[httperf] httperf + libevent

[Mark Nottingham]

+1; I'm wary of using the libevent http code until I know it's been  
well-tested and worked over.

Cheers,


On 03/07/2008, at 11:27 AM, Adrian Chadd wrote:

> On Wed, Jul 02, 2008, Theodore Bullock wrote:
>
>> Glad to see some extra hands on this topic.
>>
>> I also have a partial patch that I have been working on for this very
>> topic.
>
> I saw.
>
>> The problem is that it's rather an all or nothing approach, before  
>> the
>> whole port will work.  Not exactly a big deal, but I haven't had a
>> consistent opportunity to work on the problem since I have been out  
>> of
>> school.
>
> You're trying to use the whole libevent http framework though,  
> aren't you?
> Its not a complete http client/server implementation which can be  
> reused
> everywhere.
>
> I figured it was an easy win to convert httperf over to just use the  
> libevent
> comm framework. It should perform a lot better now but I just can't  
> seem
> to tweak the parameters right to do a heavily parallel test (I can't  
> seem
> to convince it to use more than about 3 concurrent connections.)
> The difference at my current load (~ 500req/sec) is pretty striking -
> system time is 5% CPU versus 95% CPU. Very nice. :)
>
> Other bits can then be used later - like the timeout/timed events -
> to replace the duplicated work inside httperf. The HTTP code would  
> be nice
> in the distant future; it would be nice if a number of projects could
> reuse the libevent http code.
>
> I reiterate - testing please! :)
>
>
> Adrian
>
> _______________________________________________
> httperf mailing list
> httperf at linux.hpl.hp.com
> http://www.hpl.hp.com/hosted/linux/mail-archives/httperf/

--
Mark Nottingham       mnot at yahoo-inc.com