Problem with XP scheduler?
in [Win32 Kernel]
|
From: Anne & Lynn Wheeler on 31 Dec 2009 13:54 re: http://www.garlic.com/~lynn/2009s.html#54 Problem with XP scheduler? http://www.garlic.com/~lynn/2009s.html#55 Problem with XP scheduler? http://www.garlic.com/~lynn/2009s.html#59 Problem with XP scheduler? the folklore in a.f.c. is that NT starts out as VMS by some people hired from DEC. VMS had specialized in some amount of commercial dataprocessing ... but didn't particularly have very long SMP support heritage. post with old email about vax/vms SMP product announcements http://www.garlic.com/~lynn/2007.html#email880324 http://www.garlic.com/~lynn/2007.html#email880329 in this post http://www.garlic.com/~lynn/2007.html#46 How many 36-bit Unix ports in the old days? then windows continues as desktop platform and NT becomes the (somewhat compatible) server platform. then when the two platforms were consolidated ... more was taken from the desktop platform than the server platform ... possibly implying that some amount of the SMP work was dropped(?). this possibly accounts for the stories in the press about intel having to explain to the CEO in redmond why single processor chips couldn't just continue to get faster ... and why there was the move to multi-core (multiprocessor) chips ... AND why windows (& desktop applications) would have to significantly improve its SMP support ... referenced in this article: http://www.theregister.co.uk/2007/05/01/mundie_mundie/ mentioned in this post http://www.garlic.com/~lynn/2007i.html#78 John W. Backus, 82, Fortran developer, dies note ... a.f.c. tends to have some amount of topic drift ... when I was at SJR, Backus' office was just a couple doors down the corridor. In any case, above post has references to old email regarding boca/os2 group asking me about "scheduling": http://www.garlic.com/~lynn/2007i.html#email871204 http://www.garlic.com/~lynn/2007i.html#email871204b in this post http://www.garlic.com/~lynn/2007i.html#60 John W. Backus, 82, Fortran developer, dies In any case, some of the decktop platform smp/multicore was that a lot of the desktop apps were strictly single-threaded and only ran faster when the processor got faster (additional processors didn't help, multiple processors help multiple different applications to run concurrently, but didn't help with running a single-threaded application faster) ... and for these apps to show increased thruput ... they would have to be rewritten for multi-thread (&/or parallel) operation. for other topic drift ... some past posts about working on design for 5-way SMP in the mid-70s (which got canceled before being announced): http://www.garlic.com/~lynn/submain.html#bounce which was almost immediately followed by working on design for 16-way SMP ... which also got canceled before being announced ... some recent references: http://www.garlic.com/~lynn/2009o.html#10 Microprocessors with Definable MIcrocode http://www.garlic.com/~lynn/2009o.html#14 Microprocessors with Definable MIcrocode http://www.garlic.com/~lynn/2009o.html#17 Broken hardware was Re: Broken Brancher http://www.garlic.com/~lynn/2009q.html#74 Now is time for banks to replace core system according to Accenture http://www.garlic.com/~lynn/2009s.html#32 Larrabee delayed: anyone know what's happening? and previous mention ... lots of past posts mentioning SMP (and/or compare&swap instruction) http://www.garlic.com/~lynn/subtopic.html#smp -- 40+yrs virtualization experience (since Jan68), online at home since Mar1970
From: Don Burn on 31 Dec 2009 14:00 "Anne & Lynn Wheeler" <lynn(a)garlic.com> wrote in message news:m3hbr76kjo.fsf(a)garlic.com... > > this possibly accounts for the stories in the press about intel having > to explain to the CEO in redmond why single processor chips couldn't > just continue to get faster ... and why there was the move to multi-core > (multiprocessor) chips ... AND why windows (& desktop applications) > would have to significantly improve its SMP support ... referenced in > this article: > http://www.theregister.co.uk/2007/05/01/mundie_mundie/ > It is ironic that Gelsinger complains about single processor chips, years before it was Intel that pushed Microsoft to release that disaster known as Windows ME because they were concerned about the lower end multi-processor systems taking business away from the high end Pentiums. I knew people from the Windows NT space that were discusted with the large amount of $$$ that Intel threw at the ME group to get it out and promoted. -- Don Burn (MVP, Windows DKD) Windows Filesystem and Driver Consulting Website: http://www.windrvr.com Blog: http://msmvps.com/blogs/WinDrvr Remove StopSpam to reply __________ Information from ESET NOD32 Antivirus, version of virus signature database 4732 (20091231) __________ The message was checked by ESET NOD32 Antivirus. http://www.eset.com
From: tanix on 31 Dec 2009 14:19 There is one more unpleasant thing about Windows scheduler. It looks totally synchronous with the mouse and keyboard events. From statistical standpoint, the chance is probably > 90% that a task switch will occure at exactly the same time when you click on something. It is like task switching is some kind of a synchronous event that only happens when you service the mouse click or keyboard events. What happens as a result, and especially if your box is uder a heavy load, is that when you click on something, what you think you clicked on is not actually what happened as far as windows interpreted it. But what happened is that other event from a different tast got processed. As a result things get totally screwed up in terms of them happening according to your expectation. Anybody knows the scoop on this one? -- Programmer's Goldmine collections: http://preciseinfo.org Tens of thousands of code examples and expert discussions on C++, MFC, VC, ATL, STL, templates, Java, Python, Javascript, organized by major topics of language, tools, methods, techniques.
From: Alexander Grigoriev on 31 Dec 2009 17:42 A task (actually thread) switch occurs when: 1. A time slice of a currently active thread expired _and_ there is another thread of equal current priority ready to run; or: 2. A thread of dynamic priority higher than the currently running is made ready to run. The important words here are: "ready to run" and "dynamic priority". An input message from mouse or keyboard will signal the input queue event for the target thread, making the thread ready to run. At the same time, KeSetEvent call will boost the dynamic priority of the target thread, likely elevating it over all other threads with the same static priority. Thus, the newly awaken thread will get the CPU. "tanix" <tanix(a)mongo.net> wrote in message news:hhitg3$i7f$2(a)news.eternal-september.org... > There is one more unpleasant thing about Windows scheduler. > > It looks totally synchronous with the mouse and keyboard events. > > From statistical standpoint, the chance is probably > 90% that a > task switch will occure at exactly the same time when you click > on something. > > It is like task switching is some kind of a synchronous event > that only happens when you service the mouse click or keyboard > events. > > What happens as a result, and especially if your box is uder a > heavy load, is that when you click on something, what you think > you clicked on is not actually what happened as far as windows > interpreted it. But what happened is that other event from a > different tast got processed. As a result things get totally > screwed up in terms of them happening according to your > expectation. > > Anybody knows the scoop on this one? > > -- > Programmer's Goldmine collections: > > http://preciseinfo.org > > Tens of thousands of code examples and expert discussions on > C++, MFC, VC, ATL, STL, templates, Java, Python, Javascript, > organized by major topics of language, tools, methods, techniques. >
From: Peter Flass on 31 Dec 2009 17:59
Anne & Lynn Wheeler wrote: > re: > http://www.garlic.com/~lynn/2009s.html#54 Problem with XP scheduler? > http://www.garlic.com/~lynn/2009s.html#55 Problem with XP scheduler? > http://www.garlic.com/~lynn/2009s.html#59 Problem with XP scheduler? > > the folklore in a.f.c. is that NT starts out as VMS by some people hired > from DEC. VMS had specialized in some amount of commercial > dataprocessing ... but didn't particularly have very long SMP support > heritage. post with old email about vax/vms SMP product announcements > http://www.garlic.com/~lynn/2007.html#email880324 > http://www.garlic.com/~lynn/2007.html#email880329 > in this post > http://www.garlic.com/~lynn/2007.html#46 How many 36-bit Unix ports in the old days? According to Pascal Zachary, and some memos on Bitsavers, it appears Cutler & Co. were working on a new OS for a new system (Prism/Mica) at DECWRL. When DEC, or should I say Digital, canceled that Cutler was happy to move to M$, where he could still develop a brand-new OS from scratch. |