dr|z3d
*** i2pd 0.9.57 now shipping ***
orignal
yes, I commited it
orignal
waiting for R4SAS
obscuratus
I'm running into some unusual issues on my testing network with the latest I2PD and Java I2P.
obscuratus
It's taking 50% more bandwidth to complete a torrent test run across the network.
obscuratus
And routers are encountering unusual spikes in throughput aren't normally there.
obscuratus
I'm in the process of reverting the I2PD and I2P versions to 2.44.0 and 2.0.0 respectively to confirm this wasn't something there all along that I missed.
obscuratus
On my testing network, my logs fill up with messages like: InboundEndpointProcessor: Invalid IV, dropping at IBEP
obscuratus
I don't see this near as much on the live network.
orignal
interesting
zzz
hmm
zzz
that's the bloom filter
obscuratus
That's what I was beginning to suspect.
obscuratus
I see lots of duplicate IVs
zzz
I assume your torrent clients are only on java i2p routers?
obscuratus
zzz: Yes.
zzz
so i2pd is just being routed through
obscuratus
My testing network has 24 Java I2P routers, and 40 I2PD routers.
zzz
did you try reverting or not?
obscuratus
Yes, this appears to be something that's been there, but I missed it because it's intermittent.
zzz
is it always at IBEP?
obscuratus
zzz: No. If I expand my logging I also see OutboundTunnelEndpoint messages and HopThrottle messages.
zzz
I'm asking specifically about IV messages
zzz
HopProcessor.java: _log.warn("Invalid IV, dropping at hop " + _config);
zzz
InboundEndpointProcessor.java: _log.warn("Invalid IV, dropping at IBEP " + _config);
zzz
OutboundTunnelEndpoint.java: _log.warn("Invalid IV, dropping at OBEP " + _config);
zzz
hop throttle should be disabled with allowLocal
obscuratus
zzz: Lol, I re-enabled that since we were making changes there.
zzz
so for "invalid IV", is it only at IBEP or all three places?
obscuratus
zzz: All three places.
zzz
ok what kind of bandwidth usage are the routers seeing, and what's the wrapper max memory configured to? your bloom filter may just be too small
obscuratus
Each I2PSnark instance has the upload bandwidth set at 36 for this particular test. My wrapper is configured with wrapper.java.maxmemory=256
zzz
what's the router bw limit?
obscuratus
It varies by router. I try to spread them out approximately similar to the live network.
zzz
ok
zzz
have you injected delay and/or drop with tc qdisc?
obscuratus
Should I be seeing that warning in the bloom filter if that's not balanced?
obscuratus
zzz: 35 ms delay with tc qdisc
zzz
no drop?
obscuratus
No drop
zzz
are the routers SSU2-only or both transports?
obscuratus
The I2PD routers are now SSu2-only. I think for Java I2P, that doesn't matter, right?
zzz
I mean, is NTCP2 enabled or disabled?
obscuratus
NTCP2 is enabled, and I try to encourage about 50% NTCP tunnels on the Java I2P routers.
zzz
and enabled for i2pd as well,right?
obscuratus
zzz: Yes. but it does seem like SSU2 is preferred now.
zzz
ok
zzz
I have a theory, which I can reveal now or not, but kill all the i2pd routers and see if all the Invalid IV messages stop
dr|z3d
SSU2 is preferred on a random basis in i2pd. on java i2p, ntcp is still the preferred transport.
obscuratus
zzz: OK, I'll repeat my run without the I2PD routers.
RN
anything I should be watching for on -14rc?
obscuratus
zzz: Your premise seems correct. If I just use the Java I2P routers (I2PD disabled), the Invalid IV messages disappear, and bandwidth resembles what is expected for this test.
zzz
RN, nothing special
zzz
thanks obscuratus
zzz
my theory, orignal, is that i2pd doesn't have a SSU2 Bloom filter, or it's broken
RN
:D
dr|z3d
i2pd has a bloom filter implementation.
dr|z3d
(or did)
dr|z3d
appears it's been binned since I last refreshed the repo.
dr|z3d
libi2pd/BloomFilter.cpp: /** @brief decaying bloom filter implementation */
dr|z3d
not in the current tree: github.com/PurpleI2P/i2pd/blob/openssl/BloomFilter.cpp
obscuratus
The i2pd BloomFilter implementation looks like it was removed here:
dr|z3d
there's no looks like, that's 100% certainty. blame orignal.
dr|z3d
I don't know enough about bloom filters to know if that's potentially problematic for network health.
dr|z3d
no doubt zzz has a view :)
obscuratus
Diplomatic and subtle as always. :)
dr|z3d
subtlety often gets lost in translation, not always the best policy :)
dr|z3d
orignal you hooligan, what you done to the bloom filters?