Skip site navigation (1)Skip section navigation (2)
Date:      Sat, 1 Jun 1996 01:57:14 +1000
From:      Bruce Evans <bde@zeta.org.au>
To:        freebsd-hackers@freefall.freebsd.org, sextonr.crestvie@squared.com
Subject:   Re: Disktab and SCSI Performance.
Message-ID:  <199605311557.BAA11037@godzilla.zeta.org.au>

next in thread | raw e-mail | index | archive | help
>	Do the disk parameters in disktab have any meaning?  It seems that when 

For some disks.

>we do the mapping from the drive's own private geometry to the scsi block 
>number, we lose the connection with the drives' own performance 
>characteristics.  I am specifically wondering about the selection of 

Usually not for SCSI disks.  It wouldn't hurt to make the disktab geometry
match the actual geometry (if any) for the most active part of the disk.

>drive speed, and to what degree the filesystem can use this information.  

The default newfs parameters are chosen to inhibit the parts of ffs that
use this info:

-d 0
Rotdelay is only useful for slow disks.  All small nonzero rotdelays are
equivalent to skipping every second block, i.e., to using an interleave
of 2 for blocks.  This is in addition to any hardware interleave, and has
much the same disaadvantages as hardware interleave: an interleave of N
reduces the best-case performance by a factor of N.  Disks with their
Iown read _and_ write buffering don't need to be interleaved at the block
level.  For disks with read but not write buffering, you can use -d 0
to get the best possible read performance but abysmal write performance,
or -d 4 to get medicocre performance for both read and write.

-n 1
This value essentially disables the rotational position table.  I think
the interleave, rpm and track skew parameters are only used to construct
this table, so this option makes them irrelevant.  The cylinder skew
parameter is never used.

-t 1 -u 4096 (one track/cylinder, 4096 sectors/track)
This limits the damage from any "optimizations" that weren't stopped by
`-d 0 -n 1' to 4096-sector boundaries.

Bruce



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199605311557.BAA11037>