[Seaside] Squeak async i/o performrance, was Re: HTTP Performance

Stephen Pair stephen at pairhome.net
Fri Nov 21 04:10:50 UTC 2003


Jimmie Houchin wrote:

> Stephen Pair wrote:
>
>> Ned Konz wrote:
>>
>>> On Thursday 20 November 2003 8:36 am, Stephen Pair wrote:
>>>
>>>> ----
>>>> Linux:
>>>> ab running from a Linux 1Ghz PIII laptop to a server on the same
>>>> maching...apache serving a very small hello world file, comanche 
>>>> serving
>>>> a hello world string (in memory):
>>>>
>>>> Apache 1.3.29: scored consistently around 240 rps (no failed requests)
>>>> Comanche 6.2 with priority set to 80: scored consistently around 
>>>> 230 rps
>>>> (no failed requests)
>>>> ----
>>>>   
>>>
>>>
>>>
>>> I'm curious as to what Unix VM you're using.
>>>
>>> Could you try my latest Linux VM (you can get it from
>>> http://bike-nomad.com/squeak/linux-squeak-vm-3.6g-2.zip)?
>>>  
>>
>> I just tried using your VM and I observed about 15 - 20% more 
>> requests per second...nice.  This is using "ab -n 1024 -c 16 <url>" 
>> fetching a 7620 byte document held in memory.  So, do tell, what 
>> magic is in that VM?  ;)
>
>
> Do I understand this to mean that with this VM it outperforms Apache 
> 1.3.29?  That would be cool. :)
>
> Jimmie Houchin


No, I found a few bugs in my previous tests...first, I was relying on 
the documentIndex feature of apache for serving the test file and that 
appears to significantly affect apache's performance...when I give the 
full URL to the file, I get better performance with apache:

These results are running "ab -n 1024 -c 16 <url>" against a server 
running on the same machine (a 1Ghz PIII linux laptop)

apache 1.3.29: ~650 rps for a 15 byte document
comanche 6.2 (Ned's unix vm): ~230 rps for the same 15 byte document (it 
appears that Ned and Ian's VM aren't noticeably different when serving a 
small document)

With a 7620 byte document:

apache: ~640 rps
comanche: ~87 rps

So, with a realistic sized document, comanche/squeak serves at ~13% the 
rate of apache/linux.

- Stephen





More information about the Squeak-dev mailing list