home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.sys.hp
- Path: sparky!uunet!elroy.jpl.nasa.gov!sdd.hp.com!col.hp.com!fc.hp.com!jk
- From: jk@fc.hp.com
- (John Kessenich)
- Subject: Re: HP 7xx problem with virtual memory (thrashing)
- Sender: news@fc.hp.com (news daemon)
- Message-ID: <C0E6xu.E77@fc.hp.com>
- Date: Tue, 5 Jan 1993 17:51:30 GMT
- Distribution: comp.sys.hp
- References: <1993Jan3.034944.3556@alchemy.chem.utoronto.ca>
- Organization: Hewlett-Packard Workstation Kernel, Ft. Collins, CO
- X-Newsreader: TIN [version 1.1.1 PL6]
- Lines: 54
-
- System Admin (Mike Peterson) (system@alchemy.chem.utoronto.ca) wrote:
- : In article <1992Dec31.193847.6419@csi.com> bkp@csi.com (Brian K. Pheiffer) writes:
- : >
- : >I have found a serious deficiency with the HP-UX memory management and
- : >was wondering if anyone else has seen this problem: whenever a process
- : >require more memory than the amount of physical RAM that the system has,
- : >the HP-UX operating system (8.05,8.07 & 9.0) start ``thrashing'', a
- : >term HP uses to mean that the disk continues to swap indefinately
- : >while the process makes no progress!
- : >HP mentioned that it solved this problem with HP-UX 9.0, however, we
- : >haven't seen that to be true. Does anyone else know about this
- : >problem?
- :
- : I reported this problem on 8.07 last June, and have tried several
- : patches, including the one for 8.07 that is supposedly based on the
- : 9.0 swap/paging algorithms. None of them have made any difference.
-
- The patches mentioned and 9.0 have certainly improved paging problems
- for some situations. Much of the effort for the patches and 9.0 was in
- keeping some interactive responsiveness of the system once it started
- "thrashing". Unfortunately, it is difficult to predict what situations
- will be improved.
-
- That the system thrashes means the amount of memory the CPU touches in a
- short amount of time is larger than the amount of physical memory
- available. The main variables are the amount of available memory (less
- means more thrashing), the speed of the CPU (faster means more
- thrashing), the size of the process (smaller means less thrashing),
- speed of paging devices (faster/more discs means less thrashing), and
- the access pattern of the process (more locality means less thrashing).
- These will vary considerably from system to system, and knowing which
- one changes when performance changes requires some work.
-
- System paging algorithms also play a role, but their impact is mainly
- the level of performance between the point where a system starts paging
- and where it starts thrashing. Changing the points where the system
- starts to thrash requires changing the variables above. You chose to
- change amount of available memory:
-
- : Our "solution", which seems to have worked so far, was to buy
- : enough 3-rd party memory (192MB), then only allow 1 task to run at a time,
- : with the user data size set to 128 MB max. This way one process
- : can not steal all the physical memory, though 2 or more could
- : gang up and do it.
-
- : Our system would die within a week with 128 MB memory.
-
- 9.0 is a considerably different system than 8.0*. If you have a major
- problem in this area with 9.0 which was reported against 8.0*, but you
- still see as unsolved, it might be best to report it in the context of
- 9.0.
-
- --------------
- John Kessenich
-