[THIN] Re: Page file max out
- From: "Tim Mangan" <tmangan@xxxxxxxxxxxx>
- To: <thin@xxxxxxxxxxxxx>
- Date: Fri, 21 Jan 2005 23:00:07 -0500
Steve
Not on Win2k. The problem with Win2k and other "non hyperthread aware" OSs is
that when one of the processors in the physical/logical pair becomes idle (I
suspect only idle for a while) it can send a control to the processor that has
an effect on the logical paired processor (e.g. 0/2, 1/3). The effect of this
isn't clear to me. It looks to me that this effect probably clears out too
much cache (decreasing performance), but perhaps at the wrong time it might
cause problems such as Chris relates.
By the way, that quote looks suspicious. Of course the processor is SMT aware
-- how could Intel develop a pseudo-dual processor that wasn't? Portions of
the Xeon processor are shared, portions (including part of the on-die cache)
are separate. It is technically possible for one logical processor to have
impact on cache lines for the other processor, but not to the level implied in
that quote. The nature of hyperthreading saves cost (over a full dual
processor design) by taking advantage of logical processor stalls that occur in
the cache lines (even on a uniprocessor). This is why we see HT increasing
performance (in an HT aware OS) of about 70% when the second processor is
enabled.
And you don't need HT or SMT to get the benefits of add on CPU/Memory products.
I might be biased on which is better though...
Tim Mangan
-----Original Message-----
From: thin-bounce@xxxxxxxxxxxxx [mailto:thin-bounce@xxxxxxxxxxxxx] On Behalf Of
Steve Parr
Sent: Friday, January 21, 2005 7:07 PM
To: thin@xxxxxxxxxxxxx
Subject: [THIN] Re: Page file max out
The logic then if I am hearing it correctly is: for Win2K hyperthreading can
offer performace benefits when the load is generally âunder controlâ and
the limits are not being reached. However if the load is approaching the limits
more often than not hyperthreading is likely to cause problems and outweighs in
my opinion the perf. boost when the load is lighter(and exactly when a perf.
boost is probably not as much needed)
On a diff. But related note
CPU/memory control software seems to be an interesting option. Is Tscale the
recognized leader in this niche? If so do they deserve that repuation?
Who are the worthy competitors? Will using this software universally help the
situation or are there qualifications ie. Only certain apps and under certain
conditions. I know TScale offers a free Trial which is the way to go as you can
test firsthand for improvements.
-----Original Message-----
From: Rick Mack <Rick.Mack@xxxxxxxxxxxxxx>
Date: Fri, 21 Jan 2005 18:11:26
To:thin@xxxxxxxxxxxxx
Subject: RE: [THIN] Re: Page file max out
Hi Steve,
There's a lot of evidence to show that under most scenarios, a hyperthreaded
CPU will give you a significant CPU performance boost, always provided your
system isn't getting too busy.
I'd like to quote part of a very good article on what hyperthreading is all
about "Introduction to Multithreading, Superthreading and Hyperthreading" by
Jon "Hannibal" Stokes.
Of particular interest is the following:
"Each of the Xeon's caches--the trace cache, L1, L2, and L3--is SMT-unaware,
and each treats all loads and stores the same regardless of which logical
processor issued the request. So none of the caches know the difference
between one logical processor and another, or between code from one thread
or another. This means that one executing thread can monopolize virtually
the entire cache if it wants to, and the cache, unlike the processor's
scheduling queue, has no way of forcing that thread to cooperate
intelligently with the other executing thread. The processor itself will
continue trying to run both threads, though, issuing fetches from each one.
This means that, in a worst-case scenario where the two running threads have
two completely different memory reference patterns (i.e. they're accessing
two completely different areas of memory and sharing no data at all) the
cache will begin thrashing as data for each thread is alternately swapped in
and out and bus and cache bandwidth are maxed out."
Basically, once the CPU starts cache thrashing, memory access time blow out
and you end up with a very slow system. Strtaight performance tests show
that hyperthreading (HT) is great until things get really busy, then an HT
system will collapse a lot faster than a non-HT system.
If it happens to be a back-end file/print server, the end result can be
catastrophic for a TS farm.
I don't use hyperthreading on file/print servers, regardless of O.S.
version. That response isn't totally logical, but my experiences with
hyperthreading problems have just been too painful.
But back to your original question about hyperthreading and win2k. Microsoft
genarally recommend you don't use hyperthreading on win2k.
However, provided you use some sort of CPU control (Armtech, Appsense,
TMUlimit, threadmaster etc) my experience is you can run with hyperthreading
on win2k system, safely and with significant benefit.
But mix heavy I/O, a virus checker and heavy apps an an HT system without
CPU resource control, regardless of the O.S. version and you're going to
have some interesting problems.
regards,
Rick
Ulrich Mack
Volante Systems
_____
From: thin-bounce@xxxxxxxxxxxxx on behalf of Steve Parr
Sent: Sat 22/01/2005 5:10 AM
To: 'thin@xxxxxxxxxxxxx'
Subject: [THIN] Re: Page file max out
Are there Citrix Admin's out there disabling hyperthreading if running Win2k
OS?
Have seen recommendations that it is better to disable for Win2k but for
Win2k3
hyperthreading generally will give a performance boost depending on load and
type of app.
Is there anyone out there who has tested\experinced Win2k OS for
hyperthreading vs. disabled
hyperthreading when running Citrix/TS? I just hear people saying that
hyperthreading "could" lead
to system stability, but perhaps there are people out there who have
switched from one mode to
another and seen noticeable improvements?
