kern/113885: [gmirror] [patch] improved gmirror balance algorithm

[freebsdpr]

The following reply was made to PR kern/113885; it has been noted by GNATS.

From: freebsdpr <freebsdpr at satin.sensation.net.au>
To: bug-followup at FreeBSD.org
Cc: freebsdpr <freebsdpr at sensation.net.au>
Subject: Re: kern/113885: [gmirror] [patch] improved gmirror balance algorithm
Date: Tue, 21 Jul 2009 17:45:37 +1000 (EST)

 I was also surprised to discover that gmirror, regardless of the algorithm 
 used, does not seem to offer either random or sequential read performance 
 any better than a single drive. I have a new SATA backplane which shows 
 individual drive activity indicators - with these you can easily see that 
 the "load" algorithm seems to be selecting (and staying on) only a single 
 drive at a time, for anywhere between 0.1 - 1 seconds. Some simple testing 
 confirmed that there's no discernable read performance benefit between 1 
 or >1 drives - so much for my 4 drive RAID1 idea!
 
 In comparison, a 5 drive graid3 array offers a sequential read speed of 
 nearly 4 times a single drive... with read verify ON.
 
 ----
 
 Onto the "load" patch above - it doesn't seem to work for me. I thought it 
 may have been because I had 4 drives in the array, but even after dropping 
 back to 2 it still only reads from a *single* drive. Any ideas? I'm using 
 7.1R-amd64.
 
 Geom name: db0
 State: COMPLETE
 Components: 2
 Balance: load  <--- ***