Crash Dump Configuration

In the HP-UX operating system prior to release 11.0, the only configuration option was the ability to specify which devices would be used to store a crash dump. This option was set during kernel build and only took effect after rebooting the system on the new kernel

Now, by default, the system dumps to the swap partition. If the root disk is VxFS, the system dumps to a swap partition on the root disk; for example, /dev/dsk/cntndn. During runtime the default location can be changed to dump to another (secondary) disk, but the following limitations currently apply if you do that:

In addition to the kernel and logical volume device crash dump configurations, there are two other methods and some additional options.

Configuration of dump devices is modeled on configuration of swap devices Consider this comparison between them:

An important feature listed in the table above is the ability to read /etc/fstab and configure all devices marked "dump".

In addition to configuring dump devices, crashconf (both the system call and the command) can also be used to configure the classes of memory that will be included in, or excluded from, the dump. If desired, this information can also be configured at kernel build time, by setting the bitmask tunables alwaysdump and dontdump. Refer to the /usr/include/sys/crashconf.h file or the output of the crashconf(1M) command for the current list of memory classes.

In most circumstances, it should not be necessary to modify the list of individual memory classes to be dumped. The default settings provide adequate information for debugging nearly all system panics while eliminating as much extraneous data from the dump as possible.

In actual practice, if crash dump is reconfigured at all, it is usually done to force a full dump, or to disable dumps entirely.

Full dumps are enabled by including all pages in the dump. This is done with "crashconf -i all" or by setting CRASH_INCLUDED_PAGES="all" in /etc/rc.config.d/crashconf.Dumps are disabled by excluding all pages from the dump. Again, this is done with "crashconf -e all" or by setting CRASH_EXCLUDED_PAGES="all" in /etc/rc.config.d/crashconf.

Finally, crashconf(1M) uses the pstat interfaces pstat_getcrashinfo(2) and pstat_getcrashdev(2) to provide a report of which devices are configured, which memory classes will be included, and how much space a selective dump will take. This information can be used to determine the amount of dump space that should normally be configured. Specifying "-v" on the crashconf(1M) command line will display the current crash dump configuration to verify that the expected changes have taken effect.

How Much Dump Space Do I Need?

For selective dumps, use crashconf(1M) to find out how much space would be needed for a selective dump of your machine. (Use the -v flag, and run it while your system is under its normal or higher than normal workload.) The space needed will vary depending on the workload of the machine, so add another 25% or so to be safe. The total dump space should meet or exceed this amount.

For machines that are relatively stable and don't expect to dump often, this is enough. If a full dump is needed from such a machine, additional space can be configured for it on the fly, anyway. This space can also be used for swap unless reboot times are critical.

For machines on which full dumps are required, the full size of physical memory, plus a little bit for dump headers and tables, should be configured as dump space. At least the amount needed for a selective dump should be configured on a device that is not used for swap activity.

For example, the size of a selective dump for a 1GB RAM system typically ranges from 100MB to 200MB (10% to 20%). Extrapolating this to a 64GB system, a dump could be as large as 12GB.

Whenever you have dump devices that are not also used for swap activity, make sure that they are configured last. This will cause them to be used first (dump goes from the end backward), which will minimize the chance of writing into an area shared by swap. Writing into swap space is undesirable because it will slow down your reboot processing; see “Post-Reboot Dump Processing”, for details.