/var/adm/crash size

[Adam Garsha]

Does /var/adm/crash filesystem space need to allow full dump of memory? I am having a hard time keeping both swap and /var/adm/crash big enough on larger memory systems. Or does it only need enough space for system portion of memory.

If it does need to have freespace be as big as physical mem, does /var/adm/crash need to live on a vg00 filesystem? How about links? What about if I wanted to link it to a SAN filesystem vs. local storage (since I don't have space); I suppose then I'd have to be sure it gets mounted at a lower runlevel?

Maybe monkey with /etc/lvmrc?

Thanks for any ideas on how to get a big space if I need it but don't have drive slots on my 7640 to support bigger /var?

[smatador]

Hi,
Because, you could boot on san, you could also put /var/adm/crash on san. Why not effectively put on SAN if you don't have space.
Hope it helps

[Mel Burslan · Accepted solution]

to the best that I know, /var/adm/crash can be on any volume group. The dump device (which usually is the same thing as your primary swap) on the other hand is, what needs to be on the vg00 volume group. /var/adm/crash size, can be anywhere from 1/2 to 1/4 of the amount of physical memory you have, as the crash dump is getting written to this mount point in a gzipped form.

[smatador]

Look this thread
http://forums11.itrc.hp.com/service/forums/questionanswer.do?threadId=1251649

[Steven E. Protter]

Shalom,

The minimum size for dump is half of RAM. This is the size of a typical crash dump.

I would not put a crash dump on a SAN because a simple cable issue can crash the system and disable your dump device.

You may wish to use Ignite to migrate your system to a larger system disk to accommodate this need.

SEP

[Bill Hassell]

Actually, only about 25% of RAM is needed for the majority of crash dumps. And for kernel-related crashes, I've used 5%-15% crash dumps. There's a LOT of memory that is not usually involved in crash analysis. Things like shared memory and the buffer cache are seldom used for crash analysis. Look at crashconf for some additional controls.

As far as a disk, any disk, anywhere will work. The crash dump is just a simple collection of files created by savecrash which is run after local filesystems are mounted. In fact, you can disable saving crash dumps and run savecrash to a tape instead. If you send it to a SAN and the SAN goes down, you very likely have much bigger problems to solve than where /var/adm/crash is located. Note that all filesystems are mounted at the same level (actually, local filesystems including fibre/SAN systems are mounted at rc0). Don't mess with lvmrc unless you have have an Ignite backup handy.