Immanuel-

I don't have direct experience with a configuration like that, but maybe 
I can help with a few things.  The dual-core CPUs like the Core Duo and 
Athlon64 X2 (I have one of these) are basically an SMP-on-a-chip, so a 
single dual-core should be equivalent to two SMP single-core chips.

In the scenario you mention, which processes are using the most CPU 
time?  What does top (or even better, htop) show?  I imagine it is 
Firefox using the most CPU.  Complicated web pages can take a lot of CPU 
time to render, and if they have a lot of Flash animations can eat a lot 
of bandwidth to the terminal. 

A quick way to view the total activity of a system is good old xosview, 
with the following options:
/usr/bin/xosview -load -page -int +net

This will give you CPU load and total network bandwidth.  If it is 
Firefox, before you replace good hardware, is there anything you can do 
to reduce the load?  If they are accessing in-house web pages, perhaps 
some tweaks to the HTML would help, or disabling extraneous flash 
animations with a plugin.  Also, it wouldn't hurt to try another 
browser.  Perhaps Konqueror handles these pages more efficiently.  It 
really depends on if they are accessing the same pages over and over, or 
is it all over the place.  MySpace pages are the worst, because they 
almost all have text and images over a fixed background image with 
transparency everywhere, so blocking MySpace could help, if that was the 
culprit. 

Skepticism is always healthy with dealing with benchmarks or performance 
predictions.  It really depends on the workload of YOUR system.  Your 
workload could have some peculiar requirement that causes (for instance) 
firefox to need 5 seconds of CPU time per page load.  If 60 users 
require 5 seconds apiece at about the same time, then 2 CPUs are going 
to churn for 60 x 5 / 2 = 150s of wall time, plus overhead.  There is no 
magic that Linux or LTSP can do to avoid that fact; something will need 
to change at the application or data level (or run this application 
locally on the terminal, as a last resort).  If the HTML is a simple 
data entry application that uses simple forms and FF can render the page 
in .1 seconds, then you could possibly put 200 users on that same 
machine and it would be snappy.

-Todd

Immanuel Derks wrote:
> Hi all,
>
> Just wondering at the moment how people feel about the difference in
> performance characteristics between the latest Xeon core duo breed
> processors and the old fashioned dual Xeon server layouts for LTSP.
>
> We have multiple IBM X235/x236 with double Xeon 3GHz processors
> outfitted for a failover LTPS configuration at our school (80-100
> clients) and are wondering whether we can simplify the configuration by
> making use of a single server with double Xeon dual core processors (say
> an X3650 that comes with 2 dual-core Xeon 2.33/2.66 or 3.0GHz)
>
> Can anybody testify a configuration with similar loads who made such a
> swap? I know dual core processors don't have a similar performance as 2
> separate xeons at the same speed, but one might wonder with all the
> threaded apps en memory use in LTSP, that it could stack up to it...
>
> I must say I feel a bit skeptical sometimes at the performances quotas
> that I sometimes see here on the list, since we noticed (even on our
> glass backbone) that x236 servers with 3GHz processors and 8GB ram
> fitted are really dropping performance to a slow when 60 people login (2
> classes) and start working on office and firefox stuff at the same time.
> That's why we needed 2 similar servers and we are doing fine now.
> As soon as the CPU loads get over the 65% peak loads, we get a real drop
> in performance, but no iowait or memory hogs or anything....
> (we basically run standard RedHat 4 edu edition)
>
> Kind regards,
>
> Immanuel Derks
>
>   


-------------------------------------------------------------------------
Take Surveys. Earn Cash. Influence the Future of IT
Join SourceForge.net's Techsay panel and you'll get the chance to share your
opinions on IT & business topics through brief surveys - and earn cash
http://www.techsay.com/default.php?page=join.php&p=sourceforge&CID=DEVDEV
_____________________________________________________________________
Ltsp-discuss mailing list.   To un-subscribe, or change prefs, goto:
      https://lists.sourceforge.net/lists/listinfo/ltsp-discuss
For additional LTSP help,   try #ltsp channel on irc.freenode.net

Reply via email to