[erlang-questions] performance of websocket

Eric Moritz eric@REDACTED
Wed Jun 13 19:53:47 CEST 2012


I made a dumb mistake at 3am on this table.  Those ms times are actually
seconds.  I've updated the table at this link.  Disregard the table in this
email.

https://github.com/ericmoritz/wsdemo/blob/master/results.md

If you look at the raw data you will also notice that the handshake times
for 2,500 of the 10,000 connections are greater than 742ms which is pretty
lousy as well. While none of the TCP connections timed out, 25% of the
connections had to wait at least 743ms to start sending messages.

Once the handshake eventually finished, the per-message latency was really
good compared to the others with a mean of 17ms.

Eric.

On Wed, Jun 13, 2012 at 12:49 PM, envelopes envelopes
<sunwood360@REDACTED>wrote:

> from:  https://github.com/ericmoritz/wsdemo/blob/master/results.md
>
> It seems Erlang's connection time is a little bit high.
> Summary ImplementationConnection Time (mean) Latency (mean)Messages
> Timeouts Erlang00.865ms00.017ms 28492940 Java (Webbit)00.567ms 00.835ms
> 1028390157 Go00.284ms 18.503ms2398180225 Node.js00.768ms 42.580ms1170847
> 4299 Python (ws4py)01.561ms 34.889ms10529965208
>
> Both the Python gevent/ws4py implementation and the Node.js websocket
> implementation failed hard with around half of the connections hitting the
> 2 second TCP connection timeout.
>
> I expected Go to kick Erlang's ass in the performance department but the
> message latency was much higher than Erlang's latency and we had 225
> unhappy customers. Go only reached C9.775k; close but no cigar.
>
> I did not know what to expect with the Java Webbit implementation. I
> expected it to perform somewhere close to the Go implementation. The Webbit
> implementation did do much better than the Go implementation but it still
> had 157 connection timeouts which is unacceptable.
>
> I was very surprised that node.js fell down after 5001 connections and
> that gevent fell down at 4792 connections. Both platforms were specifically
> built to for the C10k problem and both platforms could barely handle C5k.
>
> _______________________________________________
> erlang-questions mailing list
> erlang-questions@REDACTED
> http://erlang.org/mailman/listinfo/erlang-questions
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://erlang.org/pipermail/erlang-questions/attachments/20120613/7ee481ce/attachment.htm>


More information about the erlang-questions mailing list