Re: Does XFS on hardware RAID5 have perfomance issues?

[John W <jwest@nic.com>]

> The issue with the internal log with software
> raid5 is that the alignment (or lack thereof) of the logwrites causes
> invalidation of the internal cache in the raid code which is used for
> parity generation.

All,

If this is the case, then the RAID code get flushed and then sucked back
into 
cache too often?  Guess the dinky cache of the Intel Boxen is the weak
point.
There once was a way, in a distant land, to mark executables as
stickier, so they
would not get flushed as often as data... thus the code would more
probably stay 
in cache. Do multiple CPU's even compound matters by killing coherency
even more?

The idea of a software RAID implementation spoils the purity of the
whole SCSI thing. 
More thrashing of the CPU for disk transactions, sounds more like IDE!
Hardware RAID 
all the better...

Using separate partitions for log on same disk would make the disk head
reposition to a 
different location,  wouldn't that be bad for a big writes?  Having a
log on 
a different or even dedicated disk sounds like a great idea... but that
is intuition!

Is this whats happening?  Learning lots on the list!

John Westerdale

# if you cant xylem, phloegm #