On 9/29/06, [EMAIL PROTECTED] <[EMAIL PROTECTED]> wrote:
My past installations have used 4 same-sized v-disk areas with ascending priorities. I was under the impression that Linux references an entire v-disk when it goes to swap, so smaller v-disks would be more efficient. But I'm not sure that's completely true. I want to simplify the fstab configuration and reduce the potential memory footprint, so I'm thinking of 1 moderate size v-disk, backed by 1 larger lower-priority DASD. The pair can be sized to whatever the applications require.
You're correct that it *does* make sense to have multiple vdisks for swap. The reason for that is the fact that Linux prefers to take fresh slots on the swap disk rather than re-use old slots that became free. Unfortunately VM does not know that Linux does not need those freed slots so it will go and write the pages to paging disk. That will eventually make your swap device slow. If you have vdisks for swap with different priority, Linux is forced to re-use the top one before using something else. The frequent references to that top disk also prevent that from being paged out when possible. The optimal size for those disks depends on the workload, but there's the rule of diminishing damage. Because the 2nd disk does not get used that much, the impact of slightly worse behavior is less as well. I think I once suggested to start with half the virtual machine size and double the size at each layer. That rule of thumb breaks with virtual machines > 4GB :-) Your performance monitor will show you various things of the virtual disk. Like virtual I/O rate, resident pages, and paged out pages and paging rate. You want the top disk to be large enough that it still is used for swapping by Linux (and not sits there with swapped out stuff that is never referenced) and small enough that it does not get paged a lot by VM. If your memory requirements in Linux are very dynamic, you want the same analysis for the 2nd layer. Swapping to disk is only good for one thing, and that is to slow down the Linux server. While z/VM 5.2 does not slow it down as good as z/VM 5.1, swapping to disk is still pretty slow. If you don't mean to slow down Linux but only allocate swap space on disk "just in case" and not swap at it, why not give that to CP for paging device and give the virtual machine another big virtual disk. If you don't use the virtual disk, it's almost for free. When you do need to use it, it may be pretty quick if CP has real resources for you. I know someone who runs virtual machines with 16GB of swap space on vdisk each. You would need to monitor the behaviour of the system and set your tuning parameters right to prevent damage, but it can be very effective. Rob -- Rob van der Heij Velocity Software, Inc http://velocitysoftware.com/ ---------------------------------------------------------------------- For LINUX-390 subscribe / signoff / archive access instructions, send email to [EMAIL PROTECTED] with the message: INFO LINUX-390 or visit http://www.marist.edu/htbin/wlvindex?LINUX-390
