On Mon, Apr 27, 2009 at 8:31 AM, Renil Thomas <[email protected]> wrote:
> Hi Experts,
>
>
> _write
> value ------------- Distribution ------------- count
> 4096 | 0
> 8192 |@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@ 1161825
> 16384 | 2754
> 32768 |@ 42041
> 65536 | 1175
> 131072 | 649
> 262144 | 492
> 524288 | 143
> 1048576 | 50
> 2097152 | 31
> 4194304 | 54
> 8388608 | 275
> 16777216 | 27
> 33554432 | 23
> 67108864 | 5
> 134217728 | 0
>
>
> If you look at the statistics value for the second last interval, the value
> is 5 writes for 67ms. There are instances when the statistic is spoiled by
> just few writes like this.
> My question related to this is " can we isolate why this write is taking so
> much time?" or "can we isolate during which write this data was seen?"
Yes, you can get more information about the long writes. Use the
difference you're using to create this distribution (e.g., timestamp -
self->ts) in the predicate for a separate clause, like this (assuming
you're tracing the write(2) system call):
syscall::write:return
/ self->ts && timestamp - self->ts > 33554432 /
{
/* record data */
}
Chad
_______________________________________________
dtrace-discuss mailing list
[email protected]
