Re: raid and cgd

[Mouse <mouse%Rodents-Montreal.ORG@localhost>]

>> RAIDframe operates on units called stripes.  [...RMW...]
> Your description matches a paragraph in raidctl(8) that starts with
> 'Tuning RAID 5 sets is trickier.' while my setup is RAID 1

Ah, that's important information.

> and recommended values for SectPerSU are 32 to 128.  I set it to 128
> but I'll try 64 to see if makes a difference.

In theory, RAID 1 - mirroring - does not have to have a "stripe" size;
writes could in principle just be sent to all devices with the size and
offset unchanged.  I don't know whether RAIDframe actually does that,
though; raidctl(8) on one of my systems says, in part,

     For a RAID 1 set, a SectPerSU value of 64 or 128 is typically sufficient.
     Since data in a RAID 1 set is arranged in a linear fashion on each compo-
     nent, selecting an appropriate stripe size is somewhat less critical than
     it is for a RAID 5 set.  However: a stripe size that is too small will
     cause large IO's to be broken up into a number of smaller ones, hurting
     performance.  At the same time, a large stripe size may cause problems
     with concurrent accesses to stripes, which may also affect performance.
     Thus values in the range of 32 to 128 are often the most effective.

which certainly makes it sound as though it insists on doing I/O in
units of stripes even in cases like RAID 1 where it's not actually
necessary.  If RAIDframe is similar on your version, you probably want
to make sure your filesystem boundaries align with your RAID stripe
boundaries.

Similar issues apply if you're saddled with an "advanced" format drive
which semi-lies about its sector size (typically, it claims half-K
sectors when it actually has 4K sectors - I say "semi-"lies because
they generally do provide an interface with which the OS can determine
the actual sector size and alignment).  In this case, the drive will be
doing RMW cycles unless your I/O requests' sizes and alignments match
those of the underlying physical sectors.  Unfortunately, such drives
are becoming more and more common.  (While I know a good deal less
about them, I think SSDs have some related-but-different issues.)

/~\ The ASCII				  Mouse
\ / Ribbon Campaign
 X  Against HTML		mouse%rodents-montreal.org@localhost
/ \ Email!	     7D C8 61 52 5D E7 2D 39  4E F1 31 3E E8 B3 27 4B