Hi,
On Thu, 13 Jan 2000 13:38:38 -0800 (PST), Linus Torvalds
<torvalds@transmeta.com> said:
> Yes. But I'm getting worried about too many changes now. The patch from
> Stephen looked ok, although his new /proc interface wasn't, and I don't
> understand the point of having a separate program to parse /proc. If it
> isn't readable already, then why?
It's the same concept as vmstat. "vmstat 1" gives VM summaries every
second; "vmstat 10" gives it every 10 seconds.
The standard unix "sar" tool works the same way. To achieve the same
effect for a load-averaged statistic such as the IO queue utilisation or
the average queue length, you can't let the kernel munge the basic event
counts into a time-averaged value by itself: it doesn't know, on its
own, what interval the user is going to be time-averaging over.
By presenting the stats as raw event counts, user-land can take the
difference between _any_ two points in time and compute the average IO
behaviour between those two points, so even if one user is doing a "sard
1" while another is doing a "sard 100", the results will still be
correct for both.
If we just presented a time-average in /proc, as we do for the load
average, the sard program simply could not emulate the expected sar
behaviour. sard needs those event counts just as much as vmstat needs
event counts for cpu-mode cycle counts and interrupt counts. vmstat
could not work given only the load average in /proc.
--Stephen
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/
This archive was generated by hypermail 2b29 : Sat Jan 15 2000 - 21:00:25 EST