[K12OSN] LTSP Lab Slow to Boot

Terrell Prude' Jr. microman at cmosnetworks.com
Sat Sep 13 05:18:45 UTC 2008


On Friday 12 September 2008 21:36, James P. Kinney III wrote:

> People: you MUST do some engineering on these installations. You can't
> just expect every machine you have to run off a crappy little
> network/server/used hard drive, etc. The TC environment is CPU
> intensive, VERY RAM INTENSIVE and a substantial impact on network
> bandwidth. I've been there, done that and I can't emphasize enough to
> spend the $$ on the server and switches and go cheap on the clients. The
> little, tiny clients will not drain the network as badly as they run
> slower and can't be such hogs. If the clients are 2GHz machines, that X
> server is gonna FLY!! Which means it's pounding the server for data.

I have to agree.  I have said countless times that one TuxType session *by 
itself* will suck up 72Mb/sec of data.  You get similar numbers with TuxMath.  
I know that because I took a little time and actually measured it.  You've 
got to do this sort of thing.  BTW, that's on an older client running an ATI 
3D Rage Pro video card from 1998, a Pentium-166 MMX CPU, with a 3Com 10/100 
network card.  Old cheapie stuff, and it *still* sucked up bandwidth.

We've also seen folks actually seriously want to run 20 clients on a 1GB 
DRAM "server" with a single CPU in the 1.4GHz range...without so much as 
Gig-E on this "server".  They try it, and they wonder why it doesn't perform 
to their expectations.  That's like trying to run Windows Vista Ultimate with 
full MS Office 2007, Adobe Photoshop, and a big Flash animation...in 256MB 
DRAM.

--TP




More information about the K12OSN mailing list