dr|z3d
1.9.0-3+ up on the /dev/ url, SSU2 speedups.
mildseven
is there anything i can do to stop the irc timeouts? :/
shiver
mildseven, did you try rising the time for timeouts with /set net_ping_timeout
shiver
its in seconds, i have it at 300
shiver
otherwise i dont think you can
shiver
mildseven, did you see what i wrote?
mildseven
shiver: no sorry i disconnected
shiver
mildseven, did you try rising the time for timeouts with /set net_ping_timeout
shiver
its in seconds, i have it at 300
shiver
how many tunnels are set for irc at /configtunnels
shiver
mine is at 3
mildseven
2 tunnels
shiver
don't think you can do more than that, thats just the way the network is
shiver
dr|z3d, log still looks like the one after the first dev update
shiver
which means better, but still 1.5sec spikes over 5min
dr|z3d
shiver: ok, will dig deeper.
dr|z3d
shiver: on /info is your system classed as slow?
dr|z3d
ok, I've made some mods to the Iterative Search Job to lessen the load all round, and more so for systems with <= 4 cores. let's see if this helps. update is uploaded to /dev/
shiver
dr|z3d, sorry, had to go afk
dr|z3d
np
shiver
not classed as slow
dr|z3d
ok, thought not, had to check.
shiver
i often see <1ms joblag now, so something is working.
dr|z3d
good, that's what you should be seeing after the router's settled down after a restart when you're a floodfill.
shiver
but i dont think that the cpu is overloading when the joblag spikes, i have munin running and see no cpu spikes
dr|z3d
hopefully the latest build will help keep that job lag low and smooth out the spikes.
dr|z3d
no, it's not a cpu load thing, it's that jobs are taking a while to finish.
dr|z3d
under 200ms and the router shouldn't do anything to affect the tunnel count, but the sidebar indicates an average, not a realtime readout.
shiver
i saw that zzz said around 0.7% of the net have ssu2 enabled, maybe after the next update more will see the spikes..or my system is something special
dr|z3d
haha, something "special" for sure. special needs :)
dr|z3d
ssu2 shouldn't be causing spikes in any event, as far as I can tell it's the way I2P+ looks for leases and router infos.
dr|z3d
shiver: if you now look on /jobs, you should see less instances of the Timeout Iterative Search job running concurrently.
shiver
i saw 4 at once
dr|z3d
should be fine
shiver
did you change "Lookup LeaseSet & Reply to Client" too?
shiver
asking because that was in the thread list last time when is saw spikes
dr|z3d
not in this update, but a couple of updates previously, yes, timeout's a lot lower now.
shiver
lag sits around 3-10ms
dr|z3d
that's pretty good. it'll go lower after the first 1/2 hour of uptime.
dr|z3d
essentially in the first 1/2 hour your router will try a lot harder to validate existing routers in your netdb to ensure you have a good set. then it calms down.
shiver
unrelated question, when i think version 1.6 still was a thing i had i2p running on win10 and had much more active peers, gui said around 4000/9000. this was with normal i2p but when i switched to linux i could not reproduce those numbers. checked everything, had no firewall or other connection issues. still scratch my head
shiver
what could that be?
dr|z3d
different mechanism for calculating the numbers.
dr|z3d
active peers is now over 1 minute / 1 hour
shiver
no, i mean with the same version
shiver
just switching form win to linux
dr|z3d
oh, i2p not i2p+ ?
shiver
normal i2p
dr|z3d
and the win10 router was up and running for a long time, the linux one just installed?
shiver
i had to switch i2p to a different machine because i got audio errors from the router running lol
shiver
no, both were running for a long time
dr|z3d
open file descriptor limits on linux perhaps?
dr|z3d
what does ulimit -n tell you in the account you're running i2p from? 1024?
shiver
iirc that i looked up and it wasn't the issue
dr|z3d
and now.. ulimit -n ?
shiver
1024
dr|z3d
oh dear. not great.
