IDE -- mount partitions for better performance

[Adam Vande More]

On Tue, Mar 15, 2011 at 3:59 PM, Polytropon <freebsd at edvax.de> wrote:

> Keep in mind that performance "across" ad0 and ad2 is best.
> Masters are always good. Slaves are slower. Using primary
> and secondary in parallel works good, working on a master
> and a slave simultanously is worse.
>

Your statement about master being faster than a slave is simply not true for
almost every scenario when using devices with same capabilites.  All
master/slave really controls is enumeration, and shouldn't effect
performance in and of itself.  Other variables can effect that of course,
like using a slower device as an ATA Device-1 with a faster Device-0.  Even
that example isn't ubiquitous as many, maybe most controllers are able to
support mixed devices each in their fastest mode.

The whole IDE device contention really isn't much of a bottle neck in this
scenario.  It's only a big factor when there's *a lot* of simultaneous IO
going to both, say dumping one disk to another.

The highest preforming setup in something like this is likely to be
something along the lines of a 4-way /boot gmirror, and a 4-way gstripe with
a smaller stripe size eg 32k across the remaining usable space.  If you
aggregate your disk IO in this manner, IDE channel contention shouldn't be
much of a bottleneck.



-- 
Adam Vande More