-----Original Message-----
From: Tim Mangan [mailto:tmangan@xxxxxxxxxxxx]
Sent: Friday, January 21, 2005 8:29 AM
To: thin@xxxxxxxxxxxxx
Subject: [THIN] Re: Page file max out
Pavlo,
I looked back at the thread, because I thought you mentioned Win2K servers,
but I couldn't find the reference. If so, Hyperthreading should be disabled
(Hyperthreading is OK on Win2k03 and WinXP). I have not seen it cause these
symptoms - but who knows? Hyperthreading is disabled in the BIOS.
Originally, you mentioned that "Page Swaps 100%". What does that mean? So
let's explore if that is really what is going on.
If I thought paging was an issue, I would first look to the Windows Task
Manager to see if the "Total Commit Charge" exceeds the "Physical Memory".
Constant paging will not occur until physical memory is exceeded. In the
numbers you show, it looks like (at least for that snapshot) you are below
this threshold. So I would use the performance monitor to watch these
values. (Note: you should run the Performance Monitor GUI from another
system and gather stats remotely). Next, I would look a the Performance
Manager "Memory:Pages/Sec". This counter includes paging of code/data pages
of applications, plus paging due to application file system cache. The
counter is in pages per second. So how many pages per second, when the
problem is evident and when it is not?
Tim Mangan
_____
From: thin-bounce@xxxxxxxxxxxxx [mailto:thin-bounce@xxxxxxxxxxxxx] On Behalf
Of Pavlo Ignatusha
Sent: Wednesday, January 19, 2005 9:17 AM
To: thin@xxxxxxxxxxxxx
Subject: [THIN] Re: Page file max out
OK,
I spent some time in the Perfmon and here is what I found while CMC reports
full load based on Page Swap:
1. Perfmon %DPC and %Interrupt for each processor are almost flat (1-8%)
2. Perfmon Process Page Faults/sec shows no single leader. A lot of apps
and windows services spike every so often but no one is claiming High page
Faults for longer then 2 sec. The most Page Faults were reported for
WinMgmt
some instances of Dynamics
some instances of Winlogon
some instances of Winword
some instances of guiengine (SQL client for patient information system)
some instances of Outlook
some instances of Accuterm97
spoolsv
some instances of CSRSS, wfcshell
naPrdMgr
3. Taskmon stats:
Totals
Handles 62313
Threads 3946
Processes 334
Physical Memory (K)
Total 2096604
Available 585365
System Cache 362476
Kernel Memory (K)
Total 211100
Paged 136364
NonPaged 74740
Commit Charge (K)
Total 1841980
Limit 7162272
Peak 2428548
Since I do not see a single source of the problem then maybe I'm running out
of resources? But interestingly I have 3rd server HP DL 380 with
Hyperthreading on and more memory but it is still hosting almost same number
of users and is getting maxed out on Page swap in CMC just as well. I'm
really puzzled.
Thanks for any help with this,
Pavlo Ignatusha
Systems Network Coordinator
Pembroke General Hospital
Tel. (613) 732-3675 ext.6150
Fax. (613) 732-9986
www.pemgenhos.org
"All that matters is love and work" - Sigmund Freud.
############################################################################
#########
This e-mail, including all attachments, may be confidential or privileged.
Confidentiality or privilege is not waived or lost because this e-mail has
been sent to you in error. If you are not the intended recipient any use,
disclosure or copying of this e-mail is prohibited. If you have received it
in error please notify the sender immediately by reply e-mail and destroy
all copies of this e-mail and any attachments. All liability for direct and
indirect loss arising from this e-mail and any attachments is hereby
disclaimed to the extent permitted by law.
############################################################################
#########
############################################################################
#########
This e-mail, including all attachments, may be confidential or privileged.
Confidentiality or privilege is not waived or lost because this e-mail has
been sent to you in error. If you are not the intended recipient any use,
disclosure or copying of this e-mail is prohibited. If you have received it
in error please notify the sender immediately by reply e-mail and destroy
all copies of this e-mail and any attachments. All liability for direct and
indirect loss arising from this e-mail and any attachments is hereby
disclaimed to the extent permitted by law.
############################################################################
#########
NYÃÂ*ÅÃIÃâÃ>VzÃË:j)jgEÃÂÃÃârâÂfÅvyÃjÃÃÂÃâzââÃÃÃÂÂzÃÂyÂÂÃNz|âjÂËÂÅÃ
yÂzÃâÂÂÃRÃÂTÅpâÃ
âÂÃâÂâÃyÂÅVâÃâÃÃÃÅyÂÃÅK~dÅuâ'ÅkÅÃÃÃâÃ
'ÂÃjâÅXâÃÂÃÃ
ÂÃ)râÂÂrzÃR{nÃâÂÂÂÂ&ÂÂÂ8zÃÂVjÃÅjzÃÂzZâ)âÃÃÃÅyÂÃÂÂâ+
********************************************************
This Weeks Sponsor SeamlessPlanet.com Domain Names
Register your .com domain name for as low as $7.85
One of the lowest prices on the web! Part of The Kenzig Group.
http://www.seamlessplanet.com
**********************************************************
Useful Thin Client Computing Links are available at:
http://thin.net/links.cfm
ThinWiki community - Awesome SBC Search Capabilities!
http://www.thinwiki.com
***********************************************************
For Archives, to Unsubscribe, Subscribe or
set Digest or Vacation mode use the below link:
http://thin.net/citrixlist.cfm
Other related posts:
