Hi, On Tue, 18 Oct 2005 13:27:06 -0400 Ken Schneider <suse-list@bout-tyme.net> wrote:
For 2.6 kernels I'd suggest a swap file. There's nearly no overhead compared to a swap partition. [...]
Sorry but you have that backwards. A swap partition is much more efficient than a swap file. You don't have the overhead of the filesystem i/o to deal with. Same reason many databases are run on raw partitions.
Sorry, but it's not me who didn't his homework lately, see: http://www.kerneltraffic.org/kernel-traffic/kt20050828_319.html#3 Short summary: I was right... Maybe I should quote the most important sentence (by A.Morton): "The kernel generates a map of swap offset -> disk blocks at swapon time and from then on uses that map to perform swap I/O directly against the underlying disk queue, bypassing all caching, metadata and filesystem code." Back in the old 2.4 era, you were of course right. And databases tend to run in userspace that doesn't offer the new path easily, don't they? -hwh