@eyedeekay
&zzz
+R4SAS
+T3s|4
+dr|z3d
+eche|off
+nilbog
+orignal
+postman
+qend-irc2p
+sourceress
Arch
Birdy
Irc2PGuest30010
Irc2PGuest36077
Irc2PGuest49364
Irc2PGuest51117
Irc2PGuest6564
Irc2PGuest65656
Irc2PGuest67278
Irc2PGuest74235
Irc2PGuest83482
MatrixBot
Onn4l7h
Over
Sleepy
T3s|4_
Teeed
Yotsu
_bob_
aargh3
ac9f
acetone_
ahiru
anontor
b3t4f4c3__
cims
dr4wd3_
duanin2
f00b4r
hababam_
hagen_
leopold
makoto
marek
marek22k
n2_
noidea
not_bob_afk
nyaa2pguy
o3d3_
poriori
profetikla
r00tobo
rapidash
solidx66
stormycloud[m]
test7363673
uop23ip
urist_
user_
w8rabbit
zelgomer
dr|z3d
zzz: any scope for shunting some of the eventpumper load onto NTCP Finisher threads? they seem underutilized right now.
dr|z3d
NTCP reader threads are being exercised more, writer threads, not so much, and then finisher threads least of all.
zzz
dr|z3d, we just did a big shunting with the ntcp write changes
zzz
there's no need to balance workload across threads
zzz
and jogger would tell you the opposite
dr|z3d
ok, just noticing the finisher threads are less noticeably less utilized than both the reader and writer threads.
dr|z3d
*noticeably less utilized
dr|z3d
if I constrain the pumper thread it's happy to use around 50% of one core under loads of ~1MB/s
zzz
probably a candidate for removal of the finisher threads
zzz
jogger hates all of it
dr|z3d
yeah, jogger's not a huge fan of the "let's throw threads at it approach", understandably, given android.
dr|z3d
android/arm
dr|z3d
more arm than android, but yeah.
zzz
not sure if arm is worse at threads than x86, or if it's just worse at everything
dr|z3d
probably the latter for the most part, given the specs of the usual candidates vs x86 (raspberry pis)
dr|z3d
still, if the finisher threads aren't serving much purpose, sounds like a reasonable idea to eliminate them.
zzz
just takes time to analyze and test, and of course be extra careful not to introduce deadlocks