Hi,
I'm a newbie to tool 'iostat' and I've read the
manual for iostat several times. But it doesn't help.
I still get confused with the output of 'iostat', the
manual seems too abstract, or high-level, for me.
Let's post the output first:
avg-cpu: %user %nice %sys %idle
5.70 0.00 3.15 91.15
Device: rrqm/s wrqm/s r/s w/s rsec/s wsec/s
rkB/s wkB/s avgrq-sz avgqu-sz await svctm
%util
/dev/sdb1 0.60 4.70 12.60 1.50 105.60 49.60
52.80 24.80 11.01 1.54 10.92 8.65
12.20
I'll ask about the rrqm/s, r/s, rsec/s, avgrq-sz,
avgqu-sz, await, svctm and %util in the above output.
First question: How many physical disk I/O read
requests are sent to hard drive by kernel driver? is
it the subtract of (r/s - rrqm/s), or just r/s? if it
is r/s, then it means user&sys applications send
(r/s+rrqm/s) read requests to kernel per second?
Second question: ( r/s * avgrq-sz ) is 30% bigger than
rsec/s, why? they should be equal or little difference
related to calculation omission.
Third Question: What's the UNIT of avgqu-sz, is it
NONE, or sector, or something else? If it is NONE,
then does it mean that the unit is 'read request'?
4th question: (await + svctm) is the time span for a
read request from being dispatched (by kernel driver)
to being served? If so, could we use this number as a
criteria for (disk + file_system) performance ?
5th question: %util is which percentage of CPU time?
it looks too abstract in manual, does it means (disk
I/O opertions time) divided by (%user + %nice + %sys)?
Or it is (%user + %nice + %sys) divided by all the
system time lots (%user + %nice + %sys +%idle)?
I got lost completely here, Please help.
Thanks a lot.
__________________________________________________
Do You Yahoo!?
Tired of spam? Yahoo! Mail has the best spam protection around
http://mail.yahoo.com
