General
cancel
Showing results for 
Search instead for 
Did you mean: 

memory pages high swap rate and very low write cache hit ratio

E.Gopi Kannan
Occasional Advisor

memory pages high swap rate and very low write cache hit ratio

Hi All,

Daily I'm getting incdients "memory pages high swap rate and very low write cache hit ratio" from my HPUX server.

Please let me how to overcome this issue?
1 REPLY
Matti_Kurkela
Honored Contributor

Re: memory pages high swap rate and very low write cache hit ratio

This is actually two semi-separate issues.

1.) "memory pages high swap rate"

Is it a high page-in rate or a high page-out rate?

High page-out rate indicates your system does not have enough RAM, so it must move data extensively in and out of swap.

If only page-in rate is high but page-out rate is low, you may have an application that is loading a lot of data from the disk. A high page-in rate can be normal in that situation.

2.) "very low write cache hit ratio"

Either your disk cache is too small to be effective for your workload (perhaps because your system needs RAM for more important purposes than caching?), or you have an application that is processing a lot of data in a sequential fashion.

If the application loads a data item, does something to it, then writes it to disk and then moves on to the next data item (i.e. it never re-reads or re-writes the data items it has already processed once), then extensive write caching is not very beneficial for the application. If an application processes a lot of data in this way, a low write cache hit ratio is expected and normal.

The write cache is more useful when the same data items are changed again and again frequently.


Without knowing the users' opinion of the current system performance and/or without having more information about the application(s) on the system, it is impossible to decide an appropriate action to overcome this "issue". See the HP-UX Performance Cookbook:

http://h21007.www2.hp.com/portal/download/files/unprot/devresource/docs/techpapers/uxperfcookbook.pdf

On the 2nd page, it includes some rules of thumb:

- Don't fix what is not broken. If your users are happy with the current performance level, no major actions are needed. However, if your system's workload is increasing, your "issue" might be an early warning that you will need to increase the system's capacity (more RAM and/or more/faster disks) in the future.

- If you don't understand your system, you're likely to make bad decisions (buying extra hardware that is not really required, or making things worse by tuning the wrong things, for example)

MK
MK