Linux Help

[_DeeJay_]

I'm looking for some ideas around a particular issue we're seeing on some of our (64 bit) RHEL4 boxes.

I'm not a Linux person (really) but can generally muddle though.

So anyway, the problem is that on some 4 socket dual core systems (of which we have many), /proc/cpuinfo shows the 2.8Ghz processors running at just over 1Ghz (with corresponding bogomips).

On machines which report the speed simple benchmark tests but the CPU's just less than 3 times faster. We've not running anything within Linux to take advantage of any of the speed reduction technology built into the chip, so it should just run at full speed all of the time.

So, the question is, what's the hell's going on? I presume this has to be something to do with the hardware/BIOS rather than Linux?

[JamieBeeston]

Depending on the machines, and their location (are they in a DataCentre?) the BIOS may be able to permanently disable any speed reduction technology (on our Dells it can) however it might be the machines are overheating / too hot and are being forced to slow speed..

[_DeeJay_]

JamieBeeston said:

Depending on the machines, and their location (are they in a DataCentre?) the BIOS may be able to permanently disable any speed reduction technology (on our Dells it can) however it might be the machines are overheating / too hot and are being forced to slow speed..

They're in a datacentre, not reporting any temperature related errors and I believe (but will need to double check) than the speed reduction technology is disabled is the BIOS.

To make matters worse on some of them we have 6 cpu's running at 1Ghz and the other 2 running at 2Ghz.


edited to add:

As it's a recent kernel, the cpufreq kernel modules are installed.
However, the cpuspeed daemon is disabled within the OS (for whatever reason)

Edited by _DeeJay_ on Sunday 9th December 20:14

[JamieBeeston]

_DeeJay_ said:

JamieBeeston said:

Depending on the machines, and their location (are they in a DataCentre?) the BIOS may be able to permanently disable any speed reduction technology (on our Dells it can) however it might be the machines are overheating / too hot and are being forced to slow speed..

They're in a datacentre, not reporting any temperature related errors and I believe (but will need to double check) than the speed reduction technology is disabled is the BIOS.

To make matters worse on some of them we have 6 cpu's running at 1Ghz and the other 2 running at 2Ghz.

Are you sure this isnt a software/driver issue, i've never experienced that before.

[_DeeJay_]

JamieBeeston said:

_DeeJay_ said:

JamieBeeston said:

Depending on the machines, and their location (are they in a DataCentre?) the BIOS may be able to permanently disable any speed reduction technology (on our Dells it can) however it might be the machines are overheating / too hot and are being forced to slow speed..

They're in a datacentre, not reporting any temperature related errors and I believe (but will need to double check) than the speed reduction technology is disabled is the BIOS.

To make matters worse on some of them we have 6 cpu's running at 1Ghz and the other 2 running at 2Ghz.

Are you sure this isnt a software/driver issue, i've never experienced that before.

In theory, the working and non-working boxes were built with the same kickstart scripts, so they should be identical...

[TheLearner]

Poke around dmesg and /var/logs, see if there is anything mentioned during bootup. Sounds very odd it'd run different cpu's at different speeds and lock them there.

Push comes to shove, you're on RHEL, give them a call... you are paying for a support package, use it.

[qube_TA]

Linux will auto throttle the CPU if it supports it to save power when it's idle, get the boxes to run something strenuous and cat /proc/cpuinfo again you'll see it jump up.

[_DeeJay_]

qube_TA said:

Linux will auto throttle the CPU if it supports it to save power when it's idle, get the boxes to run something strenuous and cat /proc/cpuinfo again you'll see it jump up.

Not if the cpuspeed daemon has been stopped by some bright spark. Apparently, the kernel was using the user mode process (i.e. daemon) as a guide to setting CPU speed. However, since (at Redhat's recommendation) the Unix bods has disabled the daemon it was stuck at the speed the daemon had last told it to run at (permanently, reboots and all).

So, the fix was to either unload the kernel modules, or tell it to run in performance mode (i.e full speed all the time).

That seems to have sorted it.

[_DeeJay_]

qube_TA said:

Linux will auto throttle the CPU if it supports it to save power when it's idle, get the boxes to run something strenuous and cat /proc/cpuinfo again you'll see it jump up.

You'll need to logout then back in again for the security token to be updated to have the new permissions.
Alternatively, if you can't afford to do that, you could use runas with a new user account to launch a shell as an admin.