[rescue] hardware RAID, esp. EMC, vs SS20 & SSA100 w/ VXVM

[Derrick D. Daugherty]

It's rumored that around Tue, Apr 02, 2002 at 03:42:58PM -0500
"Greg A. Woods" <woods at weird.com> wrote:
> [ On Tuesday, April 2, 2002 at 18:36:49 (+1000), Robert Rose wrote: ]
> > Subject: Re: [rescue] somewhat OT: secondary market storage?
> >
> > At 06:49 PM 1/04/2002 -0600, some un-named being posited:
> > >
> > > Hardware RAID will always be better than software RAID.
> > 
> > Sorry to disagree with you there, but when it comes to EMC at least, that's 
> > not the case.  A few years ago I found once you fill the write cache (I 
> > think it was 1GB), performance drops significantly,
> 
> Yeah, well what do you expect!?!?!?  
> 
> > such that a Sparc 20 - 
> > SSA100 - VXVM combination performed better.  That had the EMC techs running 
> > in circles, then they claimed that my test wasn't fair (vxbench) because it 
> > wasn't a real-world simulation.
> 
> Hmmm....  Same disks?  Same RAID config?  Did you scale the benchmark
> parameters to match the relative scaling of the host's RAM too?

this is because their "little" caching alg's take up 25% PLUS of your
cache.  so you don't really get that cache.  their 'os' lives there.
you lose.

again, their BUS arch sucks.  they keep saying their cache takes care of
it, it doesn't.  their cache guys will tell you this, well, they'll tell
the salesdroids this.  if you lose a cache card, you're pretty much
done.

does it happen?  oh hell yeah, at the worst time.