iostat confusion??

From: Eric (etailor_at_jpl.nasa.gov)
Date: 10/18/04 

Date: Mon, 18 Oct 2004 10:19:21 -0700

I am trying to analyse a performance problem. I began looking
at the iostat program but I cannot understand the results.

My test case is to have a program that simply writes a 1 gig file.
It just does a loop of write calls. Here is the iostat line that results.
(running on EL 3).

Device: rrqm/s wrqm/s r/s w/s rsec/s wsec/s
/dev/hda2 0.00 24645.22 0.00 1779.62 0.00 211449.68

rkB/s wkB/s avgrq-sz avgqu-sz await svctm %util
0.00 105724.84 118.82 15799.36 890.26 1.79 318.47

My questions:

1) How can %util be greater than 100% (I have a dual processor, so I could conceive
of a value as high as 200% to mean both processors). I don't understand how this is
computed.

2) I don't understand the avgqu-sz value. The man page says:

 avgqu-sz
                     The average queue length of the requests that were issued
                     to the device.

What is meant by the queue length? Length of what. Surely it cannot be
the number of device requests (15799) pending? What queue is this
referring to?

3) Does iostat measure actual disk activity with or without the disk cache?
If all requests on, say a read, can be satisified by the disk cache, what will
iostat report. A value of 105724, or 105megabytes/sec seems a bit high to me
for an average 120gig ide drive, 7200rpm 8meg buffer. I can see possibly
33 or even 66 meg/sec, but not 105. How is this value computed?

thanks
eric