:54 <jnewsome> We already have 1.5 tb
17:27:58 <ahf> uf!
17:28:02 <mikeperry> jnewsome: so after trying that stack/heap copy avoidance for clients, we could try some value of the client scaling that still uses guards, but with a smaller client multiplication factor
17:28:29 <mikeperry> jnewsome: yeah, possible
17:28:50 <hiro> ok so mikeperry I will aggregate over all the shadow clients and see what I take out of it
17:29:04 <jnewsome> The stack/heap setting I was talking about doesn't save much ram, just makes the accounting simpler for profiling
17:29:10 <mikeperry> ah
17:29:44 <jnewsome> Hiro it'd just be the perf clients right? I'm not saving the logs for the BG traffic clients
17:30:13 <hiro> I think it is just the perf clients yes
17:31:05 <mikeperry> yeah, we just want the perf clients. we may also want some level of exit logs, but I don't see us needing BG clients
17:31:20 <jnewsome> ah, finally have matrix up on my laptop instead of just my phone. fighting some hardware issues :P
17:31:48 <ahf> :-D
17:31:51 <jnewsome> i'm saving a few BG clients so we at least have a sample of their logs, but yeah  we run out of storage trying to save them all
17:33:10 <jnewsome> ok, so i'll see if can get a 10% network size going, still using < 100% client scaling for now. probably 10%
17:33:17 <jnewsome> and continue memory profiling in the meantime
17:34:05 <jnewsome> from prelim results it looksl ike there might be a small leak in shadow, but most of it is just having thousands of tor clients in the steady-state.
17:35:46 <mikeperry> so long as relay utilization looks similar between shadow and live, and perf characteristics match, that should be enough
17:36:28 <mikeperry> guards are important still ofc, because we do want those clients to avoid overloaded relays through CBT, but beyond that, it is likely OK if there is a multiplier on their activity
17:36:44 <jnewsome> ok, great
17:39:05 <ahf> anything else? :-)
17:39:12 <mikeperry> GeKo: anything from you?
17:39:37 <GeKo> not much
17:39:45 <GeKo> i was mostly in okr mode last week
17:39:56 <GeKo> and then needed to help tor browser folks
17:39:56 <mikeperry> ok
17:40:10 <GeKo> i which will continue this week (the browser folks stuff)
17:40:25 <GeKo> but i looked over the overload parts for the support portal
17:40:36 <GeKo> and an upcoming mail to tor-relays@
17:40:47 <hiro> yeah I have that ready to go as soon as it is merged
17:40:48 <GeKo> and created a key for our network-report@ help alias
17:41:09 <GeKo> so, we are finally in shape, i think, for flipping the switch
17:41:21 <GeKo> to show some overload on relay search
17:41:32 <GeKo> progress :)
17:42:05 <GeKo> i hope i can get back this week at least to start analyzing the overload i am now collecting for a while
17:42:13 <GeKo> to start notifying relay operators
17:42:16 <GeKo> we'll see
17:42:28 <GeKo> that's all from me
17:42:42 <mikeperry> nice, ok
17:43:25 <GeKo> faravahar moved to sbws now, too
17:43:36 <GeKo> we need to keep an eye on that as well
17:43:55 <GeKo> only gabelmoo and moria are left on torflow now
17:44:23 <ahf> cool. very cool
17:44:31 <GeKo> yeah
17:45:03 <ahf> are we otherwise done? anything else?
17:45:15 <mikeperry> I am good
17:45:35 <ahf> excellent
17:45:45 <ahf> thanks all
17:45:47 <ahf> #endmeeting