I'm sorry to say I don't have any performance numbers, and this change isn't really important enough to me to spend any time generating them. I actually noticed the 1023-byte IOs while debugging a problem that led me to instrument ffs_read() activity on executable files, and the change to a power-of-2-sized IO was so simple that I took a few extra minutes to submit a PR for it.
It takes our 180mhz ARM platforms unreasonably long to run all the /etc/rc.d scripts, over two minutes. If this change buys back a couple seconds from that it'd be nice, but the real motivation for the patch wasn't that I expected the performance change to blow me away, it was just that something deep in my core rebels at knowing there's a 1023 byte IO going on when it could just as easily be 1024. I don't know of any hardware devices that work better with a 1023 byte IO size. :) -- Ian On Wed, 2011-10-19 at 22:40 +0000, jil...@freebsd.org wrote: > Synopsis: [patch] sh(1) /bin/sh: read files in 1024-byte chunks rather than > 1023 > > State-Changed-From-To: open->feedback > State-Changed-By: jilles > State-Changed-When: Wed Oct 19 22:33:38 UTC 2011 > State-Changed-Why: > Although this change looks like an improvement, it does not seem > fully satisfying. I would like to see performance numbers for the > change on your slow embedded platform. Also, why use 1023 or 1024? > Another buffer size may be better. > > > Responsible-Changed-From-To: freebsd-bugs->jilles > Responsible-Changed-By: jilles > Responsible-Changed-When: Wed Oct 19 22:33:38 UTC 2011 > Responsible-Changed-Why: > Take. > > http://www.freebsd.org/cgi/query-pr.cgi?pr=161756 _______________________________________________ freebsd-bugs@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-bugs To unsubscribe, send any mail to "freebsd-bugs-unsubscr...@freebsd.org"
