[Building Sakai] Strange load testing results

[Remi Saias]

Thanks for the input. I tested with a remote grinder as well as a local 
one, in some of my test configurations (I used around 15 combinations 
out of 3 servers, 2 laptops and one desktop).

The configuration which was the fastest actually became faster when I 
tested it from another machine (on the same network). The resources 
freed by the grinder not running locally improved the performance of 
Sakai even with the network overhead. This is a home network, not even a 
real 100Mb (I use a hub, not a switch) and our servers are on a Gigabit 
Ethernet so I figure that when I use one to load-test another, the 
network is not a bottleneck... Anyway the CPU where tomcat runs is 
always at 99% when under load.

I just realize now that my desktop machine which is the fastest for 
running sakai up to now is on a 64 bit hardware. Could it be enough to 
explain the better performance even if it's less powerful?

Remi
P.S: my grinder script is rather basic (and totally undocumented!) right 
now as it only test a basic scenario but once this rush is past I should 
be able to share something more interesting!

-------- Message original --------
Sujet : Re: [Building Sakai] Strange load testing results
De : Steven Githens <swgithen at mtu.edu>
Pour : Remi Saias <remi.saias at hec.ca>
Copie à : sakai-dev <sakai-dev at collab.sakaiproject.org>
Date : 2009-11-09 08:36
> I find a lot of the time, my desktop machine is way faster, but often 
> assumed it wasn't going over the network, and skipping a bunch of 
> other real life infrastructure things.  Now that you mention, it would 
> be a good thing to look into and quantify.
>
> -s
>
> Remi Saias wrote:
>> Hello all,
>>
>> In the past weeks I have been performing some load tests on our 
>> Sakai+OpenSyllabus server at HEC Montreal as we are planning to go 
>> live for a pilot project in January.
>>
>> I used The Grinder which I found very efficient and easy to use, 
>> considering that OpenSyllabus is strongly dependent on Ajax requests 
>> and for this reason I could not use more classic tools relying on 
>> defined pages with fixed URLs etc.
>>
>> I tested various configurations with a single scenario and was very 
>> surprised to find out that our development laptops (regular laptops 
>> with a CoreDuo under XP Pro) were running twice as fast as our linux 
>> server which has 4 Xeon CPU at 2.8Ghz. In each case the database 
>> server (mySQL 4 and 5, tested both) was running on the same machine.
>>
>> As I suspected something was incorrectly configured on the linux 
>> server, I tried it at home on a less powerful linux box (Dual Core 
>> 3Ghz vs Quad Xeon) and, surprise, I got a performance 4 times better. 
>> I used the exact same set of sakai tools, sakai version and tomcat 
>> version.
>>
>> I am still investigating this but I am starting to run out of ideas! 
>> I have read confluence and added query caching (as per 
>> http://confluence.sakaiproject.org/display/DOC/MySQL+Administration 
>> <http://confluence.sakaiproject.org/display/DOC/MySQL+Administration>), 
>> but it didn't change anything at all. I tried tweaking various mySQL 
>> parameters without much impact!
>>
>> Does this ring a bell to someone!? I have not yet investigated the 
>> server OS/hardware configuration (maybe the disk setup, raid or 
>> such...) but I thought that it might be faster to simply ask! :-)
>>
>> BTW, Sakai 2.6.0 was 20% faster than 2.5.x in my tests.
>>
>> Thanks for any help!
>> -- 
>> Rémi Saïas
>> Analyste en informatique - Technopédagogie
>> Gestion des technologies de l'information - HEC Montréal
>> Projet Sakai-OpenSyllabus: 514.340.6776 - Édifice Decelles: 4521
>>   
>> ------------------------------------------------------------------------
>>
>> _______________________________________________
>> sakai-dev mailing list
>> sakai-dev at collab.sakaiproject.org
>> http://collab.sakaiproject.org/mailman/listinfo/sakai-dev
>>
>> TO UNSUBSCRIBE: send email to 
>> sakai-dev-unsubscribe at collab.sakaiproject.org with a subject of 
>> "unsubscribe"
>
>

-- 
Rémi Saïas
Analyste en informatique - Technopédagogie
Gestion des technologies de l'information - HEC Montréal
Projet Sakai-OpenSyllabus: 514.340.6776 - Édifice Decelles: 4521