|
Home > Archive > Macromedia Flash Server > November 2005 > FCS hyperthreading
You are viewing an archived Text-only version of the thread.
To view this thread in it's original format and/or if you want to reply to
this thread please [click here]
| Author |
FCS hyperthreading
|
|
| Michael Weck 2005-11-21, 5:45 pm |
| Hi All,
I noticed a post on Slashdot the other day regarding Intel
Hyperthreading that said in certain situations it actually caused a
decrease in performance under higher loads on certain applications.
Does anyone have any experience with disabling hyperthreading on loaded
servers and seeing any significant change in performance, either good or
bad?
Is FCS/FMS written to take advantage of special HT instructions that
Intel provides?
Thanks,
Mike Weck
=-----------------------------------------------------------
Supported by Fig Leaf Software - http://www.figleaf.com
=-----------------------------------------------------------
To change your subscription options or search the archive:
http://chattyfig.figleaf.com/mailman/listinfo/flashcomm
| |
| Kevin Day 2005-11-21, 5:45 pm |
|
On Nov 21, 2005, at 10:27 AM, Michael Weck wrote:
> Hi All,
>
> I noticed a post on Slashdot the other day regarding Intel
> Hyperthreading that said in certain situations it actually caused a
> decrease in performance under higher loads on certain applications.
>
> Does anyone have any experience with disabling hyperthreading on
> loaded servers and seeing any significant change in performance,
> either good or bad?
>
> Is FCS/FMS written to take advantage of special HT instructions
> that Intel provides?
>
> Thanks,
>
> Mike Weck
>
Hyperthreading doesn't add any instructions really, it adds a second
pseudo-processor... A HT enabled CPU appears to the OS as two CPUs.
You still only have one "brain" per CPU, but the operating system can
tell it to do two things at a time.
This lets the CPU more quickly change from one task to another when
it's stuck waiting for something, but only one is actually doing real
work at once. A good analogy would be like having two computers on
your desk. You can still only really pay attention to one of them,
but if one is stuck doing something for a long period you could
switch to a second computer to get some stuff done.
The downside is that you now have two tasks trying to make use of the
same resources (same memory bus, same actual processing power, etc).
If the OS doesn't understand this and schedule tasks accordingly, or
if this is causing the real "brain" of the CPU to spend more time
switching back and forth between tasks than doing actual work,
overall performance can suffer. (Like the distraction level of two
people trying to talk to you at once)
Overall, I've found the performance for most internet server based
tasks to be about the same if it's on or off. I haven't measured FCS
specifically, but I'm willing to bet it's not affected either way, if
the server is otherwise unused.
=-----------------------------------------------------------
Supported by Fig Leaf Software - http://www.figleaf.com
=-----------------------------------------------------------
To change your subscription options or search the archive:
http://chattyfig.figleaf.com/mailman/listinfo/flashcomm
| |
| Michael Weck 2005-11-21, 5:45 pm |
| Thanks for the quick answers 
Frank Hoang wrote:
> We have production servers with both HT enabled and disabled,
> Honestly we don't see a difference.
>
> Going to be testing on Opterons next month to compare performance
> between Intel and AMD
>
> Kevin Day wrote:
>
>
>
>
>
=-----------------------------------------------------------
Supported by Fig Leaf Software - http://www.figleaf.com
=-----------------------------------------------------------
To change your subscription options or search the archive:
http://chattyfig.figleaf.com/mailman/listinfo/flashcomm
| |
| Frank Hoang 2005-11-21, 5:45 pm |
| We have production servers with both HT enabled and disabled,
Honestly we don't see a difference.
Going to be testing on Opterons next month to compare performance
between Intel and AMD
Kevin Day wrote:
>
> On Nov 21, 2005, at 10:27 AM, Michael Weck wrote:
>
>
>
> Hyperthreading doesn't add any instructions really, it adds a second
> pseudo-processor... A HT enabled CPU appears to the OS as two CPUs.
> You still only have one "brain" per CPU, but the operating system can
> tell it to do two things at a time.
>
> This lets the CPU more quickly change from one task to another when
> it's stuck waiting for something, but only one is actually doing real
> work at once. A good analogy would be like having two computers on
> your desk. You can still only really pay attention to one of them,
> but if one is stuck doing something for a long period you could
> switch to a second computer to get some stuff done.
>
> The downside is that you now have two tasks trying to make use of the
> same resources (same memory bus, same actual processing power, etc).
> If the OS doesn't understand this and schedule tasks accordingly, or
> if this is causing the real "brain" of the CPU to spend more time
> switching back and forth between tasks than doing actual work,
> overall performance can suffer. (Like the distraction level of two
> people trying to talk to you at once)
>
> Overall, I've found the performance for most internet server based
> tasks to be about the same if it's on or off. I haven't measured FCS
> specifically, but I'm willing to bet it's not affected either way, if
> the server is otherwise unused.
>
>
>
> =-----------------------------------------------------------
> Supported by Fig Leaf Software - http://www.figleaf.com
> =-----------------------------------------------------------
>
> To change your subscription options or search the archive:
> http://chattyfig.figleaf.com/mailman/listinfo/flashcomm
--
Frank Hoang
=-----------------------------------------------------------
Supported by Fig Leaf Software - http://www.figleaf.com
=-----------------------------------------------------------
To change your subscription options or search the archive:
http://chattyfig.figleaf.com/mailman/listinfo/flashcomm
| |
| Edward Chan 2005-11-21, 5:45 pm |
| No. There is no special HT code in FCS/FMS. We'll just leave it to the processor (whether it's a real one or fake one in the case of HT) to do the thread scheduling.
-----Original Message-----
From: flashcomm-bounces-1Ss2GqJETD3yZ38Mhd3e/9ZfFG6BLHNm@public.gmane.org on behalf of Michael Weck
Sent: Mon 11/21/2005 8:27 AM
To: flashcomm-1Ss2GqJETD3yZ38Mhd3e/9ZfFG6BLHNm@public.gmane.org
Subject: [FlashComm] FCS hyperthreading
Hi All,
I noticed a post on Slashdot the other day regarding Intel
Hyperthreading that said in certain situations it actually caused a
decrease in performance under higher loads on certain applications.
Does anyone have any experience with disabling hyperthreading on loaded
servers and seeing any significant change in performance, either good or
bad?
Is FCS/FMS written to take advantage of special HT instructions that
Intel provides?
Thanks,
Mike Weck
=-----------------------------------------------------------
Supported by Fig Leaf Software - http://www.figleaf.com
=-----------------------------------------------------------
To change your subscription options or search the archive:
http://chattyfig.figleaf.com/mailman/listinfo/flashcomm
|
|
|
|
|