IRCaBot 2.1.0
GPLv3 © acetone, 2021-2022
#ls2
/2022/03/11
@eyedeekay
+R4SAS
+RN
+RN_
+Xeha
+acetone
+hk
+not_bob_afk
+orignal
Irc2PGuest59134
Irc2PGuest66424
Irc2PGuest81747
Leopold
Nausicaa
Onn4l7h
Onn4|7h
T3s|4_
T3s|4__
anon3
cancername
eyedeekay_bnc
l337s
qend-irc2p
shiver_
u5657_
x74a6
RN does SAM get no indication of the requesting app?
RN just trying to understand
RN I would think once per app, but if SAM is blind then 1 sounds most logical to me
zzz yes RN, there's nothing in the protocol to identify it other than a username when auth is enabled
zzz eyedeekay, near the end of the meeting I said you probably want to do it later than the HELLO; I think that was bad advice
zzz agreed that at the HELLO (#1) is by far the cleanest and easiest
zzz zlatinb, good morning, ping re: SSU testnet
zlatinb zzz: ping
zlatinb what's up
zzz as I said, 45 KBps, 0-hop, 25 msg delay, no drop
zzz nothing obvious, but shitloads of dup tx
zzz went back to SSU 1, same thing
zzz went back to 1.7.0 release, same thing
zzz you organized enough to find what your setup and results were for the last time you did SSU-only testing?
zlatinb hmm that doesn't sound right
zzz haven't looked too hard as to root cause, but it's stuck in fast retx mode
zlatinb I'll have to look on gitlab, I usually publish results there, but of the top of my head 0 hop without delay was 100+ Mbits
zlatinb 25ms delay was definitely more than 45kb
zzz not sure if it's a quirk of my test setup, which I'll explain here:
zzz 16 routers, same delay on all. Half floodfill, a couple firewalled
zzz all standard NTCP+SSU 1, except for two, which are my HTTP client and server: SSU 1/2 only
zzz the client and server are both 0-hop
zzz I'm getting about 10-15% retx at the SSU layer
zlatinb well lets try the simplest thing first:
zlatinb for i in $(seq 1 15); do lxc-attach i2p$i -- /sbin/tc qdisc del dev eth0 root; done
zzz for a 2 MB download, about 10,000 pkts, about 1000-1500 retx
zlatinb then see what you get, should max out the CPU and be in the 10+MB/s
zzz gotta fire it up for the day, no qdisc by default anyway
zzz no change with 0 delay
zlatinb then something isn't right
zlatinb time to enable debug logging ... :-/
zzz if pkts are dropping somewhere, I haven't found it yet
zlatinb last time I ran the benchmarks was before the changes to pass priority and switch to a priority codel queue
zlatinb is the rtt calculated correctly on the web console?
zzz yes
zzz I'm not seeing codel drops in any queue
zzz let me change from 0-hop to 3-hop which will put me closer to your setup
zlatinb it's harder to dig through logs of 3-hop tunnels
zlatinb with 0 hop you know exactly where to look
zzz just for a quick test
zzz 30 KBps
zlatinb still too low
zlatinb but not as drastically
zlatinb also keep an eye on cpu usage when doing the 0-hop test, it does max out my 8-core xeon
zlatinb zzz: also make sure you didn't by accident execute qdisc on your host laptop :)
zlatinb /sbin/tc qdisc ls
zzz no cpu problems at 45 KBps, verified qdisc not on host
zlatinb something else is happening because when I was doing the interop testing I was able to download at expected speeds between java and i2pd
zzz that was a udp-only test?
zlatinb the java node was udp-only, i2pd doesn't support such mode
zlatinb I could try SSU2 if you show me how to enable/force it
zzz no, I want to fix SSU 1 first
zlatinb did you ever change the bandwidth limits in the testnet?
zlatinb cause 45 sounds alwfully close to the very slow default
zzz oooh thats it
zzz 61 KBps out
zzz thanks, will reconfigure, retest, and report
zzz 0 hop: 1.25 MBps w/ 25 ms delay; 9.25 MBps w/ no delay
zlatinb that's more better
zzz 300 KBps w/ 25ms delay, 1% drop
zzz zero dup rx/tx at the SSU2 layer when no qdisc dropping
zlatinb 1 or 0.1? 1% is high
zzz it's zero-hop
zzz zlatinb or eyedeekay would you please bump that reddit thread about i2pd < 2.41 crashing, I think I'm about ready to throw this back on the live network
R4SAS zzz: just throw
eyedeekay I'll bump it anyway
zzz orignal, the problem was it trying to decode the "i" value in SSU? if we change it to "k" would that prevent the crashes?