[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]

Re: [K12OSN] Client RAM requirements? / Athalon ATA server vs.Dual-Xeon with SCSI - feedback?



The X server and the kernel run on the client side.  I think you
would want at least 16MB to be comfortable, 32MB would be plenty.

-Don

Greg Long wrote:
> 
> Thanks for the input. I guess I'm missing something fundamental about
> the client side:  I rather thought they hardly need any RAM since
> everything runs on the server...I was referring to client system RAM,
> not video RAM.  For video RAm, I figured at 800x600 or 1024x768 2 or 4mb
> PCI DRAM video would be fine for 16bit color
> 
> I'll try it out and see for myself how things go.
> 
> If more RAM helps the clients, there must be more running on the client
> side than I thought.
> 
> Greg
> 
> -----Original Message-----
> From: k12osn-admin redhat com [mailto:k12osn-admin redhat com] On Behalf
> Of John Baillie
> Sent: Monday, May 27, 2002 11:07 AM
> To: k12osn redhat com
> Subject: Re: [K12OSN] Client RAM requirements? / Athalon ATA server
> vs.Dual-Xeon with SCSI - feedback?
> 
> Here is our experience with 35 clients.
> Specs:
> 
> Dual PIII 933, 2GB RAM, 3  18 GB SCSI  W/ Hardware RAID 5 10/100 NICS on
> all 2 24 Port SMC Switches. Clients P 133 - P II 233 all have 32 - 160
> MB RAM
> 
> We very little swapping taking place.
> 
> Sitting side by side is one P 133 32 MB RAM and a PII 233 with 160 MB
> RAM both have 8 MB Video RAM. The performance is identical
> 
> Performance is excellent up to 24 clients.
> 
> After 24 clients the %Processor Utilization starts climbing and when it
> reaches 80% the performance is very slow, it will peak at 97 to 98 %
> then drop down to about 65% at which point the performance is
> acceptable.
> 
> Staggering log-ons and launching Mozilla / OpenOffice in groups of 10 in
> 15 to 30 second increments helps a great deal in getting all 35 clients
> to a working OpenOffice / Mozilla desktop in short order.
> 
> A couple of impatient students triple clicking on Mozilla bogs down the
> system.
> 
> We started with 1 GB RAM and doubled it, this didn't make noticeable
> difference during the log on process.
> 
> My hunch is a network bottleneck might be responsible for the high
> processor utilization.We might try adding a 3rd NIC one for each switch.
> 
> As for video RAM this has been my observation, anything less than 8 MB
> has resulted in less than adequate redrawing of windows. I noticed sort
> of a strobe effect ie. there is a pause then the window is smaller
> another pause then the window is smaller again and so on.
> 
> For those using a 1000 MB NIC I'd be interested to know the performance
> change after adding it. Can it be throttled back to 100 then back up to
> full speed in order to gage the performance increase?
> 
> I'd like to note that as long as everyone does not log on at the exact
> same time and launch apps at the same time the system works very good.
> Our overall experience has been very good and we are currently trying to
> get the funds together to put 3 or 4 clients into every classroom.
> 
> Well that's my .02 worth and it goes without saying YMMV.
> 
> Eric and Paul, you guys have done quite an amazing thing here with
> K12LTSP.
> 
> Happy Memorial Day. Remember the Vets.
> 
> John
> 
> On Mon, 2002-05-27 at 01:09, Greg Long wrote:
> > I am getting involved with our local schools to explore the
> > possibility of a K12LTsP deployment. After some initial reading, it
> > appears for K12LTSP (2.0.2 ISO's just downloaded) almost any old
> > legacy box can become a client.
> >
> > My main question is how much RAM is required?  I should think if
> > everything runs from the server, even 4 or 8mb RAM would do fine, and
> > even a pentium 75 with a 2mb DRAM PCI video would perform nicely with
> > a 100bT NIC.  We can most likely make extensive use of existing
> > clients.
> >
> > Attached is some ballparking I did for a single CPU Athalon Server for
> 
> > a target of 12 clients utilizing 3 ATA/100 drives and a Dual Xeon for
> > a target of 24 clients.  The prices come from a little poking around
> > on pricewatch - with the recognition that we may pay a little more in
> > one area and skimp a little in another.  My thought is that is due to
> > the larger storage space per dollar (as well as client) the IDE drives
> 
> > might be a favorable option over the SCSI drives. 8.7ms seek for 12
> > clients vs 5.2ms seek for 24 clients is the number that really seems
> > to make this acceptable - plus the estimated cost per client.  I am
> > also figuring putting a separate 9 or 10gb drive for a swap-only
> > partition should help out the whole equation.  128mb/client ought to
> > help reduce swapping in the first place, until client load gets heavy.
> 
> > Certainly a smaller drive than a 10gb 7200rpm IDE
> >
> > The Xeon server could also be built more cheaply - such as 1.7 Xeons,
> > or even a single CPU Xeon.  From the reading, I've been getting the
> > impression that CPU load isn't really the performance bottleneck, but
> > rather the RAM and seek time of hard drives.
> >
> > I'll be setting up a test K12LTSP server soon at home to play with all
> 
> > this on a 1.2 Athalon /ATA100 drive and 3 or 4 existing Winboxes of
> > various speeds, booting to floppy to connect to the server - I'm not
> > up on how to configure those yet for the various Xf86 drivers, but
> > I'll read the documentation. :)
> >
> > Another question, on the site, it was mentioned that 2 SCSI drives
> > were suggested for the larger server - one for /home, although I think
> 
> > in the example /home was mounted on another server's SCSI.  I'm
> > thinking putting /home on the same box should not adversely affect
> > performance too much.
> >
> > Finally, and this is a bigger curiosity than some of the others:
> > Network congestion.  Pushing graphics over a 100bT can get slow.  Not
> > being too knowledgeable about X, my experience with this is limited to
> 
> > running VNCserver/client over my 100bT here - but I'm the only user.
> > With 24 or 32 active clients, isn't the network a limiting factor?
> > Would performance be noticeably increased with just 12 clients?
> >
> > I'd welcome any feedback on these 6 issues as described above:
> >
> > 1) RAM/CPU for clients
> > 2) IDE vs. SCSI
> > 3) Athalon vs. Dual Xeon
> > 4) Separate swap drive - would it help?
> > 5) /home on separate drive on same box
> > 6) Network congestion with 24 or 32 clients vs. 12 clients
> >
> > Big thanks,
> >
> > Greg Long
> > Klamath Linux Unix Group
> > http://www.maneuveringspeed.com/klug/index.html
> > ----
> >
> 
> .00
> (est)
> 
> Total
> 
> $1,212.00
> 
> $2,906.00
> 
> Average
> cost per
> client
> 
> $101.00
> 
> $121.08
> 
> RAM
> (mb/client)
> 128.0
> 
> 85.3
> 
> _______________________________________________
> K12OSN mailing list
> K12OSN redhat com https://listman.redhat.com/mailman/listinfo/k12osn
> For more info see <http://www.k12os.org>
> 
> _______________________________________________
> K12OSN mailing list
> K12OSN redhat com
> https://listman.redhat.com/mailman/listinfo/k12osn
> For more info see <http://www.k12os.org>

-- 
Don Christensen       Senior Software Development Engineer
djc cisco com         Cisco Systems, Santa Cruz, CA
  "It was a new day yesterday, but it's an old day now."





[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]