evaluation of sar -d

[Nikhil Mistry]

Hi All,
can anybody help me to solve this problem?

#sar -d 5 5

HP-UX hp-k260 B.10.20 D 9000/800 10/02/00

11:52:03 device %busy avque r+w/s blks/s avwait avserv
11:52:08 c2t0d0 1.40 0.50 4 70 5.28 4.91
11:52:13 c0t6d0 1.80 0.50 3 9 5.50 6.38
c2t0d0 0.20 0.50 1 19 8.12 0.97
11:52:18 c0t6d0 1.20 0.50 2 25 6.14 6.21
11:52:23 c0t6d0 1.80 0.59 2 30 6.34 39.08
c2t0d0 3.80 0.50 8 115 5.64 6.16
11:52:28 c0t6d0 5.60 1.12 7 102 11.96 25.57
c2t0d0 0.80 0.56 6 96 7.13 2.32

Average c2t0d0 1.24 0.52 4 60 6.19 4.33
Average c0t6d0 2.08 0.82 3 33 8.91 21.02
hp-k260:/ my disk # c2t0d0 has avwait 6.19
and avserv 4.33 so looks like bottleneck right there.how to solve this problem?

[John Palmer]

What makes you think that you have a disk bottleneck? The figures that you posted indicate that the disks are very lightly loaded.

The avwait and avserv time are in milliseconds so look pretty good to me - not sure how accurate sar is though.

The second column (%busy) is usually a better indicator of how active a disk is.

[Bill Hassell]

There is no simple answer to increasing performance except to add more disks on a separate channel and move the files and/or directories that are most heavily accessed to the new disks...basic performance steps.

The problem is how to determine which application(s) is hitting the disk and which files or directories are involved. It is quite easy for an ordinary user to slow a system down by starting a find / or a du / which on a small box is probably OK but not on a big server.

You can use the various sar reports (and top and vmstat) but they require a lot of training on how to interpret the results and which options provide the best information. Or you could install the trial version of Glance and use it's many screens to narrow down the busiest application(s) and which mountpoints are most affected.

For ongoing performance monitoring, Measureware is also available as a trial version on the Application CDROMs. You can use the extract command in Measureware to pull out needed data and graph the data or send it to a spreadsheet program.

[Bill Hassell]

There is no simple answer to increasing performance except to add more disks on a separate channel and move the files and/or directories that are most heavily accessed to the new disks...basic performance steps.

The problem is how to determine which application(s) is hitting the disk and which files or directories are involved. It is quite easy for an ordinary user to slow a system down by starting a find / or a du / which on a small box is probably OK but not on a big server.

You can use the various sar reports (and top and vmstat) but they require a lot of training on how to interpret the results and which options provide the best information. Or you could install the trial version of Glance and use it's many screens to narrow down the busiest application(s) and which mountpoints are most affected.

For ongoing performance monitoring, Measureware is also available as a trial version on the Application CDROMs. You can use the extract command in Measureware to pull out needed data and graph the data or send it to a spreadsheet program.

[Devbinder Singh Marway]

The basic rule with sar -d is if %busy is > 50% then there is a bottle neck and if avwait is greater than avserv there is a disk bottle neck. SO what you can try doing is see what processess are run on the disks and try sharing the load between them i.e. stripe the data whhich requires high access to it.

[Tan Shirley]

Hi,

From the %busy, looks like your hard disk are not really that busy. If you suspect that there is a bottleneck, try using the GlancePlus utility to check on the CPU and Memory and see if the bottleneck is there rather than the disks.