Date: Mon, 25 Dec 1995 14:34:24 -0600 (CST) From: Joe Greco <jgreco@brasil.moneng.mei.com> To: bde@zeta.org.au (Bruce Evans) Cc: Mattias.Gronlund@sa.erisoft.se, bde@zeta.org.au, freebsd-hackers@FreeBSD.ORG Subject: Re: iostat and msps Message-ID: <199512252034.OAA04004@brasil.moneng.mei.com> In-Reply-To: <199512250133.MAA13380@godzilla.zeta.org.au> from "Bruce Evans" at Dec 25, 95 12:33:27 pm
next in thread | previous in thread | raw e-mail | index | archive | help
> >OK, if we can't calculate msps on new disk's, what is the reason for
> >it to still exist?
>
> They can be calculated accurately for some disks (for floppies and old
> uncached wd drives :-) and inaccurately for other cases.
Maybe the inaccurate count is better than nothing at all. Solaris does a
convincing attempt at gathering statistics on SCSI disks, and even if it is
not 100% accurate, it helps an administrator locate the most likely
bottleneck drive in a busy filesystem. It also helps when you want to go to
management and say "hey, the system statistics claim that the drive is 95%
saturated, and the performance tuning book says you have a problem if it's
over 30%" (Solaris has a NICE performance tuning book)....
Tools I don't have under FreeBSD but would like :-)
slowaris% iostat -x 30
extended disk statistics
disk r/s w/s Kr/s Kw/s wait actv svc_t %w %b
fd0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0 0
sd0 11.8 1.2 60.0 6.5 0.0 0.3 28.2 1 29
sd1 6.0 1.9 21.5 7.3 0.0 0.2 26.7 0 14
sd12 44.0 15.9 313.9 102.5 1.1 4.3 90.2 3 75
sd2 10.8 4.6 36.7 15.9 0.0 0.5 34.9 0 24
sd3 3.0 1.5 16.4 10.5 0.1 0.1 57.9 1 10
sd7 0.3 0.4 1.6 1.8 0.0 0.0 27.8 0 2
This is a fairly useful display - "wait" is the avg number of transactions
waiting for service, "actv" is the avg # of transactions actively being
serviced, svc_t is the average service time in ms, %w is the percent of time
that there are transactions waiting for service, and %b is the percent of
time the disk is busy (both derived from wait/actv).
It clearly shows that sd12 is in some pain....
> >> Lots of things would break even if DK_NDRIVE only increased. systat
> >> output wouldn't fit in 80 columns...
>
> >So, I have checked, the only system binaries that uses the dk_xxx
> >variables from /dev/kmem is systat and iostat and I guess that it
> >should be possible to fix them if they break :-)...
>
> I think Joe Greco fixed some things here.
I raised DK_NDRIVE (on my local system) and then fixed a bunch of
formatting type issues that arose from 4-character device names in
systat and iostat, etc.... might hold us for another half a decade
:-) I do NOT know whether or not these changes were merged into
-current as I don't have any boxes running -current currently.
There was one problem I didn't track down relating to the numbers mode of
systat -io and unusually sized xterms. My changes were sufficient for a
moderately sized system.
> >Is it possible that an dynamic implementation of thous statistics could
> >be intergrated into FreeBSD-current?
>
> Someone has to do the work. It would be nice if every driver didn't
> have to know how to manage the statistics. They could call (inline)
> functions but the placement of the calls is driver-dependent.
Always the problem... somebody's gotta do the work :-) :-)
Merry Christmas everyone,
... Joe
-------------------------------------------------------------------------------
Joe Greco - Systems Administrator jgreco@ns.sol.net
Solaria Public Access UNIX - Milwaukee, WI 414/342-4847
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199512252034.OAA04004>