Reconfiguring kernel

[Ninad_1]

Hi,

I want to change the following kernel parameters on a HP 11.00 box.

Param Existing New

maxswapchunks 4197 16384
maxusers 64 128
shmmax 2147483648 3221225472
swapmem_on 0 1

After changing the parameters a kernel reconfiguration will be required and hence a reboot will be required. I just want to know what all files I need to backup so that incase the server does not boot I can revert to previous configuration and where I need to copy the files - like in /stand or anywhere else.

Thnks

Nad

[RAC_1]

Ninad,

Do not backup anything but /stand/vmunix, /stand/system and /stand/dlkm.
This should be standard practice during kernel reboot. So if anything goes wrong, you can boot from old kernel.

Anil

[Pete Randall]

You can also use SAM to make these changes and SAM will make the appropriate backups for you.


Pete

[Ninad_1]

Sorry , one more query - I have 12 GB memory on the server and swap space configured is 8 GB , thats why I am making the swapmem_on as 1 and maxswapchunks = 16384 , I just want to know that if the swap is not getting used do I need to add additional swap space to make it equal to the physical memory ?

swapinfo -tam shows all swap as reserved and nothing used :
Mb Mb Mb PCT START/ Mb
TYPE AVAIL USED FREE USED LIMIT RESERVE PRI NAME
dev 4196 0 4196 0% 0 - 1 /dev/vg00/swap
dev 4196 0 4196 0% 0 - 1 /dev/vg00/swap2
reserve - 7847 -7847
total 8392 7847 545 94% - 0 -


Regds,

Nad

[A. Clay Stephenson]

As long as the swap is not being used then there is no need to increase swap space; that would simply result in even more disk that you aren't using; however, eventhough it's not actually being used, it is a very good idea to mirror every dab of swapspace that you have so that you won't need to shutdown to replace a failed disk. I have many servers with much more physical memory than swap space.

[RAC_1]

When you set swapmem_on to 1, what happens is swap space (only for kernel calculation)
is calculated as 75 % of RAM + swap space configured.

This suit more to high RAM systems. typical systems having 16 GB of RAM and above.

As do not swap, there is no need to add swap space. However you need to think about dump space. If you use swap spave as dump space, you should put some additional swap space so as to get full crash dump in case system crashes.

Anil

[Ninad_1]

Stephenson
Thanks a lot for a very good practical suggestion of mirroring the swap.

RAC,
Yes you have pointed out a very right thing , but just I wanted to know - is there anything like full crash dump is required or even the partial dump of the memory on the smaller swap device is useful and acceptable ? Please can you answer this query as well, as I want to know that in case of a crash will the crash dump whatever is available on the swap is acceptable by HP to analyze or it is not at all acceptable and only a full dump is acceptable ?

Thnks,
Nad

[A. Clay Stephenson]

I am a very firm believer that swap and dump should have absolutely nothing to do with each other; they serve entirely different purposes and on large systems, having to compress and save the image during bootup (so that swap can be used for its intended purpose) can add significant delays to the reboot process. As I said before, always mirror swap but never mirror dump. As a practical matter, very, very seldom will HP want a full dump image because the things of interest will be in the lower addresses. A good rule of thumb is that you should have about 25% of RAM configured as dump space on systems with your amount of memory. Note: You can have multiple dump devices so everything doesn't have to be on the boot disk.

[RAC_1]

You can check what the dump size of the crash with crashinfo -v command. HP would say to reserve at leat about 70 % area for crach dump. Most of the admins use swap space as crash dump space. As Clay pointed you can have a very small swap space. But this would be OK, if you have seperate enough dump space.

Anil

[Ninad_1]

Thanks for the clarifications. Regarding dump I read that from HP 11.00 onwards dump device can be added even while the system is running and not reboot is necessary. Please can you tell me if this is correct and if I need to do any configuration for dump device - ( currently I have a separate dump device but is very small in size ) which may require a reboot. I am planning to make the changes in the kernel config as mentioned earlier and hence would like to know if I can add the additional dump device later without the necessity of reboot and if anything needs to be configured in order to be able to do this while system is online. ( like I am keeping the maxswapchunks to max value so that if required in future I can add swap online. ) - Please advice and correct me if I am wrong.

Thanks again - hope I am not too greedy always asking more and more doubts.

Nad

[A. Clay Stephenson]

You can run the lvlnboot command to define new dump areas; however, the actual changes do not take place until the system is booted. Note that you can specify multiple dump devices but each must be on a separate lvlnboot command.

[Ninad_1]

I guess that lvlnboot is required for configuring the primary dump device , what I am asking is if I already have a primary dump device configred , can I add more dump devices to increase the total dump size using the crashconf command while the system is still running and do these changes come into effect immediately without requiring a reboot. ( only for the additional dump devices )
Please advice.

Thanks

[A. Clay Stephenson]

The crashconf command changes take place immediately; however, these changes are lost with the next boot. You can create dump entries in /etc/fstab that crashconf will read upon the next boot (if you have configured /etc/rc.config.d/crashconf correctly) or choose to use lvlnboot. In any event, it's far safer to go through a boot cycle to make sure that you have the dump areas properly configured; otherwise you are just begging for a booboo when you need it the least.