Don MacAskill wrote:
Arjan van de Ven wrote:On Mon, 2006-03-06 at 16:27 -0800, Don MacAskill wrote:We have an 8-way RHEL4 AS system (4 dual-core Opterons) with no swap (because swapping is still broken for DB loads. It swaps the DB out in preference of disk cache).zero swap is a really bad idea. At least make a 10Mb or so swap.Created a 10MB swap file and turned it on. kswapd3 immediately stopped using 100% of a CPU and I'm seeing much more normal usage. (And only like 5-7MB of so of swap used, generally. That may be because of the swappiness and vfs_cache_pressure settings, though...)Thanks Arjan!I'd still love to hear more about why it's a bad idea to have no swap, particularly given the number of diskless nodes we have and plan to deploy. Apparently my swap knowledge is lacking. :)
Spoke too soon. As load increased, and the 10MB was used, the problem returned. kswapd3 is again using 100% of one of my cores, and my system is dog slow.
Strangely, there's almost no I/O on the disk that has the swap. It's not like it's churning on that 10MB swap a ton.
Someone else from Red Hat has seen more of these reports and thinks they've narrowed it down to this patch: 'linux-2.6.9-vm-wired-pagetables.patch'
More as I get it, but I'd love to hear any other advice if anyone's got some.
Don