How much space for /var/adm/crash

[Olivier Masse]

Is there a rule of thumb to determine from crashconf how much space I need in /var/adm/crash to hold a standard selective dump with 11.31. The size is in pages, and I guess a page is 512kb, but I'm not sure.

I'm used to create a dedicate, 4Gb /var/adm/crash but the system administrator's guide doesn't give specifics and there's no way I'm allocating 32Gb to it!

I'm not looking into fancy solutions such as dedicated parallel dump devices, as conserving disk space is more important to me than the time needed to reboot. But if you have any suggestions, I'll be glad to hear them.

Thanks

[sujit kumar singh]

hi


Well i shall keep my /var/adm/crash to be able to hold a full kernel dump in case i should facs that one day.

But due to space limitations even i shall make that always big so as to hold a Selective dump completely.

So that in case the system is critical and sometime say if there are RCAs and serious issues for the system , well i think a good and sound dump shallhelp me actually what was in the Black-Box of my system then.
regards
sujit

[sujit kumar singh]

hi
Well i shall keep my /var/adm/crash to be able to hold a full kernel dump in case i should facs that one day.

But due to space limitations even i shall make that always big so as to hold a Selective dump completely.

So that in case the system is critical and sometime say if there are RCAs reqd and serious issues for the system , well i think a good and sound dump shall help me know actually what was in the Black-Box of my system then.
Dumps and crashes are not a regular event but once it happens and that system is critical , a good dump analysis helps that actual reason is pin-pointed and that we make the system ( well that depends how critical is the system) more prepared and sound.
regards
sujit

[Tingli]

#crashconf -v
to find out the size of selective dump, then plus 25%.

[Tim Nelson]

As a side note:

You would only need this space in the event you would like to review a core dump.

During boot the core dump if any is kept on your configured dump space. Then converted and saved in /var/adm/crash.

I only configure /var/adm/crash if I have a core I want to look at. Then add the space, run savecrash. Do the Q4. Send to HP. Remove the /var/adm/crash when done.

It is simply a preference.

Crash dumps now a days do not dump the unused memory areas so extra space is not needed.

To be safe, configure = to physical memory. If you system always runs at 100% the tack on a little more ( 1.5 x ).

And with that, if your dump area is not configured for at least = phys ram then you will never get a complete dump anyway.

[Olivier Masse]

I know about running crashconf -v but my question is specifically: how much data is a page? Is it 1Kb? 512bytes?

Furthermore I had a crash yesterday and it was much bigger than anticipated. The console logs show that it tried to dump 5Gb of data, while crashconf says that it should take 1227155 pages. That's why I'm mystified.

[Olivier Masse]

Just to clarify, I run a swap+dump volume, along with a /var/adm/crash, strictly to save disk space.

For example, one of my servers has 144Gb. I don't want to allocate a lvol just to hold the dump. That's why I prefer doing a selective dump on the swap volume, then have savecrash copy it to a (I hope big enough) /var/adm/crash that has reserved space. Might be oldschool but I like it that way.

Crashes don't happen often and my production is under ServiceGuard, so saving disk space takes precedence over server boot-up time for me.

Thanks

[Don Morris_1 · Accepted solution]

I'm not going to speak as to sizing /var/adm/crash (being a very atypical consumer of crash dumps as a developer).

But as to your recurring question:

getconf _SC_PAGE_SIZE

Usually, 4096 bytes [4kb]. On later versions of v3 (I think 11.31.0803?), the base_pagesize tunable could result in this being larger.

[Olivier Masse]

getconf _SC_PAGE_SIZE

Seems to do the trick!

Thanks to everyone.

[Olivier Masse]

closed