dr|z3d
try adding the following to /etc/security/limits.conf .. system will need to be rebooted for them to take effect:
dr|z3d
* soft nproc 65535
dr|z3d
* hard nproc 65535
dr|z3d
* soft nofile 65535
dr|z3d
* hard nofile 65535
dr|z3d
root soft nproc 65535
dr|z3d
root hard nproc 65535
dr|z3d
root soft nofile 65535
dr|z3d
root hard nofile 65535
dr|z3d
you can set the limit temporarily with ulimit -n 65535 but it won't persist a restart.
dr|z3d
that config about will set max processes and max open files to 65535 globally.
dr|z3d
*above
dr|z3d
tweak to taste :)
shiver
done, does the have an impact in i2p?
shiver
*that
shiver
i'll restart brb
dr|z3d
shiver: potentially, yes, increasing max open files can help i2p.
shiver
i restarted but the ulimit is still 1024
dr|z3d
and you made those mods to /etc/security/limits.conf ?
dr|z3d
for now you can just do: -> crontab -e * * * * * ulimit -n 65535 <- for the i2p user account.
dr|z3d
but limits.conf should work unless whatever linux you're on is preventing that.
shiver
ubuntu with xfce
dr|z3d
hmm, odd.
shiver
will look at
shiver
thank you
dr|z3d
np
shiver
its not commented out and is see no log error
shiver
the other somment says changing the default in /etc/systemd/system.conf and /etc/systemd/user.conf too, should i?
dr|z3d
and you went with the values I pasted?
shiver
*comment
dr|z3d
no, leave that, you can set the globals from limit.conf
shiver
look at rrx4hxejr4pod3tqrkj5ckzuauwy2aqsfvpiac2tkebp2iuv5qla.b32.i2p/view/HXbo9pm7SF_Why9xbXt4m7XlnhjguuBUk70NhctIu_5IOPhnsyst/HXbo9pm7SF.txt
shiver
this is how i wrote it, after the #end of file
dr|z3d
should be ok
shiver
i set it with ulimit -n for now
dr|z3d
in your i2p account, add the following 2 lines to crontab -e
dr|z3d
@reboot ulimit -n 65535
dr|z3d
* * * * * ulimit -n 65535
dr|z3d
that will do the same thing.
Irc2PGuest40951
retroflex.
dr|z3d
you can do the same for the root acct if you want.
shiver
okay
shiver
afk for a while
dr|z3d
ok, when you get back let me know how your job lag is looking.
shiver
dr|z3d, peak of 600ms over 10min
shiver
then a short peak of 300 and 150ms
shiver
while it peaks i still see Lookup LeaseSet & Reply to Client running for 3s
shiver
often on all 8 threads
dr|z3d
shiver: ok, thanks for the feedback.
dr|z3d
how do your tunnels fare while job lag is spiking?
dr|z3d
I'm looking at obscuratus' latest patch here: git.idk.i2p/i2p-hackers/i2p.i2p/-/issues/364
dr|z3d
zzz is also testing something which hopefully will turn up tomorrow.
shiver
dr|z3d, at the 600ms spike tunnels fell from 900 to 500
dr|z3d
shiver, ok, and how long are those spikes lasting?
shiver
peak of 600ms over 10min
shiver
smaller one maybe 5min
dr|z3d
ok, that's not great. I'm not seeing anything like that behavior here on various systems.. do you know how much ram you've got allocated to the jvm? .. that's also something to look at.
shiver
2G
dr|z3d
ok, plenty then.
shiver
i set it so high to never have to care again
dr|z3d
you could try disabling floodfill and see if that makes the spikes go away.. floodfill definitely asks more of the system than non-floodfill mode.
dr|z3d
regarding upstream patches, if zzz's pushing something out tomorrow or thereabouts, I'm inclined to see how that changes things before making any more changes here. might fix things.
shiver
when i look at the logs i never see the cpu cores at max clock and with 1.8 everything was fine.
shiver
oh, i remember something unrelated but kind of not
shiver
one moment
dr|z3d
yeah, it's looking more like destination lookups.. with obscuratus' patch, most new lookups won't fail first time anymore, though it may have introduced issues which zzz is proposing to fix a different way.