Intermittently terminal sessions on our system "freeze" for up to about 45
secs. That is, screens displays cease building or [Return] gets no
response. I have run out of ideas re:
(1) what problems to looks for
(2) how to trap the problem, anyway.
Any ideas and suggestions will be most appreciated.
---------------------------------------------------------------------------
System: AlphaStation 600A 255Mb 1 cpu
O/S V4.0C. Major use of system is supporting a university library
running the Dynix library management system. Dynix uses the
UniVerse operating system running on top of Unix.
Storage Advfs domain A 1 vol contains applications
domain B 2 vols contains data files
using RAID. Defragmented very recently.
NFS used for some /usr directories from central campus
computer services
Net connected to about 100 dumb terminals and 50 PCs in-Library,
and the campus and world generally.
Usage generally about 100 concurrent users. Main use is for staff and
student use of a library management system. Typically a user
spends most of the time reading screens; modest I/O, very little
data processing. Xload barely registers much of the time.
Activity is typically 1%-5%.
Problem sessions freeze intermittently, several times an hour. There is
no obvious pattern related to fluctuating user demand or known
activities. Freezing affects more than one active session, and
there is no obvious relation to physical location and therefore
network effects.
The problem started several months ago, after the system had been
installed for almost a year. Nothing substantial has changed
that seems to impact. Rebooting doesn't affect the occurrence or
duration of the freezes.
Tests Performance Manager is NOT installed. Programs have been written
to run the following at (generally) 3-sec intervals, to capture
the results and monitor values for extremes etc. Some of the
tests have been susceptible to the freezing, others haven't. All
tests have been run on terminals; some have been run in parallel
on the console, and the results have been similar on the console.
1. sleep followed by date display, run as a Unix script and as
a program within UniVerse. No difference in results. Test
not noticed to be affected by freezes
2. "top" run and top 10 users recorded. No obvious heavy users
detected. Test affected by freezes.
3. advfsstat run. No obvious problems detected. I/O generally
surprisingly light; occasional con values of 2k but no
correlation with freezes. Test affected by freezes.
4. netstat -i run. No problems detected. Traffic almost 100%
clean, very few collisions. Test affected by freezes, and
higher than normal packets read indicated afterwards, but
its ambiguous whether that reflects a build-up due to the
freeze or the freeze is due to a traffic hump.
5. swapon -s run. Free swap space never detected at <50%. Test
affected by freezes.
Other standard performance tests have been run. My conclusions are:
1. we've got adequate memory
2. our disks are in good condition and functioning well
3. network traffic is exceptionally clean and manageable
4. the problem cannot clearly be isolated to UniVerse operations
What should I test next? How can I determine what is actually going on
immediately before and during a freeze (this last bit has proved tricky,
because during a freeze no commands are accepted as input).
I repeat, any ideas or suggestions will be welcome.
Mike
Mike Shearer, Mike.Shearer_at_jcu.edu.au
Information Systems phone 61 07 4781 4482
Information Technology & Resources fax 61 07 4781 6691
James Cook University
Townsville QLD 4811 Australia
Received on Fri Mar 12 1999 - 04:19:41 NZDT