Before You Begin

Post #99,650

5/1/03 11:50:44 AM

Before You Begin

I have a new dual-Xeon IBM Server x225 with external CPU cache. What version of the Linux kernel is known to be acceptable to this processor setup? This is a big deal, I can't afford to fuck it up.

-drl

Post #99,651

by admin

5/1/03 11:53:26 AM

IBM probably has a redbook for it...

Regards,

-scott anderson

"Welcome to Rivendell, Mr. Anderson..."

Post #99,780

by folkert

5/1/03 9:31:24 PM

RedHat 7.2/7.3 or SuSE8.1

I've done those version on most Netfinity *X* series boxes.

RH-8.0 should work as well. Debian Woody does too.

Stay away from Mandrake and RedHat 9.0.

[link\|mailto:curley95@attbi.com\|greg] - IT Grand-Master for Anti-President
[link\|http://www.iwethey.org/ed_curry/\|REMEMBER ED CURRY!]

-----BEGIN GEEK CODE BLOCK-----\nVersion: 3.12+\nGAT d+ s+:++ a C++++ UBHLO++++ P+ L+++ E---/E---- W+++ N+ o--\nK--- w--- O+ M+ V-- PS-- PE Y+ PGP++ t+ 5++ X+ R tv+ b+++ DI+++\nD++ Q2+++ Q3A+++ UT+++ UT2K3+++ G e* h--- r+++ z+++*\n------END GEEK CODE BLOCK------

Post #100,105

by broomberg

5/3/03 11:17:50 AM

I have 4 x345s which I think are close

We use RH AS but might sysadmin thinks it
is a waste of money with mediocre support,
so we may be moving off to something else.

The ext3 SUCKS when doing high IO.
Load average goes to 15, systems crawls.
Use something else if you can. Ext2
if you don't need journalling, JFS or XFS
if you do. Setup multiple file systems to
test before committing to any.

Also, turn of hyperthreading. Linux will
not know how to balance across the CPUs and
will swamp a single CPU with 2 processes while
leaving the other idle.

Don't use jumbo frames on the GB port. Wicked
fast (90MB a second!!!), then hard locks.

Yours:
[link|http://www.pc.ibm.com/us/eserver/xseries/x225.html|http://www.pc.ibm.co...xseries/x225.html]
Mine:
[link|http://www.pc.ibm.com/us/eserver/xseries/x345.html|http://www.pc.ibm.co...xseries/x345.html]

Feels like close to the same otherboard in a different
box.

I'm very happy with them. Less hassle than the Dells
I have.

What will you be doing with it?

Post #100,114

by deSitter

5/3/03 1:39:31 PM

Re: I have 4 x345s which I think are close

I did use ext3, because I can turn it off in 5 minutes. I was iffy about this, and will evaluate.

I instinctively did disable hyper-threading, so I guess your experience says that was correct.

SuSE installed fine, with SCSI and network that I'd never had experience with (Fusion MPT and Broadcom). I was impressed by how simple it was. I think IBM did this deliberately, because they actively support RedHat, SuSE, and Caldera.

It will be the heart of an inventory management system I'm developing. Is Postgres on a par with Oracle for smaller tasks?

As for the machine itself, what can one say? What a doll! Everything came installed - in the past this would have pissed me off because I like to tinker, but the saved time was welcome. IBM servers are still the bomb engineering-wise. I love the turbo cooling on startup - press the on button, WHOOSH!

The IBM people are also great about throwing deals around. Basically, I got a DDS4 tape drive for about 200 bucks!

-drl

Post #100,159

by broomberg

5/3/03 8:42:11 PM

Sounds good

Keep a CLOSE eye on the Broadcom though.
Will you be using GBit or 100 Mbit?

If GBit:
Run a process looking for missed interrupt in the messages
file. I've lost the exact text (my logs cycled) so
I can't tell you exactly what to look for, sorry.

We pretend the problem "went away", so we are not
actively searching for it. In our case we have
a limited highly technical user population, so
it seems we are more forgiving of problems. A hung
compute server simply means that one piece of the job
does not complete NOW, and then get tossed of to another
compute server.

Our problem might have been Dell specific though.

We went through HELL with the Broadcom / Tigon drivers
for a while but it seems to have settled down.

Another issue that bites us occasionaly is a stuck
process trying to access a file via NFS. They become
zombies when killed since they are waiting a system
call return that will never happen. The only obvious
problem this causes is it drives up the load average
without slowing the box down. Since atd then thinks
the box is busy it then refuses to "at" tasks.

For a small inventory system (< 1 million records across
ALL tables) PostGreSQL should be fine. I'd expect to
to be able to cache almost your entire base in main memory.

Oh.

How much memory did you get?

But I'd have to go with Greg's generalized SQL direction
here. Use SAP-DB if you can. It is VERY close to Oracle
(syntax wise) and in general stability/recoverability
features. On the other hand, PostGreSQL has far more
flexible back end engine for your triggers so if you
do a lots of user defined functions and trigger,
PostGreSQL is the winner. Admin VS Coder issues here.

This sounds like a vertical mission critical
app. Load test the system for WEEKS of automated
pounding before you release it. We always seem to be
bitten by something when under load. Better to find
it now and reconfigure around it, then depend on the
current configuration and get nailed.

Welcome to IWETHEY!