I'm through with my first level of changes to swap_pager.c and the results are better than even I had hoped. The new per-device round-robin allocation performs a tad _better_ than the old striping when it comes to distribution of I/O requests over multiple devices, given that there now is no upper limit on the number of swap devices, this is a clear win-win. For systems with less than the (previously) hardconfigured NSWAPDEV limit, kernel malloc usage is lower, now we only allocate that space which we need. The next level of changes will be to go directly to GEOM for disk devices, rather than take the detour over vnodes and specfs. This will not happen in the first couple of weeks I suspect. -- Poul-Henning Kamp | UNIX since Zilog Zeus 3.20 phk_at_FreeBSD.ORG | TCP/IP since RFC 956 FreeBSD committer | BSD since 4.3-tahoe Never attribute to malice what can adequately be explained by incompetence.Received on Wed Aug 06 2003 - 03:35:12 UTC
This archive was generated by hypermail 2.4.0 : Wed May 19 2021 - 11:37:17 UTC