Allocating a space for /var/adm/crash

[Lukasz Geleta]

Dear All,
can anybody give me some guidelines for allocating space for /var/adm/crash on a system that has 2 GB of RAM. These dumps get compressed pretty well and I wonder if 1.5 GB will be enough.

Many thanks in advance.

Lukasz

[Stefan Farrelly · Accepted solution]

Take a look at this;

http://docs.hp.com/cgi-bin/fsearch/framedisplay?top=/hpux/onlinedocs/5187-2760/5187-2760_top.html&con=/hpux/onlinedocs/5187-2760/00/00/7-con.html&toc=/hpux/onlinedocs/5187-2760/00/00/7-toc.html&searchterms=crashdump&queryid=20030708-035317

In my opinion 1.5GB for a crashdump on a 2GB server is easily enough.

[twang]

In my option, that filesystem has to be at least equal to the size of your RAM.
So for your case you need a 2GB fileystem space. I suggest you to create the separate filesystem mounted as /var/adm/crash

[Bill Hassell]

For operating systems prior to 11.0, the amount of space available for a crash dump was the size of RAM although a truncated crash dump (as little as 30% of RAM) is usually sufficient to debug most kernel problems). At 11.0 and higher, the crash dump mechanism changed significantly due to the typical size of RAM (4Gb and larger). The new method will compress the crash dump files to about 25-30% of the size of RAM.

It is a good idea to create a separate lvol for the crash directory and make the lvol about 500 to 1000 megs. By making /var/adm/crash a separate mountpoint, /var will be protected from filling and taking down most applications.

[Geoff Wild]

I agree with Bill (suprise suprise) - make /var/adm/crash a seperate lvol - here's my standard for vg00:

VG00
The root volume group VG00 should be mirrored and only contain OS files.
Primary swap should be 1 x memory
/ - 140MB (HP Default)
/usr - 1 GB
/var - 1GB
/tmp - 512MB
/stand - 256MB
/var/adm/crash - 1 x memory + 512MB
/opt - 1GB
/home - 512MB with quotas (32MB soft, 64MB hard)
Note: /var/tmp should be a symbolic link to /tmp. /tmp should be 1777 (sticky bit).

Rgds...Geoff

[Caesar_3]

Hello!

Yes, it's good to made the crash as difrent lv
The size of crash is as the size of RAM.
You didn't wrote the OS verions that you use
so in the new realises the crash is compresed
and stripted from some data so became less size

Caesar

[jmb]

I pulled this old thread up because I'm doing more research into this. There seems to be a big difference of opinion on sizing the crash space.

Bill stated that 25-30% of RAM should be sufficient. Does that than require that dumps be specifically configured to be compressed? I've gone through the "Abnormal System Shutdown" doc from HP, and am not seeing the logic for this clearly stated.

This is for 11.11 systems, generally 54xx, 74xx sizes, and frequently with 8 GB, 16 GB or more of RAM.

[Bill Hassell]

99% of all crashes occur within the kernel so really only a few hundred megs are needed to troubleshoot the problem. Partial crash dumps are used all the time to determine root causes for a crash. The other gigabytes of RAM are full of applications and buffers and shared memory, etc, which are not normally needed to troubleshoot a problem. Naturally, with a 16Gbyte RAM system, waiting for a hour or two while RAM is written to the dump area, then another hour or two while the dump area is copied to a filesystem is not acceptable for most servers. The only way to speed this up is to reduce the size of the crash area. You can use crashconf to do some work in this area and typical crashes on large RAM systems will need as little as 25% of RAM for the filesystem crash area. A lot depends on the compressability of data in RAM.

[jmb]

Bill,

Thanks for the quick response. So is what you described pretty automatic, or will there need to be config changes to get the compression? Still trying to figure out how much the system simply defaults to.

"savecrash" contains savecrash=1, dir=/var/adm/crash and compress=2.

[Bill Hassell]

Those values will work fine. If the system ever crashes, you won't need much more than 25-35% RAM space on your crash filesystem.