home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!vtserf!black-ice.cc.vt.edu!valdis
- From: valdis@black-ice.cc.vt.edu (Valdis Kletnieks)
- Newsgroups: comp.unix.aix
- Subject: Re: AIX 3.2.3 on RS/6000-560 - strange scheduler behavior
- Message-ID: <10382@vtserf.cc.vt.edu>
- Date: 11 Dec 92 19:39:45 GMT
- References: <10325@vtserf.cc.vt.edu> <1gaavoINNjv0@life.ai.mit.edu>
- Sender: usenet@vtserf.cc.vt.edu
- Organization: Virginia Tech, Blacksburg, VA
- Lines: 44
-
- In article <1gaavoINNjv0@life.ai.mit.edu> mycroft@hal.gnu.ai.mit.edu (Charles Hannum) writes:
- >In article <10325@vtserf.cc.vt.edu> valdis@black-ice.cc.vt.edu (Valdis
- >Kletnieks) writes:
- >> F S UID PID PPID C PRI NI ADDR SZ RSS WCHAN TTY TIME CMD
- >> 201011 R 0 14097 13392 10 65 20 29474 30776 30772 pts/0 7:15 ./mem1
- >> #define SWTED 0x00000010 /* stopped while traced */
- >
- >If you *look in the man page for `ps'*, you'll find:
-
- Charles:
-
- Time to file a DOC APAR against something. I'm not sure what yet.
-
- My copy of the man pages identifies it as
-
- 10 Waiting for a page default, or forking
-
- Really confidence inspiring. Never heard it called a 'default' before...
-
- Meanwhile, on page A-62 of the 'Performance Monitoring and Tuning
- Guide' (SC23-2365-01) it gives *different* meanings for the 'flags'
- field, which just happen to correspond to the comments in
- /usr/include/sys/proc.h - note that the 'man ps' flags are only
- identifed up to x'100', and I've got a *few* more bits in the output.
- So in *any* case, *one* of the manualss is wrong (either 'Perf Mon' in
- that it isn't just a printf("%x",proc->p_flag); or 'man ps' is
- overlooking a lot of bits...
-
- It's interesting that the manuals agree (more or less) on the meanings
- of the x'01', x'02', x'20', x'40' flags, but disagree on the meanings
- of x'04', x'08', x'10', and x'100', and 'man ps' has no higher flags
- identified...
-
- In any case, it doesn't explain why of 20 identical processes,
- the first 12 are page faulting and making progress, and the other
- 8 are *not* page faulted, listed as 'R'unning by 'ps', but not
- making any progress at all.
-
- It also doesn't explain why when the load average gets to around
- 20, interactive response falls through the floor. I suspect the
- problem is that the scheduler is using a "oldest job first" scheme
- rather than "shortest job first" or "round robin"...
-
- /Valdis
-