Hpux 11.11 need to know ideal size of /var/adm/crash ?

[Sammy_2]

I have 24 gb of physical memory.
Do I need to have 24 gb of disk space in /var/adm/crash for hp to analyze dump ? How much ? I don't have that much space in root vg.

I understand 11.11 has selective dump.

I have configured 24 gb swap on lvol2 which is also dump device ?

[B. Hulst]

Hi,

24Gb should be enough.

Some OS versions can also compress the dump...

:-)

[HGN]

Hi

The size equal to your swap is good.

Rgds
HGN

[Jeff Schussele]

Hi Sam,

My rules of thunmb are:

Min => 1/4 RAM
Opt => 1/2 RAM
Max => RAM

HTH,
Jeff

P.S. One rarely needs = RAM because you don't ever need *every* byte to properly decode a dump.

[Gary L. Paveza, Jr.]

Run crashconf on your system. It will tell give you an estimate of how much space it will need.

[Sammy_2]

Does this mean I need 24 gb of disk space ? Do you need entire contents of memory for hp to analyze the problem. This is 11.11
Thanks

# crashconf -v

CLASS PAGES INCLUDED IN DUMP DESCRIPTION
-------- ---------- ---------------- -------------------------------------
UNUSED 5120157 no, by default unused pages
USERPG 14368 no, by default user process pages
BCACHE 628777 no, by default buffer cache pages
KCODE 2690 no, by default kernel code pages
USTACK 782 yes, by default user process stacks
FSDATA 317 yes, by default file system metadata
KDDATA 233865 yes, by default kernel dynamic data
KSDATA 289988 yes, by default kernel static data

Total pages on system: 6290944
Total pages included in dump: 524952

DEVICE OFFSET(kB) SIZE (kB) LOGICAL VOL. NAME
------------ ---------- ---------- ------------ -------------------------
31:0x021000 396128 25165824 64:0x000002 /dev/vg00/lvol2
----------
25165824

[A. Clay Stephenson]

The ideal size is 0 because your box never crashes.

The next to ideal size is still 0 because you have configured separate swap and dump areas and therefore there is no need to compress and save the dump image in /var/adm/crash before it gets overwritten by swap activity. Swap and dump really should have nothing to do with each other; moreover, swap should always be mirrored; dump should never be.

But since you have ignored Plan A, the dump image normally compresses fairly well and HP isn't going to want the whole thing anyway so typically 3-4 GB or so of free space in /var/adm/crash should suffice.

[Bill Hassell · Accepted solution]

As mentioned, you don't need all of RAM to analyze a crash dump, typically 1/4 of RAM is good for virtually every crash situation. But as little as 500 to 2000 megs will work for the majority of crash cases.

Now you don't have to map /var/adm/crash into the /var filesystem. Make /var/adm/crash a mountpoint and locate some unused space, perhaps 2-4Gb and make that the crash lvol. Note that there will be a non-useful message as the kernel boots up saying that the dump area is not big enough--you can safely ignore that message.

[Sammy_2]

Clay -Thanks. I will go with Plan A next time I build a box. You are saying to have a separate LV for swap and dump. Then I don't need crash directory because the contents of memory will be in dump and it will not get overwritten. How big the dump would be then ? Anyway, I will go with 3-6 gb for crash directory.
Bill H - Your suggestion is much like Clay and I will go with your choice. I was thinking along the lines of creating /var/adm/crash mount point and did not want to make it huge as it may (I hope) will never be used. Thanks a bunch.
Jeff- I was going to go 1/2 mem for crash until Clay made a suggestion.
Hulst and HGN - I don't think wasting 24gb space for crash is worth it although that seems to be popular choice.
Gary - I ran the command but don't know what to think of it.

Thank you so much.