How to take part in discussions on other sites? (e.g. the LiveJournal)
HTT/SMP servers instability on 5.3-STABLE
Hello Guest
  
  • Login
• Register…
• Start blog
  • Who, Where, When
• What can I do?
• What to Read?
  • Polls
• Avatars
• Interests
  • Cities and Countries
• Random blog
• Users search
  • Search
• Games
• Tests
• RYXI
  • Сообщества
• Talxy Chat
• Horoscope
• Online
 
Зарегистрируйся!

RYXI > FreeBSD > HTT/SMP servers instability on 5.3-STABLE 27 December 2004 23:57:06

  Recent blog posts: 
  They have birthday today: 
  Forums:   
  Discuss: 
  Recent forum topics: 
  Recent forum comments:
  Moderators:

HTT/SMP servers instability on 5.3-STABLE

Guest 27 December 2004 23:57:06
 Hi,

I have a Dual Xeon 2.4 and a Dual Xeon 2.8 servers running with
HyperThreading, ACPI, and SMP enabled.

The 2.8 server won't stand for more than 5 days without crashing, and the
2.4 server was up 30 days crashed, now was up 12 days, and crashed.

I didn't have a debugging kernel, I'll be building one when the datacenter
reboots the server. I also don't have any panic messages.. I have,
however, a few questions:

- machdep.cpu_idle_hl­t - I've seen a lot on google about this sysctl, but
still don't fully understand it. What does this sysctl really changes?

- HyperThreading - Do I really have a performance increase with HTT turned
on? I've heard it can penalize performance because the scheduler isn't
optimized for logical CPUs. Does having HTT enabled impacts the stability
of the system?

- ACPI - I'll be disabling ACPI along with HTT to see if the server
doesn't crash for awhile. Is ACPI on 5.3-STABLE (around November 1st, it
was pre-release) still a problem?

Last but not the least, my 5.3-STABLE version is from a few days before
the release. Since I had created a few jails by then, I didn't upgrade the
system to use the -RELEASE. Was there any last-standing problem a few days
before the release that could be causing my instability problems?


Please share some common dual processor system knowledge, perhaps I'm
missing something really obvious and making these servers unstable.

Regards,

Hugo

___________________­____________________­________
Add comment
Guest 10 December 2004 20:29:56 permanent link ]
 Hi,

I have a Dual Xeon 2.4 and a Dual Xeon 2.8 servers running with
HyperThreading, ACPI, and SMP enabled.

The 2.8 server won't stand for more than 5 days without crashing, and the
2.4 server was up 30 days crashed, now was up 12 days, and crashed.

I didn't have a debugging kernel, I'll be building one when the datacenter
reboots the server. I also don't have any panic messages.. I have,
however, a few questions:

- machdep.cpu_idle_hl­t - I've seen a lot on google about this sysctl, but
still don't fully understand it. What does this sysctl really changes?

- HyperThreading - Do I really have a performance increase with HTT turned
on? I've heard it can penalize performance because the scheduler isn't
optimized for logical CPUs. Does having HTT enabled impacts the stability
of the system?

- ACPI - I'll be disabling ACPI along with HTT to see if the server
doesn't crash for awhile. Is ACPI on 5.3-STABLE (around November 1st, it
was pre-release) still a problem?

Last but not the least, my 5.3-STABLE version is from a few days before
the release. Since I had created a few jails by then, I didn't upgrade the
system to use the -RELEASE. Was there any last-standing problem a few days
before the release that could be causing my instability problems?


Please share some common dual processor system knowledge, perhaps I'm
missing something really obvious and making these servers unstable.

Regards,

Hugo





___________________­____________________­________
Add comment
Kris Kennaway 11 December 2004 00:41:56 permanent link ]
 On Fri, Dec 10, 2004 at 05:58:06AM -0600, klr@6s-gaming.com wrote:> Hi,>
I have a Dual Xeon 2.4 and a Dual Xeon 2.8 servers running with> HyperThreading, ACPI, and SMP enabled.>
The 2.8 server won't stand for more than 5 days without crashing, and the> 2.4 server was up 30 days crashed, now was up 12 days, and crashed.>
I didn't have a debugging kernel, I'll be building one when the datacenter> reboots the server. I also don't have any panic messages.. I have,> however, a few questions:>
- machdep.cpu_idle_hl­t - I've seen a lot on google about this sysctl, but> still don't fully understand it. What does this sysctl really changes?

It doesn't put the CPU to sleep in the idle loop, which can cause a
performance loss in some cases, but also uses less power and generates
less heat.
- HyperThreading - Do I really have a performance increase with HTT turned> on?

It depends on the workload you're doing. Try it and see.
I've heard it can penalize performance because the scheduler isn't> optimized for logical CPUs. Does having HTT enabled impacts the stability> of the system?

No.
- ACPI - I'll be disabling ACPI along with HTT to see if the server> doesn't crash for awhile. Is ACPI on 5.3-STABLE (around November 1st, it> was pre-release) still a problem?

Only if the problem is with your machine, i.e. if you have a buggy
BIOS. Don't discount this; look for an update from your vendor.
Last but not the least, my 5.3-STABLE version is from a few days before> the release. Since I had created a few jails by then, I didn't upgrade the> system to use the -RELEASE. Was there any last-standing problem a few days> before the release that could be causing my instability problems?

It's quite possible. The release was made when it was and not earlier
precisely because there were bugs in earlier versions.

Kris
Add comment
John Baldwin 27 December 2004 23:57:06 permanent link ]
 On Friday 10 December 2004 03:48 pm, Kris Kennaway wrote:> On Fri, Dec 10, 2004 at 05:58:06AM -0600, klr@6s-gaming.com wrote:> > I've heard it can penalize performance because the scheduler isn't> > optimized for logical CPUs. Does having HTT enabled impacts the> > stability of the system?>
No.

Actually, for 5.3 it does. HEAD has the problem fixed, not sure about
RELENG_5 yet, but due to the way IPIs are done in 5.3, having more than 2
CPUs can be very destabilizing, so disabling HTT on a 2-way system so that it
goes from 4 CPUs to 2 CPUs can help stability very much.

--
John Baldwin <jhb@FreeBSD.org> <>< http://www.FreeBSD.­org/~jhb/
"Power Users Use the Power to Serve" = http://www.FreeBSD.­org
___________________­____________________­________
Add comment
 

Add new comment

As:
Login:  Password:  
 
 
  
 
Пожалуйста, относитесь к собеседникам уважительно, не используйте нецензурные слова, не злоупотребляйте заглавными буквами, не публикуйте рекламу и объявления о купле/продаже, а также материалы нарушающие сетевой этикет или УК РФ.


RYXI > FreeBSD > HTT/SMP servers instability on 5.3-STABLE 27 December 2004 23:57:06

see also:
Re: Pigpoints 10-17-2005 Hand #3…
a hand with a lot of spades..
пройди тесты:
see also:

  Copyright © 2001—2008 RYXI
Idea: Miсhael Monashev
Помощь и задать вопросы можно в сообществе support.ryxi.com.
Сообщения об ошибках оставляем в сообществе bugs.ryxi.com.
Предложения и комментарии пишем в сообществе suggest.ryxi.com.
Информация для родителей.
Write us at:
If you would like to report an abuse of our service, such as a spam message, please .