[httperf] httperf: connection failed with unexpected error 0 & low performance

Vikash Kumar vikash.kumar at oneconvergence.com
Fri Oct 19 11:19:02 PDT 2012


Hi Rick,

   System parameters are set up to optimal limit like fd's in server and
client are raised to 400000

   Local port range is also much higher. tcp fin-timeout has been reduced
to 30 secs etc. After setting all these things we are getting higher no. of
connection with intel back to back. TCP recycle and reuse options are also
1.

   Point is after the setup of 3 node topology and requesting connection
via haproxy, performance is dropping drastically.

   All the three NIC cards are 10G Intel NIC card and we are using SFP
cable.

Regards

On Fri, Oct 19, 2012 at 11:08 PM, Rick Jones <rick.jones2 at hp.com> wrote:

> On 10/19/2012 10:10 AM, Vikash Kumar wrote:
>
>> Hi Raufoeh,
>>
>>      With Intel back to back 10G NIC card, I was able to around 28K
>> reply rate with 1kb file size with Apache server.
>>
>>      But when using haproxy that means with 3 node setup like
>>
>>
>>                                 *haproxy host*
>> *                                /         \*
>> *                               /             \*
>> *                             /                 \*
>> *                       Client                Server*
>> *
>> *
>> **connection rate is good but request rate and reply rate are dropping
>>
>> drastically. All the three machines are 8 core desktop m/c. On the
>> haproxy site it has been published that haproxy can sustain more than
>> 30K connection.
>>
>
> Are you quite sure that you are achieving 10000 connections established
> per second?  Is that backed-up by the networking statistics on the systems
> involved?
>
>
>      I am unable to find out the reason to such a  poor performance
>> whether it is issue with host m/c or haproxy.
>>
>
> What do the networking statistics look like on all three systems? Anything
> about retransmissions or queue overflows etc?
>
> Also, how do the stacks involved deal with "wrapping" the four-tuple space
> with end-points in TIME_WAIT?  That is, if there is only the one client IP,
> and one haproxy host IP and one well-known port for the proxy, there will
> then be only as many unique "names" for the TCP connections as there are
> port numbers used on the client.  If the client code is not making explicit
> calls to bind() to, for example, try to use ports 5000 through 65535 it
> will be limited to the anonymous port range.  Often that is 49152 through
> 65535.  With a 60 second TIME_WAIT that would be (65535-49152)/60
> connections per second or higher looking for TIME_WAIT reuse.
>
> rick jones
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://napali.hpl.hp.com/pipermail/httperf/attachments/20121019/67452d84/attachment-0001.htm


More information about the httperf mailing list