Re: [9fans] Re: suicidal kfs, again

Re: [9fans] Re: suicidal kfs, again

[Russ Cox]

Yes, fastrand uses truerand to seed itself, which
in turn uses /dev/random.  All the system load I saw
was genrandom trying to keep the random bit bucket
filled.

Russ

Re: [9fans] Re: suicidal kfs, again

[Andrey A Mirtchovski]

On Sun, 29 Apr 2001 forsyth@caldo.demon.co.uk wrote:

> has this happened since you installed the last two sets of updates (march and april)?
> they fixed quite a few things in mmu handling and paging
> of shared segments/pages
>

no, I took the machines down early april and haven't had any hardware to pul
p9 on since...

Re: [9fans] Re: suicidal kfs, again

[forsyth]

>>no, it is not only you. this was happening quite frequently on all my 4
>>machines running plan9 (cpu + kfs, no real fileserver) on two different
>>types of hardware. mine seemed to be triggered by heavy usage of acme, when
>>i would have several files/windows open and i would try doing any file
>>operation like saveing them...

has this happened since you installed the last two sets of updates (march and april)?
they fixed quite a few things in mmu handling and paging
of shared segments/pages

Re: [9fans] Re: suicidal kfs, again

[andrey mirtchovski]

On Mon, 30 Apr 2001, kazumi iwane wrote:

[snip]

> If this is only me, I am beginning to
> suspect there's something seriously wrong with
> my hardware setup.
>

no, it is not only you. this was happening quite frequently on all my 4
machines running plan9 (cpu + kfs, no real fileserver) on two different
types of hardware. mine seemed to be triggered by heavy usage of acme, when
i would have several files/windows open and i would try doing any file
operation like saveing them...

no swapping was occuring (lots of ram)...

depending on the situation there were two options:

if sitting on the console -- reboot
if connected to the machine via drawterm -- just reconnect. the server
didn't crash or needed a reboot.


on the other hand, the cpu server managed to keep 60+ days uptime before i
took it down (i used it only remotely :)

andrey

[9fans] Re: suicidal kfs, again

[kazumi iwane]

This time I tried

  while(){fortune>/dev/null&&cat /dev/swap}

in a rio window. kfs died in about 3 minutes,

  kfs 7: suicide: sys: trap: fault: read addr=0xcd pc=0x000043f1

No, it wasn't swapping; /dev/swap kept saying until
the last moment,

  1383/27511 memory 0/19350 swap

and ^t^tx (after the crash) reports almost the same
figure at the last line,

  1391/27511 memory 0/19350 swap 0 iolist

I have 128MB of RAM, with *kernelpercent=15 line
in my plan9.ini. It shouldn't swap.

What do you mean by "random-bound"? Does
fortune's use of fastrand(2) have something to do
with this? If this is only me, I am beginning to
suspect there's something seriously wrong with
my hardware setup.

- kazumi