Originally posted by: SystemAdmin

I learned to use filemon to monitor the %iowait of I/O. but what value constitutes a bad or good I/O? 10, 20 ms? besides, it is a constantly changing target on different lv.
#AIX-Forum

Originally posted by: SystemAdmin

I use the filemon and iostat -t to follow the I/O couple time today during the busy hours. Many times I saw the high iowait of write on my paging lv then the oracle data part of lv (understantable). here is one snapshot
VOLUME: /dev/hd6 description: paging
reads: 200 (0 errs)
read sizes (blks): avg 8.0 min 8 max 8 sdev 0.0
read times (msec): avg 13.671 min 1.926 max 576.000 sdev 53.084
read sequences: 200
read seq. lengths: avg 8.0 min 8 max 8 sdev 0.0
writes: 882 (0 errs)
write sizes (blks): avg 8.0 min 8 max 8 sdev 0.0
write times (msec): avg [b]297.835[/b] min 11.652 max 895.069 sdev 181.844
write sequences: 263
write seq. lengths: avg 26.8 min 8 max 320 sdev 35.2
seeks: 463 (42.8%)
seek dist (blks): init 9101360,
avg 2248940.6 min 8 max 11636984 sdev 2538701.9
time to next req(msec): avg 54.897 min 0.000 max 13596.242 sdev 684.450
throughput: 72.9 KB/sec
utilization: 0.03

VOLUME: /dev/fslv03 description: /dbbaan
reads: 14 (0 errs)
read sizes (blks): avg 14.3 min 8 max 16 sdev 3.3
read times (msec): avg 7.003 min 2.456 max 10.557 sdev 2.050
read sequences: 14
read seq. lengths: avg 14.3 min 8 max 16 sdev 3.3
writes: 81 (0 errs)
write sizes (blks): avg 28.4 min 16 max 48 sdev 7.1
write times (msec): avg [b]16.080[/b] min 3.284 max 36.662 sdev 6.033
write sequences: 81
Do I need to add more paging space or add a secondary dump dir? here it the sysdumpdev -e
0453-041 Estimated dump size in bytes: 248722227
the default is /dev/hd6 (by AIX 5.3L), but the /var/adm/ras/platform (383 M) has larger space. Should I sysdumpdev -D to that?
#AIX-Forum

Originally posted by: orphy

I assume that hd6 and fslv03 are on different hdisks? If not, there's one
of the problem. However, if they are on the same hdisks, the problem is
likely that you are short of physical memory. If I remember this correctly,
this is the box that you have a total of 4 hdisks and that you have had to
put some of the non-AIX LVs/FSs in rootvg?

It depends on the usage (lsps -a) of your paging space, you might want
to consider increasing the size of the paging space (chps -s) or add
another paging space (mklv -t paging) on a different hdisk that is busy.
However, adding more paging space is still not going to fix your high
I/O wait issue. Adding memory to the box will likely be the real help.

The secondary dump dev and the size of /var have nothing to do with
your iowait issue. You should read more about AIX dump in Infocenter
to unconfused yourself.
Orphy
#AIX-Forum

Originally posted by: SystemAdmin

You remembered correctly. the hd6 is a part of rootvg, including both hdisk0 and hdisk1, and fslv03 is part of applvg on the hdisk3. However, the Oracle engine is installed on the rootvg by contractor (no copy). Normally, from the monitor, the hd6 and fslv03 (oracle data part /dbbaan) are busy ones.
So if I want to ease the io bandwidth, it is better to add another paging for fslv03 on hdisk3? "add another paging space (mklv -t paging) on a different hdisk that is busy.", as the second best option to that of adding more memory.
On the other hand, our users have not complained anything yet during the busy hours, (or they just ignore the inconvience. all users are internal users, 35+).
#AIX-Forum

Originally posted by: orphy

Oops! No, do NOT add another paging space on hdisk3! I meant to say
"add another paging space...hdisk that is NOT busy." I simply typed too
fast and missed the "NOT" part. Most of the time, my fingers simply
couldn't follow what my brain says! Sorry about that!

If you are able to get the budget approved for more memory, that's likely
your "best" option. Having another paging space on the same hdisk as
/dbbaan is probably going to make the whole situation a lot worse! With
4 hdisks you have and the mirroring needed, you are pretty much limited at
tuning at this point.
Orphy
#AIX-Forum

Originally posted by: SystemAdmin

That was my suspicious, since it does not make sense if adding new paging to a location which is already busy with I/O. Thanks you for your timely correction.
Let me talk to my Mexican manager about the memory.
On the other hand, As Jims post means, If management is happy with the database performance and the application completes in it's require Service Level Agreement time, then there is no problem, although we have no SLA currently. My intention is to prevent thing getting worse.
#AIX-Forum

Originally posted by: SystemAdmin

As with any other utilization number a "bad" value for iowait is whatever value is responsible for unacceptable user performance or service. If your users are getting what they paid for then it doesn't matter what iowait or CPU busy or paging or any such numbers are. If your users aren't getting what they paid for then it's up to you to make a judgment call as to why that is. At this point it's more art that science, I hate to say. If you decide that I/O bandwidth is the problem then reducing iowait may solve your problem.

HTH

Jim Lane
#AIX-Forum