How do we get 4% faster when running on the same endianness as the fs was
created on? (created on x86, run on amd64) Where are the bulk of the calls
to BSWAP_* happening in that case?
--matt
Dan Anderson wrote:
> Here's some performance results running " find . -exec ls -l" on separate ZFS
> filesystems created on x86 and sparc and imported/exported to amd64, em64t,
> and sun4u platforms. This shows performance gain from optimized byteorder.h
> macros.
>
> Percent savings, real time
> ZFS filesystem created originally on:
> Platform x86 sparc
> amd64 4% 3%
> em64t 3% 4%
> sun4u 4% 2%
>
> Environment:
> * Create 2 separate ZFS filesystems with 1024 directories, each with 32 files,
> are on x86 and sparc and zpool export/import to the other systems.
> * Run this command on ZFS filesystem: find . -exec ls -l {} \; >/dev/null
> * Run using NV97 with and without fix to RFE 6729208 (byteorder.h macro
> optimization)
>
> BTW, I still could use some code review comments:
> http://dan.drydog.com/reviews/6729208-bswap3/
> --
> This messages posted from opensolaris.org
> _______________________________________________
> zfs-code mailing list
> zfs-code at opensolaris.org
> http://mail.opensolaris.org/mailman/listinfo/zfs-code
