[THIN] Re: Interesting TechNet article
- From: "Mack, Rick" <RMack@xxxxxxxxxxxxxx>
- To: "'thin@xxxxxxxxxxxxx'" <thin@xxxxxxxxxxxxx>
- Date: Mon, 1 Jul 2002 08:02:39 +1000
Hi People,
We've had a couple of sites where some of the win2k/metaframe XP systems
were hanging. The resolution was totally unexpected.
We've been setting HKLM\System|currentControlSet\Control\Session
Management\Memory Management\DisablePagingExecutive to 1 for a long time.
Justification for doing this, (refer technote Q184419) seemed entirely
logical since keeping drivers and kernel components from being paged out
would both reduce paging and should increase kernel performance.
However, it looks like our system hang problems were due to setting
DisablePagingExecutive on. There is a hotfix for "fixing" the hangs,
Q323608, with a caveat from Microsoft:
"Please note that this hotfix may not resolve the issue. This depends as to
whether all of the driver code and system code can map into physical RAM.
Furthermore, with progression of time, and as drivers/services get added to
the machine, you may hit this issue again at a later time. I would advise
changing DisablePagingExecutive to 0 and allowing the pageable parts of
drivers and system code to be paged out when necessary instead forcing the
system to keep this code in RAM."
Kind of makes you wonder about on a 2 GB machine. Insufficient memory when
I've go 0.5 GB on average free? And I thought the garbage-collection
routines on win2k were better.
I guess the immediate message is don't set DisablePagingExecutive to 1 on
win2k.
--Flame On----
What we've seen with win2k is a number of issues resulting either from
Microsoft not applying fixes to the win2k codebase for problems appearing in
NT 4.0 TSE, or breaking things fixed in NT 4.0 TSE because of a perceived
need to return all win2k server code to a common baseline.
A couple of examples are Q292548 (compare to Q269214 NT 4.0 TSE) and win2k's
handling of MaxMpxCt/Maxworkitems where we've seen exactly the same issues
in win2k that were already fixed in NT 4.0 TSE, or a fairly naïve
modification of system performance to "fix" interactions with win9X at the
expense of terminal services scalability.
There are basic design flaws in the win2k kernel that seem to reflect either
a lack of understanding of the needs of scalable and stable multi-user
operating systems, or legacy code which addressed a resouce-starved
workstation scenario that really doesn't and shouldn't apply anymore.
I remember a unix system (MIPS-RISCOs) that a customer set up where the swap
partition overlapped the root partition. The system had 384 MB of RAM (1994)
and went for 3 weeks without paging before the root partition got clobbered.
Imagine an operating system that doesn't page unless it needs to, or where
the spooler can't crash the system.
I wish Microsoft could.
--Flame Off----
Regards,
Rick
Ulrich Mack
rmack@xxxxxxxxxxxxxx
Volante Solutions
18 Heussler Terrace, Milton 4064
Queensland Australia.
tel +61 7 3246 7777
**********************************************************************
This email may be confidential and/or privileged. Only the intended
recipient may access or use it. Any dissemination, distribution or
copying of this email is strictly prohibited. If you are not the
intended recipient please notify us immediately by return email and
then erase the email.
We use virus scanning software but exclude all liability for viruses
or similar in any attachment or message...,..,..,.
**********************************************************************
Other related posts:
