On Oct 10, 2006, at 11:19 AM, John Francini wrote:
Now if only Darwin would support proper swap partitions (to ensure that the pagefile was contiguous and to remove filesystem I/O overhead), I'd be a happy camper.
Contiguous pagefiles make no difference in real world VM loads. Contiguous files are only important when accessing large chunks of contiguous data. VM access patterns are the exact opposite of that - tons of tiny I/Os to random locations. So it just doesn't matter, because no matter how contiguous the pagefile, there's going to be tons of seeks anyways. Only a truly pathologically fragmented pagefile could hurt performance. (IIRC Darwin uses clusters of four pages as its atomic unit of pagefile access, so there'd have to have a significant number of those 16KB clusters fragmented before you'd notice a performance hit.)
