[Bug 230260] [FUSE] [PERFORMANCE]: Performance issue (I/O block size)

bugzilla-noreply at freebsd.org bugzilla-noreply at freebsd.org
Thu Feb 21 16:44:25 UTC 2019


https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=230260

--- Comment #15 from Conrad Meyer <cem at freebsd.org> ---
(In reply to Kenneth D. Merry from comment #14)
I don't do stable/, but anyone is free to MFC it themselves.  It shouldn't
conflict.

> Since tape drives don't do tagged queueing, the common way to get better
> performance is to use a larger block size.  LTFS supports up to 1MB block
> sizes, and in order to read tapes from other systems and get better
> performance, we set MAXPHYS to over 1MB.  (So we can get 1MB I/O regardless
> of alignment.)  DFLTPHYS goes along with that.

Yeah, that makes a lot of sense.

(I think it is probable that FUSE should move to the tunable maxbcachebuf
instead of MAXBSIZE; MAXBSIZE is nearly orphaned in base, and can probably be
removed.  But that is somewhat orthogonal.)

Thank you for reporting this and especially mentioning the non-default
DFLTPHYS.  I did not realize it was a value people changed in their own
kernels. :-)

-- 
You are receiving this mail because:
You are the assignee for the bug.


More information about the freebsd-fs mailing list