*Extremely* slow startup for trivial programs under SunOS4. Help?
montanaro at sprite.crd.ge.com
montanaro at sprite.crd.ge.com
Sat Jul 29 16:44:13 AEST 1989
The small table below demonstrates an extremely annoying problem I'm
having on my Sun (3/260HM, diskless, 8MB memory, booted from a
multiprocessor Encore Multimax, SunOS 4.0.1). The wall clock time the
first time I execute simple commands (in this case, from(1)) are
astronomical, compared with both SunOS3.5 and with successive executions.
Timing of from(1), with no mail waiting
Machine First time Asymptotic
-------------------------------------------------------------------
Sun-3/260HM real 0m4.03s real 0m0.70s
diskless, 8 MB user 0m0.01s user 0m0.06s
SunOS 4.0.1 sys 0m0.50s sys 0m0.16s
Sun-3/260C real 0m0.80s real 0m0.31s
diskless, 24MB user 0m0.00s user 0m0.01s
SunOS 3.5 sys 0m0.08s sys 0m0.16s
(Asymptotic means I executed "time from" several times in rapid
succession, and recorded the fastest time of the bunch.)
Why does it take so long for from(1) to execute the first time (4+ seconds
of wall clock time) on the 4.0 system? The only fundamental difference I
can see between the two systems is that under 4.0 I'm dynamic linking.
Can that possibly take that long? Both systems had other programs active,
but were pretty much quiescent. Neither system appeared to be paging
(eliminating physical memory shortfall as the problem), besides, programs
like from(1) are small enough that they shouldn't induce much paging.
(This was not a highly controlled test, but this behavior is quite
repeatable. It's most notable, of course, for programs that started
quickly under SunOS3.5, like pwd, ls, vi, df, mount, and so on.)
Another place I see an annoying delay is when logging in, between the time
I enter my password and the time the message-of-the-day is displayed. This
seems to run between two and five seconds.
I don't think the automounter is at fault, since /usr is mounted
statically during boot.
Can anyone explain this phenonmenon? Is it fixed in a later release? Is
there a workaround?
Thanks for any help,
Skip Montanaro (montanaro at sprite.crd.ge.com)
P.S. BTW, the kernel is about as barebones as it can be made.
More information about the Comp.sys.sun
mailing list