[THIN] Re: Hyperthreading

  • From: "Jim Kenzig http://thin.net" <jimkenz@xxxxxxxxxxxxxx>
  • To: thin@xxxxxxxxxxxxx
  • Date: Thu, 30 Sep 2004 21:29:03 -0400

Hi All I have culminated the links and discussion on this thread into a quick article on Ondemandaccess.com for future reference:


Jim Kenzig

Rick Mack wrote:

Hyperthreading (HT) creates a real moral problem for me, even on Windows
server 2003.

It has obvious performance advantages that you can easily see on an HT
aware operating system. BUT, when you get to a scenario where a system
is doing a serious amount of work, things can fall apart a lot faster on
an HT system than on a non-HT system.

So HT gives me a significant advantage at low to medium levels of CPU
utilization, but when things max out I'd rather not be running HT
because it starts falling apart. It's a bit like the redline on a tacho,
stay below the redline and life's sweet, go above and things could blow.

So you have to be a bit pragmatic.

Scale your TS systems so they won't hit the wall and you get all the
advantages of HT with no problems. When things start getting too busy,
add more servers. Some sort of CPU limitting software (TMULimit,
Appsense performance suite, Tscale/Expedian, Armtech, Threadmaster) to
control runaway threads is absolutely mandatory if you want a quiet life
with HT.

On file/print servers (2003) I've had some really painful personal
experience about what can happen with HT when a server (2003 cluster)
gets too busy. I'd really rather take a 10-20% dip in performance than
have a file server cut out on me unexpectedly, hanging all the TS

But if you scale things properly and be careful about tuning antivirus
settings etc, and add extra capacity when it's needed, you can benefit
from HT at minimal risk.



Ulrich Mack
Volante Systems Ltd
18 Heussler Terrace, Milton 4064 Queensland Australia. Ph: +61 7 3246 7704 email: rmack@xxxxxxxxxxxxxx web: www.volante.com.au

-----Original Message----- From: thin-bounce@xxxxxxxxxxxxx [mailto:thin-bounce@xxxxxxxxxxxxx] On Behalf Of Tim Mangan Sent: Friday, 1 October 2004 8:04 AM To: thin@xxxxxxxxxxxxx Subject: [THIN] Re: Hyperthreading

I did a lot of testing with it last year. I never saw it cause an
error, but things often stall due to the idle processing clearing out
too much on the chip when one of the logical processor idles. It caused
erratic performance. Under a heavy load it *might* cause a crash, but I
never saw one. I would definitely disable HT under 2000.

-----Original Message-----
From: thin-bounce@xxxxxxxxxxxxx [mailto:thin-bounce@xxxxxxxxxxxxx] On
Behalf Of Bob
Sent: Thursday, September 30, 2004 9:50 AM
To: thin@xxxxxxxxxxxxx
Subject: [THIN] Hyperthreading

Has anybody experimented with Hyperthreading On -vs- Off on Win2k
servers?  I've heard rumblings that it can be problematic, and perhaps
decrease performance.  I noticed that most SpecInt. winners have it
disabled.  I'm considering disabling it on one of my farm servers to
observe the results.  If anybody else has experimented I'd love to hear
about it.

Thanks in advance,
This Weeks Sponsor RTO Software
Do you know which applications are abusing your CPU and memory?
Would you like to learn? -- Free for a limited time!
Get the RTO Performance Analyzer to quickly learn the applications, users,
and time of day possible problems exist.
********************************************************** Useful Thin Client Computing Links are available at:
For Archives, to Unsubscribe, Subscribe or set Digest or Vacation mode use the below link:

Other related posts: