home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!usc!zaphod.mps.ohio-state.edu!moe.ksu.ksu.edu!deimos.cis.ksu.edu!mccall!infopiz!cmkrnl!jeh
- Newsgroups: vmsnet.internals
- Subject: Re: global vs. private pages on the FPL (was: VMS Tuning)
- Message-ID: <1992Aug25.122200.644@cmkrnl.com>
- From: jeh@cmkrnl.com
- Date: 25 Aug 92 12:21:59 PDT
- References: <06D94B115A9F822B58@MAX.U.WASHINGTON.EDU>
- Organization: Kernel Mode Consulting, San Diego, CA
- Lines: 49
-
- In article <06D94B115A9F822B58@MAX.U.WASHINGTON.EDU>,
- DEREK@U.WASHINGTON.EDU writes:
- > When a
- > process "gives" up a page, the memory is placed onto either the free page
- > list, or the modified page list. Process-private pages are placed at the
- > head of the free page list, while system-owned pages are placed at the
- > tail. (By "system-owned pages" I mean pages which are part of an installed
- > image, such as the DIRECTORY utility.)
-
- This is not correct. Unmodified pages that are removed from a working set by
- working set replacement are put at the end of the FPL, not the beginning, and
- it doesn't matter whether they were process-private or global pages. (Please
- don't call the latter "system-owned", that sounds as though you mean that
- they're part of the system working set. The system working set is handled
- just like a process working set as far as page replacement is concerned.)
-
- A global page is of course not put on the FPL at all unless it's been pushed
- out of the working sets of *all* of the processes that had referenced it.
-
- Modified pages go to the end of the modified page list. When they reach the
- top of the MPL and are written back to disk, they then go to the end of the
- FPL.
-
- Pages are put at the top of the free page list when they are released at
- image rundown, or, more generally, by deletion of the virtual address space
- that describes them. In the case of global pages, this happens when no more
- processes are mapped to the relevant global section. So, yes, what you call
- "system owned" pages *can* be released to the top of the FPL. It all depends
- on why the pages are being released -- working set replacement (to make room
- for a newly-faulted-in page when the working set is full) or deletion of the
- virtual address space.
-
- > When a process needs a new page,
- > it takes it from the head of the free page list. Thus, the system-owned
- > pages are more likely to "survive" if the free page list is relatively
- > large.
-
- Right idea, wrong detail. The pages in on the FPL that contain useful data are
- more likely to survive. It doesn't matter whether they were process-private or
- global pages.
-
- Lots of well-known "authorities" on VMS get these details wrong, repeatedly, in
- print. Professional courtesy prevents me from naming names.
-
- --- Jamie Hanrahan, Kernel Mode Consulting, San Diego CA
- uucp 'g' protocol guru and release coordinator, VMSnet (DECUS uucp) W.G., and
- Chair, VMS Programming and Internals Working Group, U.S. DECUS VAX Systems SIG
- Internet: jeh@cmkrnl.com, hanrahan@eisner.decus.org, or jeh@crash.cts.com
- Uucp: ...{crash,eisner,uunet}!cmkrnl!jeh
-