make_recovery backup takes longer and longer

[Richard Teasdale]

I run make_recovery (to tape) every Sunday on each of our HP-UX servers. On several of the machines, the backup takes longer every week. This is particularly noticeable on two machines which run IBM's MQSeries product. Both are at HP-UX 11.0. The installed version of Ignite-UX is B.2.4.307. The amount of data being backed up is not changing significantly from week to week. Does anyone have any insights into what's going on here? Over the course of several months, the time required for the make_recovery to complete has gone from about 1.5 hours to nearly 2.5. It is starting to be a problem.

Thanks!

[Bill McNAMARA_1]

you may be getting a lot of defragmentation on your filesystems..

this you can only check with fsadm (online jfs)

Later,
Bill

[John Bolene · Accepted solution]

Bill, defragmentation?

Maybe you meant fragmentation.

Questions.

Is Mqseries shut down for the backup?

Is there anything taking CPU or I/O during the backup?

Have you monitored the backup with glance or top while it is running?

What might be happening is a bunch of small log files or message files. Small files cause at least one directory lookup and then short I/O's to get the file.

Have you cleaned out /tmp and /var/tmp lately?

Do you have cron jobs running every few minutes?

Lots of things come to mind.

Let us know what you find out.

[Richard Teasdale]

It looks like the contents of /tmp is the culprit. I've got thousands of old socket files that appear to be added to all the time. Looks like MQSeries doesn't, or doesn't always, clean up after itself.

John, thanks for the help!

[fuji_ooh]

May you should try using a more up to date version of Ignite - B.3.6.82.
( I had problems with B.3.2.xx)
May be try to use the following syntax:
make_tape_recovery -a /dev/rmt/?mn -t "Your Commnets" -I -x inc_entire=vg00