Hi everyone, first post here... sorry to say it's a long one...

I have this quad opteron machine, based on a Tyan k8qs pro (4x opteron 846). It's meant for a database server on a high traffic dynamic website. For a few weeks, it was up and running very well with 16gb of ram on RedHat9 (w/plain old kernel 2.4.26), and the latest stable version of postgres. For storage, I'm using a powervault 220s w/14 15krpm 36gb drives in a raid 10.

The problem is this: ever since I tried to upgrade to 32gb of ram, the machine stalls periodically under heavy IO loads (disk). It comes back after a few seconds, but that's not a stable condition for a production environment. I've had this happen in RedHat 9, RedHat AS3.0 x86, and RedHat AS3.0 x86_64. This behavior still exists now that I've brought the machine back down to 16gb.

So I'm weighing OS options now, and it should be a dependable linux distro, made for or at least stable with opteron cpus. I've tried the ones above, along with Fedora Core 2. FC2 ran great, but it's not really a 'reliable production os', or at least that's what the RedHat rep said at LinuxWorld last week. Has anyone had experience w/SuSE and the opteron? I really want to get this machine back into production 1) because it made the site incredibly fast, and 2) it's an expensive thing to junk. Thanks in advance...

Aaron

suse9.1 amd 64 cut was great on my dual opteron box.

Thanks for the reassurance. I'm installing SuSE Enterprise 9 right now...

One thing to check, make sure that interrupts are being processed by all the CPU's and not just one of them. For example -

CODE
jes@cerberus[9:44pm]~> ps aufgx | grep irq
root         3  0.0  0.0     0    0 ?        SWN  Jun04   0:00 [ksoftirqd/0]
root         4  0.0  0.0     0    0 ?        SWN  Jun04   0:00 [ksoftirqd/1]
jes      24080  0.0  0.0  3952  632 pts/230  S    21:58   0:00              \_ grep irq

jes@cerberus[9:44pm]~> cat /proc/interrupts
          CPU0       CPU1
 0: 1431064056          0    IO-APIC-edge  timer
 2:          0          0          XT-PIC  cascade
 8:          0          0    IO-APIC-edge  rtc
12:      76974          0    IO-APIC-edge  i8042
14:    7766660          0    IO-APIC-edge  ide0
15:       4705          0    IO-APIC-edge  ide1
16:   15094470        122   IO-APIC-level  ide2, ide3
19:       1691          0   IO-APIC-level  ohci_hcd, ohci_hcd
24:  110654028          0   IO-APIC-level  eth0
NMI:     119763      99595
LOC: 1430211941 1430213506
ERR:          0
MIS:        122


Look how all the interrupts are being handled by CPU0. Running IRQBalance will fix that.

CODE
[root@cerberus root]# /usr/local/bin/irqbalance

[root@cerberus root]# ps aufgx | grep irq
root         3  0.0  0.0     0    0 ?        SWN  Jun04   0:00 [ksoftirqd/0]
root         4  0.0  0.0     0    0 ?        SWN  Jun04   0:00 [ksoftirqd/1]
root     24769  0.0  0.0  3952  632 pts/230  S    22:07   0:00                      \_ grep irq
root     24719  0.0  0.0  2564  456 ?        S    22:06   0:00 /usr/local/bin/irqbalance

[root@cerberus root]# find /  > /dev/null

[root@cerberus root]# cat /proc/interrupts
          CPU0       CPU1
 0: 1431391262     262621    IO-APIC-edge  timer
 2:          0          0          XT-PIC  cascade
 8:          0          0    IO-APIC-edge  rtc
12:      76974          0    IO-APIC-edge  i8042
14:    7769043       2345    IO-APIC-edge  ide0
15:       4758          0    IO-APIC-edge  ide1
16:   15097139       2284   IO-APIC-level  ide2, ide3
19:       1691          0   IO-APIC-level  ohci_hcd, ohci_hcd
24:  110675282          0   IO-APIC-level  eth0
NMI:     119772      99602
LOC: 1430801725 1430803289
ERR:          0
MIS:        122



Now both CPU's are handling interrupts, I did the "find" command to stress the disks a little.



-------------------------
The opinions expressed above do not represent those of Advanced Micro Devices or any of their affiliates.
http://www.shellprompt.net
Unix & Oracle Web Hosting Provider powered by AMD Opterons

Cool, thanx for the advice.

SuSE 9 is up and running (well). We tested w/actual data for half a day and we're going to production mode now (we need the speed boost on the site pretty badly).

When you mention "performance problems" are you still referring to the original IO problem or some other software related performance issue? Remember, by throwing more hardware at the solution you're not "solving it" only masking it.

To be honest, I pretty much haven't come across a production system with software performance problems that could *only* be solved by more hardware.

-------------------------
The opinions expressed above do not represent those of Advanced Micro Devices or any of their affiliates.
http://www.shellprompt.net
Unix & Oracle Web Hosting Provider powered by AMD Opterons