On Bringing a Sun to It Knees.

[jon at jim.UUCP]

With the talk about the performance of Sun's under load that has
been going on I thought it an opportune time to relate this.

With two images of the following silly little program our Sun3/110,
8meg core/SunOS3.2, stops!

int a[0x200000]; /* big enough to make it page without many images */

main()
{
  int c;

  while (1)
   c = a[ rand() & 0x1fffff];
}

 Suntools was the only other non-daemon type process running. Moving
the cursor out of the (only) window resulted in not being able
reselect the window and the only way out was the reset button. At one
point even the cursor stopped moving but this was with several other
processes trying to run, a few perf-meters etc.  which also stopped.

  Death did not seem to be instantaneous, it took a few minutes for
the machine to get completely bound up in paging.

Does Sun use some `clever' algorithm for paging which gets confused by
random accesses and starts paging out the things that it needs most????

Does the SCSI disk controller take lots of looking after???

I tried the same thing on a micro-vaxII/ultrix 2.0 and sure the
machine slowed down but it didn't stop. I could forground my emacs
(took 10-15 seconds) and continue editing, none of the other eight
users thought anything else but `machines running !$@%$# slow today'.
(No I didn't warn them, I knew it would work, trusting sole aren't I.)

jon


-- 
Jon Wells @ O'Dowd Research P/L.  Ph: 03-562-0100 Fax: 03-562-0616

Rogue isn't the only place that leprechauns steal your gold!