[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: hot-add unformatted drive to RAID array automagically

On Friday 04 February 2005 19:32, Wouter Verhelst <wouter@debian.org> wrote:
> > > So, what you want is hardware RAID. Not just because it does that, also
> > > because it's (usually) quite a bit faster.
> >
> > Is this proven or a myth?  I've seen analysis that say software raid is
> > faster, sometimes much.  I guess "it depends" (on CPU, if server is i/o
> > bound or cpu bound, etc.)
> Having a dedicated processor that does just one task usually implies
> faster processing than programming a general-purpose processor for the
> task.

Not if the dedicated processor has the performance of a 486 and the general 
purpose processor is a P4 or Athlon.

> Note, I said "usually". I've seen a number of RAID arrays, and while
> some are extremely fast, others are even slower than a single regular
> hard disk when running in RAID5 mode.
> I've had good experiences with Mylex RAID-controllers in that regard.

In late 2000 I did some benchmarking of a Mylex Dac960 controller.  It gave a 
maximum performance for bulk linear reads/writes of about 10MB/s.  Linux 
software RAID gave 30MB/s on the same test with slower disks.  However the 
Mylex gave good performance for random seeks, extra caching I guess.

A RAID-5 setup without NVRAM for write-back cache will give poor write 
performance.  If you want to write one sector to a RAID-5 array then the 
controller has to read the original sector and the checksum sector and then 
write to the sector and the new checksum.  This makes a single write become 
two reads and two writes (one read and re-write on each of two disks).  All 
good hardware RAID systems have NVRAM write-back caches to allow 
write-combining on RAID-5 lines.  This gives a massive performance benefit.

For RAID-1 it's a lot easier to get good performance, Linux software RAID-1 
didn't give optimum performance last time I tested.  RAID-1 should give 
double the read performance of a single disk for the case of two processes 
reading large files linearly and for the case of a large number of processes 
that are disk read bound.  Linux software RAID didn't do so last time I 

http://www.coker.com.au/selinux/   My NSA Security Enhanced Linux packages
http://www.coker.com.au/bonnie++/  Bonnie++ hard drive benchmark
http://www.coker.com.au/postal/    Postal SMTP/POP benchmark
http://www.coker.com.au/~russell/  My home page

Reply to: