home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!olivea!pagesat!spssig.spss.com!brent
- From: brent@spss.com (Brent Lambert)
- Newsgroups: comp.unix.aix
- Subject: Re: Virtual Memory curiosity
- Keywords: VM overflow response
- Message-ID: <1992Nov19.210627.11604@spss.com>
- Date: 19 Nov 92 21:06:27 GMT
- References: <bmarlowe.721422470@ka> <1992Nov12.131230.130467@jrh.uucp> <1992Nov18.193645.11129@msuinfo.cl.msu.edu>
- Sender: news@spss.com (Net News Admin)
- Distribution: usa
- Organization: SPSS, Inc. - portable code group
- Lines: 70
-
- In article <1992Nov18.193645.11129@msuinfo.cl.msu.edu>, boone@convex.cl.msu.edu (Dennis Boone) writes:
- > In article <1992Nov12.131230.130467@jrh.uucp> jrh@jrh.uucp (James R. Hamilton) writes:
- > >In article <bmarlowe.721422470@ka> bmarlowe@ka.reg.uci.edu (Brett Marlowe) writes:
- > >>I've heard rumor that AIX will pick processes to kill off to avoid
- > >>running out of virtual memory? Can anyone verify this and if this is
- > >>true a) is there a way to disable this "feature" and b) what idiot
- > >>decided that randomly killing processes was a good idea??!!
- > >>
- > > It is true that AIX will start killing process when it runs out of
- > > page space. You can disable this feature (i.e. protect a process)
- > > by it calling "signal(SIGDANGER, SIG_IGN)". WRT to the idiot that
- > > decided to randomly kill processes, the decision is not random --
- > > processes are killed youngest first (in AIX 3.2) and if they ignore
- > > SIGDANGER then they are not killed at all.
- >
- > I tend to agree with Brett. The solution is completely non-portable and
- > non-standard, so out-of-the-box software has to be modified. You can't
- > run a stable system if the o/s is going to run around whacking off
- > important processes. The fact that the decision isn't really random makes
- > no difference: it still isn't possible to tell which processes will get
- > the axe without keeping continuous track of every new process.
-
- I've previously posted suggestions for IBM to improve memory allocation to
- make our lives less "fun", so clearly I'm not happy about the current
- implementation. On the other hand, I really don't think that the problem
- is as significant as it's being made out to be.
-
- For one thing, is there such a thing as portable "out-of-the-box software",
- where a single executable file runs on both AIX and non-AIX machines? I
- think not. Therefore, "out-of-the-box software" is either coded for, or
- ported to, AIX. Part of porting software includes dealing with platform-
- specific issues. If your favorite application doesn't work well on your
- platform of choice, it's because the application vendor failed to code it
- or port it properly.
-
- Being an application developer, I don't like to deal with platform-specific
- issues like this, but if my program crashes because of AIX's memory
- allocation strategy, it's my fault, not IBM's. And since most systems have
- problems in the portability and standard area, I just expect to deal with
- them (sigh).
-
- As for whacking off (important processes), if a process is important, it
- should be doing a lot of things to look after itself, including catching
- SIGDANGER. One might even say that a program which doesn't catch SIGDANGER
- and then try to free up some paging space is not very 'neighborly'. Of
- course, if you don't have access to the code, and it's designed to be an
- "important process", then refer to my two previous paragraphs.
-
- In the same vein, if you have consistent problems with processes getting
- killed due to a shortage of page space, that's probably a good sign that
- you need more paging space.
-
- As far as knowing "which processes will get the axe" is concerned, it's
- pretty predictable, something like the youngest process that doesn't catch
- SIGDANGER goes next. There's more to it than that, but the algorithm has
- been discussed here before, so you can look it up.
-
- Finally, as others have pointed out,
- Ignoring danger could get you killed;
- Noticing danger could keep you alive;
- Doing something about the danger is your best bet.
-
- So catch SIGDANGER, then do as you please.
-
-
- --
- The above statements are not the opinions or policies of SPSS Inc.
- The above statements may not be the opinions of Brent Lambert.
- The first disclaimer is a policy of SPSS Inc.
- Subsequent disclaimers are probably the opinion of Brent Lambert.
-