geronimo-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Forrest Xia <forres...@gmail.com>
Subject Re: Performance Report for Geronimo 2.2?
Date Thu, 26 Nov 2009 13:11:04 GMT
On Thu, Nov 26, 2009 at 6:30 PM, Juergen Weber <weberjn@gmail.com> wrote:

>
> This summer I ran daytrader with the Ginder load test tool on several
> application servers. Especially with the Two-Phase commit asynchronous
> scenario there were big performance differences, I think Geronimo's
> transaction throughput was a mere third of Weblogic 10.3's (same hardware).
>
> IBM says similar of Websphere and JBoss: see slide 27 of
>
> http://public.dhe.ibm.com/software/dw/techbriefings/presentations/webspheresoa/appserver.pdf
>
> I  recorded a scenario where a trader would log on, buy some shares, sell
> them again and log out. Problem with Daytrader is that you cannot blindly
> assume that you can sell shares with constant http parameters, you cannot
> make a call like sell?s=1&s=2 as the link IDs are not indexes on the screen
> but database row ids. To change that one would have to keep a mapping from
> DB id to screen ID in the users' sessions.
> So, you need a load driver that can understand semantics.
>

Daytrader has a single servlet entry (
http://localhost:8080/daytrader/scenario), user can use it to simulate the
complex and random user actions. With it, the http parameters are changed
dynamically and randomly. Thus, user can write simple load driver scripts to
push load on server.


Forrest

Mime
View raw message