home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.unix.aix
- Path: sparky!uunet!decwrl!csus.edu!netcom.com!dave
- From: dave@netcom.com (David A. Lee)
- Subject: What is wait time in sar and vmstat
- Message-ID: <50sn#la.dave@netcom.com>
- Date: Thu, 10 Sep 92 02:26:12 GMT
- Organization: Netcom - Online Communication Services (408 241-9760 guest)
- Lines: 30
-
- I am currently testing 2 versions of database software for our product
- and the results are somewhat confusing. After running a sample test
- for an hour with exactly the same environment with the 2 versions of the
- database, I look at vmstat and sar for the duration of the tests.
- The results are that the new version uses about 30% more user CPU time
- but 400% less wait time, resulting in a 100% increase of idle time.
- (idle from 25% to 50%, wait from 40% to 10%).
-
- The problem is that I cannot find a good description of exactly what
- wait time is. "Normal" unix systems have only user,system, and idle.
- Aix describes wait time (in man vmstat) as cpu cycles checking for
- IO to complete. Does anyone know exactly what this means? Are
- wait cycles time when the cpu isnt doing useful work? Why is the
- kernal polling for IO complete instead of putting the process to sleep
- and waking it up when the IO completes? If I have software (call it DB3.0)
- that uses say 25 Idle, 40 Wait, 25 User, 10 sys, and a
- software (call it DB3.1) that uses say 50 idle 10 wait 30 user 10 system,
- do I really have 25% more cpu time to spare for other processes? or
- is wait time also usefull time? (say wait+idle == Really Idle).
-
- If anyone really knows what is going on here, please, please let me know.
- We need to decide (very soon) if it is worth the extra user time used
- to give us more idle time.
-
- Thank you very much.
- Please email response to
- dave@netcom.com,
- or post to newsgroup.
-
-
-