Stress-Testing LTSP was Re: No Sound with Flash in Firefox

Jim Kronebusch jim at winonacotter.org
Fri Jul 13 17:59:34 BST 2007


> Suppose a thin client can be made to boot up exactly as normal and instead
> of waiting for a user to login, ldm automatically logs in say an ip
> address-based user (so that every client uses a different account on the
> server).  That user's xsession runs and as part of that they launch a shell
> script.  The shell script is a "user simulator".  At this point you should
> have N users logged in, doing nothing.  The "user simulator" script runs
> certain tasks at random times, with a certain probability.  Initially those
> tasks might include:
> 
>  - load a random webpage from a list (start firefox if necessary)
>  - load a random webpage from a list in a new tab (start firefox if
>    necessary)
>  - close tab in firefox
>  - close firefox
>  - open random file from a folder in openoffice (starts openoffice if
>    necessary)
>  - close openoffice
>  - open a PDF in evince.
>  - close evince.
>  - reboot thin client
>  - any other
> 
> All of the above are pretty standard applications and tasks just to get a
> starting point.  We would initially make a stab at what a typical user does
> (eg an ordinary web surfer opens 2 pages per minute on average).
> 
> To run a test you just go around booting up thin clients and letting
> them go.  Ideally you would use wakeonlan to boot them from a master
> script.  You can then:
> 
> 1. Measure RAM usage, CPU load, network load, internet bandwidth usage.
> 2. Actually sit down and use a thin client and see how responsive it feels

Gideon Romm's modified ldm adds the ability to do auto logins with LTSP 5, you could use
this to tackle the autologin part.  In the past I have booted up clients and logged in
manually and started up Kstars since it is constantly changing, at least I could get an
idea of performance that way.  I would then sit at a terminal and check webmail and type
some garbage into openoffice and just see how it felt while I used it and 30 other
machines were running KStars.  I like your idea of a script that simulates actual user
use, maybe firefox could run and hit video.google.com and run a video while other
machines opened OOo Writer and type random characters.  If the script was written in a
simple way users could possibly modify it to test the specific applications their
environment would use.  I like the idea of wake on lan but even if it could be done with
a manual boot it would be better than nothing.  This would be great if coupled with some
sort of tool for checking server parameters.  Most of our users I think have at least
two apps open at a time, most often it will be a browser with one or two tabs and a text
editor.

If anyone wants to give such a script a try I will test the crap out of it for them.  My
110 new clients should be shipping today and I will be ready to do some load testing in
a few weeks.  If nothing else maybe I'll give it a try (my scripting abilities are a bit
lacking :-).  It would be nice to have a standard way of testing and be able to put
together some sort of matrix that could state for average student use a server of size X
could handle X amount of users.  We all know that this would never be a perfect chart
for all scenarios, but to be able to provide a baseline for average use would be a huge
step.

-- 
This message has been scanned for viruses and
dangerous content by the Cotter Technology 
Department, and is believed to be clean.




More information about the edubuntu-users mailing list