home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.unix.aix
- Path: sparky!uunet!ornl!utkcs2!darwin.sura.net!spool.mu.edu!umn.edu!csus.edu!csusac!cdsac!scotte
- From: scotte@cdsac.uucp (L. Scott Emmons)
- Subject: Re: AIX 3.2.3 on RS/6000560 strange scheduler behavior
- Message-ID: <Dec11.163914.24147@cdsac.uucp>
- NntpPostingHost: cdsac
- Date: 11 Dec 1992 16:39:14 GMT
- References: <10325@vtserf.cc.vt.edu>
- Organization: CableData (U.S. Computer Services), Sacramento, CA
- Lines: 42
-
- In article <10325@vtserf.cc.vt.edu> valdis@blackice.cc.vt.edu (Valdis Kletnieks) writes:
- [...]
- >a) the x'10' flag in p_flag is defined in <sys/proc.h> as:
- [...]
-
- If you mean the 'C' field, that is (according to 'man ps') "Processor
- utilization for scheduling".
-
- >b) I have a "gut feeling" that part of the problem is that under heavy
- >load (load average over 20 or so), the scheduling algorithm breaks down
- >and chooses "inappropriate" processes to schedule next. Does anybody know
- >of a way to make the system "prefer" interactive processes under high load?
- >I looked at the 'schedtune' stuff but that isn't what is needed I think
- >a knob to turn to bias proc.p_pri more for interactive I/O hitting return
- >should make it get scheduled sooner...
-
- With a load average greater than 20, no amount of tuning is going to
- do a heck of a lot, especially with a 'fair' scheduling algorithm that
- guarantees execution to all processes (starvation prevention).
-
- AIX is shipped with the kernel tuned in a particular way (I haven't
- been able to find out how, but I bet it's based on TPC's). The way
- that it is tuned is not the optimal way for a highdegree of
- multiprocessing, nor for interactive use concurrent with I/O bound
- processing doing very large I/O operations.
-
- The "schedtune" (available in AIX3.2.1 and later (I believe)) program
- can help the first problem (multiprocessing).
-
- Using "I/O pacing" from "smit system" > "change/show characteristics"
- can help the later problem (interactive I/O).
-
- Unfortunately, I have not been able to find good guidelines on how to
- tune the system for various types of use. Anyone have any? If you do
- use the "Tuning Guide", make sure you have the latest version, as the
- earlier versions did not address many of the available tuning options.
-
-
- L. Scott Emmons
- CableData Research Center
- csusac.csus.edu!cdsac!scotte
- KC6NFP
-