~dr|z3d
@RN
@RN_
@StormyCloud
@T3s|4
@T3s|4_
@eyedeekay
@not_bob_afk
@postman
@zzz
%Liorar
+FreefallHeavens_
+Leopold
+Xeha
+acetone
+cancername
+cumlord
+hk
+mareki2p
+profetikla
+r00tobo_BNC
+uop23ip
+weko
An0nm0n
Arch
BravoOreo
Danny
FreeRider
Irc2PGuest83914
KiloDeltaRomeo
Meow
Nausicaa
Onn4l7h
Onn4|7h
anon1
anu3
boonst
enoxa
mareki2pb
opaque
orignal
poriori_
qend-irc2p
shiver_
simprelay
solidx66
tr
u5657_1
woodwose
orignal
so, what do you think about second attempt to the same FF?
zzz
if you have no result data how do you know it's faster?
orignal
visaully
orignal
10 secs vs. like a minute
zzz
FYI I don't have a stat for it but eyeballing it, it looks like the avg response time is ~300-400 on a decently fast box
orignal
why? what if request or reponse was dropped
zzz
sure, it could happen, I don't know how likely though
zzz
you could try it and collect data
orignal
onon says often
dr|z3d
you got me interested in that stat, zzz. Having made some tweaks, looks like I'm down to around 20% failure, will know more soon.
zzz
but you're likely to get the answer from the next peer anyway, so I don't know what it buys you
orignal
second attempt might also fail
zzz
jrandom implemented parallel lookups but not sure if it was ever enabled, it isn't now
orignal
you can flood the network this way
zzz
yeah it's a tradeoff, there's no right answer
dr|z3d
Lifetime average: 0.227 (444 events)
dr|z3d
we do parallel lookups.
dr|z3d
private static final int MAX_CONCURRENT = SystemVersion.isSlow() ? 2 : 4;
zzz
the stat will drive a lot lower if you're high-bw, you're more likely to have direct conns to the ffs
zzz
hard to do apples/apples
dr|z3d
I was seeing around 40% pre-tweak, so it looks like there's an improvement, for some definition of improvement.
zzz
orignal, I do recommend that whatever timeout you have for LS lookups you do it for RI also, I don't know why it would be different
zzz
shortening it will make tunnel builds and OBEP->IBGW data messages faster
orignal
because LS lookup is through a tunnel
orignal
while RI lookup is usually direct
zzz
ok
orignal
there 1200 min + up to 400 more variace
orignal
goint to reduce depending if direct or through exploatory
dr|z3d
private static final long SINGLE_SEARCH_TIME = 5*1000;
orignal
too long
orignal
no reason
dr|z3d
private static final long MIN_SINGLE_SEARCH_TIME = 1000;
dr|z3d
unlikely to hit that limit, but a single search could query up to 32 peers.
zzz
I'm beyond trying to bring you back to sanity drz, but the flood redundancy is only 3, so querying 32 peers is bananas
dr|z3d
*** grins. ***
dr|z3d
private static final int MAX_NON_FF = SystemVersion.isSlow() ? 6 : 8; ?
dr|z3d
let's drop that back down to 16 then.
dr|z3d
it was 24 previously.
dr|z3d
(or 16 if slow)
zzz
and it's not unlikely if you're looking for a LS that's not up
dr|z3d
true. I'm working on the maybe incorrect assumption that no all ffs are reliable.
dr|z3d
I'd also like to see the lookup fail average at around or under 20%.
dr|z3d
it's not far off right now.
zzz
sure. I don't know how to do the math, but the imputed transient fail rate that would tell you the 32nd lookup (well beyond the normal keyspace window) would be successful is staggeringly high
dr|z3d
see above, dropped the max to 16 :)
dr|z3d
60 sec rate: Average: 0.166 • Highest average: 0.309 • Average event count: 251.698
dr|z3d
Events in peak period: 237 • Events in this period (ended 88 sec ago): 391
orignal
zzz, what's you aveage number of lookup attempt?
zzz
1.877 after the change, I don't have data for before the change
orignal
so, I can start re-requesting floodfill after second attaempt
orignal
e.g. tried first than second
orignal
and request third and first again
zzz
you can do whatever you want, I can't predict what's going to be useful or make tradeoffs for you
orignal
just thinking if it's usually after second attempt
dr|z3d
definite improvement on lookup failure: cake.i2p/file/kVbqd36rlT_EaPUNreafrCcsrjFD4CAdpVOhY5RH8_ECDcsMSVFg/peer.failedLookupRate.svg
dr|z3d
Lifetime average: 0.192 (248,752 events)
dr|z3d
+ sidebar new hosts refinements, more robust implementation: cake.i2p/file/M2Md77S18d_2axVQS2qX36mloXd7MqDq4lOCxg99m_j5GuFHdn9m/newhosts.webp
dr|z3d
afaik, not_bob_afk, kytv still runs plugins.i2p