[httperf] Maximum connect burst length

Vikash Kumar vikash.kumar at oneconvergence.com
Sat Dec 1 02:35:33 PST 2012


Hi Tim,

   Thanx for reply. I was trying to understand the behavior of httperf, if
some connections are lost . What happens to the socket used by client for
initiating the connection?

   I have one question.

   *Let us suppose I requests 15000 conn with 15000 conn/sec. Now out of
this, some 14000 requests are successfully served by server, but client
doesn't get the data for 1000 conn or may be the GET requests
acknowledgments lost ? *
*
*
*  What would be the behavior of httperf at that time?*
*  In what condition httperf initiate connection once again which it deemed
to lost? Does httperf initiate new connection which are lost? *

*  I asked this question because, I am observing for total 10000 no. of
connections, httperf client is initiating around 10925 conn. *
*
*
*  Is this possible that httperf client initiate more connection if the
requested connections are not completed like normal TCP connection?*
*
*
Thanx in advance if anybody explain my question.
*
*
*Thanx once again Tim.*

On Fri, Nov 30, 2012 at 7:52 PM, Tim Terrill <TTerrill at synacor.com> wrote:

>   Hi Vikash,
>
>  My understanding is this:
>
>    1. Httperf attempts to send requests in a uniform distribution across
>    every second.  So if you ask for 10,000 connections/sec, it will initiate a
>    request every 1/10,000th of a second.  It will continue making requests in
>    the subsequent seconds irregardless of it got replies to the previous ones.
>    2. Httperf does not "assume a request has completed" or "clean up
>    sockets" at the system level.  The OS reuses (cleans up) socket after the
>    TIME_WAIT period has elapsed once the connection has been closed (either
>    due to a reply, or timeout).
>
>  I have never used the maximum connection burst length, so I'm not sure
> how it plays into this.
>
>  Hopefully this helps (and hopefully I'm right! :o)
>
> Tim****
>
>
>
>   From: Vikash Kumar <vikash.kumar at oneconvergence.com>
> Date: Friday, November 30, 2012 2:24 AM
> To: "httperf at linux.hpl.hp.com" <httperf at linux.hpl.hp.com>
> Subject: [httperf] Maximum connect burst length
>
>  Hi everyone,
>
>     Httperf o/p gives :
>
>     *Maximum connect burst length*
> *
> *
>    Can anyone please explain about this?
>    What exactly it refers to?
>    Is this burst length is in millisecond or seconds?
>    How can we get connection rate and request rate from this number?
>
>     Some important question:
>
>     *Let us suppose, we fix our connection rate to 10000 conn/sec, then
> how does the httperf generate 10000 req/sec? Are all these requests are
> made parallely ? Is this has anything to do with the burst length as we get
> at output?*
> *
> *
> *   **How does httperf assume that a request has completed and how does
> it clean up the sockets in use? *
> *
> *
> *
> *
> *Thanks*
> *
> *
> *   *
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://napali.hpl.hp.com/pipermail/httperf/attachments/20121201/3b45d984/attachment.htm


More information about the httperf mailing list