High %SWPWT vs. almost zero SWR/s & SWW/s?

[skayser]

Hi,

according to esxtop our VMs are heavily bound by Swap I/O ('c'-> %SWPWT close to 100%). However, SWR/s and SWW/s (memory view, 'm') are almost always zero. Has anyone observed such a behaviour? Do I need to consider another indicator to explain high %SWPWT?

Sebastian

[Halesh]

>> according to esxtop our VMs are heavily bound by Swap I/O ('c'-> %SWPWT close to 100%).

This is because you might be running in memory over-commit situation, so you are observing swap target getting set.

>> However, SWR/s and SWW/s (memory view, 'm') are almost always zero. Has

anyone observed such a behavior? Do I need to consider another

indicator to explain high %SWPWT?

Even swap target is set this doesn't mean that we have to swap. Before swapping memory ESX will reclaim memory via transparent page sharing or balloon and also via compression in ESX 4.1. So you might not see swap happening even when it is set. All swap target is getting transfer to other memory reclaiming techniques. This improve the performance by avoiding disk i/o for swap.

Thanks.

[JeffInouye]

I know this is an old post, but I did have the same issue that was resolved by updating/installing vmwaretools.  This became apparent when the host mem overcommit got a bit steep, but continue even after TPS kicked in a further optimized the guest pages.

[roydfreeman]

Are you using 5.1? If so, this KB might explain it... http://kb.vmware.com/selfservice/microsites/search.do?language=en_US&cmd=displayKC&externalId=204341...