How to get Hyperthreading without turning on any part of ACPI?

Sheridan · 03-22-2011, 03:41 AM

Hi there,

Some time ago I've opened a thread about ACPI - as it turned out - killing my server. You may read it here if interested.

My question is how can I enable Hyperthreading support in my kernel without enabling any part of ACPI?

(Boot flag acpi=ht causes the same issue as described in above thread, so enabling any part of ACPI in any manner is not desirable)

My kernel:

[root@sheridan ~]# uname -a
Linux sheridan.******** 2.6.35.11-83.fc14.x86_64 #1 SMP Mon Feb 7 07:06:44 UTC 2011 x86_64 x86_64 x86_64 GNU/Linux

Thank you!

adamk75 · 03-22-2011, 12:47 PM

hyperthreading requires ACPI. Hence the 'acpi=ht' option which disables all of ACPI except what is needed for hyperthreading.

Adam

Sheridan · 03-22-2011, 11:52 PM

Quote:

Originally Posted by adamk75

hyperthreading requires ACPI. Hence the 'acpi=ht' option which disables all of ACPI except what is needed for hyperthreading.

Adam

Thank you for taking the time to answer me.

This is exactly what I was afraid of, because in this case, I will certainly not be able to use it. For if I enable ACPI in any way, the above linked ordeal happens.

Anyway... thank you for clarifying this. Be well..

syg00 · 03-23-2011, 12:27 AM

Do the previous kernels show the same symptoms ?.
You can't do much about the slab allocation - they are generally a symptom not the problem itself. You'll probably need to some kernel tracing to see what is doing the allocates. Not generally recommended on a prod box (but then, neither would Fedora be), but in your case it may be a necessary evil. Of course that will (hopefully) expose the miscreant, not fix it.

Sheridan · 03-24-2011, 04:38 AM

Quote:

Originally Posted by syg00

Do the previous kernels show the same symptoms ?.
You can't do much about the slab allocation - they are generally a symptom not the problem itself. You'll probably need to some kernel tracing to see what is doing the allocates. Not generally recommended on a prod box (but then, neither would Fedora be), but in your case it may be a necessary evil. Of course that will (hopefully) expose the miscreant, not fix it.

Yeah man. I know Fedora is not the best for industrial grade server, but I have little choice unless I want "them" (users I mean) to keep ***king me for new software and libraries.

I had CentOS 5.5 running on it for months, and it was fine, stable... When I was forced to be as bleeding-edge as possible, I turned to Fedora, figured - how big a mess could it be, it's essentially a beta-RedHat right? Well... apart from ACPI - it's fine... I couldn't care less about power managemenet (I run the server in Static High Performance Mode) but then, I'm not sure the performance loss from HT suffered by heavily threaded applications like MySQL, Apache, science stuff, etc etc, was worth the tradeoff... :-(

Anyway... I have no experience diagnosing this, I never seen a behaviour like this before.

Sheridan · 03-24-2011, 04:57 AM

BTW. I may have forgot to mention: the slab issue occurs if and only if I turn on any part of ACPI. I experimented with all boot options relatied that I could find on Google. If ACPI (any part of it, really) is involved, slab issue. I tried hard to reproduce some instability without ACPI, but if I leave it off, it's solid. Well, I don't have HT, but it's solid, anyway...