[erlang-questions] Throughput on Yaws 1.68 / OTP R11B-3 / Suse Linux 10.2?

Francesco Cesarini francesco@REDACTED
Mon Mar 10 12:47:02 CET 2008

Just an update for the archives in case others come across this problem 
in the future. We solved it by writing our own http client, after not 
having seen any performance degradation with tsung.

The bottle neck was not in the inets driver, but in both ibrowse and the 
inets http client, both of which did not preform that well when scaling 
up to 1000s of simultaneously open socket connections on the same node 
(It was misleading as the http client was running at 2 - 3 % CPU while 
ibrowse was at 100%). Chandru has forwarded us an updated version of 
ibrowse which we will test soon.


Francesco Cesarini wrote:
> No, not surprisingly, the bottle neck is not Yaws. We got a slight 
> improvement, which is understandable, but the system still stalls at a 
> ridiculously low  rate and a degradation of service the more we increase 
> the load. It is still a few thousand hits per second, so most people 
> would probably not have noticed unless they've run proper stress tests. 
> I guess we will have to start digging in the inets driver.
> Francesco
> --
> http://www.erlang-consulting.com
> Claes Wikstrom wrote:
>> Torbjorn Tornkvist wrote:
>>> You could perhaps try the iserve-server provided by Sean Hinde:
>>>  http://www.tornkvist.org/gitweb?p=iserve.git;a=summary
>>> This way you get the minimal turnaround in Erlang, just to
>>> get an idea if the bottleneck is in Yaws or somewhere else.
>> If you find that the bottleneck is Yaws - I want to know.
>> /klacke
>> _______________________________________________
>> erlang-questions mailing list
>> erlang-questions@REDACTED
>> http://www.erlang.org/mailman/listinfo/erlang-questions
> _______________________________________________
> erlang-questions mailing list
> erlang-questions@REDACTED
> http://www.erlang.org/mailman/listinfo/erlang-questions

More information about the erlang-questions mailing list