topic kernel parameters for disk acces in Operating System - HP-UX

kernel parameters for disk acces

Jimmy_13 — Thu, 20 Mar 2003 00:23:32 GMT

What are the kernel parameters for the i/o operations?
I have raid and striping, but
I would like have more speed in the read and write operations.
Maybe the kernels parameters give more speed.

Re: kernel parameters for disk acces

Wilfred Chau_1 — Thu, 20 Mar 2003 00:33:51 GMT

You can check out the pvchange -t <# of second> option. Which set the timeout for IO timeout.

Re: kernel parameters for disk acces

Michael Tully — Thu, 20 Mar 2003 00:46:24 GMT

There is kernel change that can be used for fastser IO time. This is called vx_ninode. You might also need to decrease dbc_min_pct and dbc_max_pct.
Have a look here:
http://www.docs.hp.com/hpux/onlinedocs/B3929-90011/B3929-90011.html

The pvchange program does not speed up IO, but will wait until IO is comfirmed. The recommended value is 180 for most external disk array units.

It is possible to increase filesystem access time by utilising these parameters:
mincache, convosync and the type of logging. You need to have online JFS to implement them.

Re: kernel parameters for disk acces

James Murtagh — Thu, 20 Mar 2003 00:54:58 GMT

Hi Jimmy,

The (documented) tunable parameters for 11i are available here:

http://www.docs.hp.com/hpux/onlinedocs/TKP-90202/TKP-90202.html

I imagine you are talking more about filesystem and mount options? This is probably where you have more scope for increasing performance. The first point of call for the options would be in the release notes for the version of VxFS filesystem you are using.

Also on JFS 3.3 is the command vxtunefs, which has several new IO and configuration tunables. The man pages are also good, look at pages like mkfs_vxfs, newfs_vxfs and mount_vxfs.

You can also alter things like the scsi queue depth (using scsictl or kernel tunable scsi_max_qdepth) which can aid performance if you have PV links to devices.

Regards,

James.