[rescue] Insane drives

[David Passmore]

On Tue, Jul 10, 2001 at 11:26:43PM -0400, Jonathan Katz wrote:
> 
> No, you don't need to do "partner pairs" -- "partner pairs" is
> a neat setup, though.

Okay, so it's not /strictly/ necessary. But I work in an environment where
reliability is critical, and the T3 only has one controller. We don't run
anything without multiple paths, period. Hence, you run it in pairs for a
reason similar to why you should run A5x00s in pairs: they only have one
power cord. But two power supplies, go figure. 

> > Not switched fabric capable
> 
> Rumor has it the next rev will be.

Well, Sun also tells me time and time again that the A3500FC will 'work' in
the next patch for Raid Manager. It's been 2 years and they're still broken.

> > Not Sun Cluster certified (!)
> 
> Yes and no-- not officially, but Sun will install them in a
> cluster config. 

And not support it when you call them. Trust me, I've been all over this
one. It's absurd that the T3 was in development for ages (I was beta-ing it
when it was the T300 in August 1999!) and it's not supported by their
enterprise clustering software.

> Yeah, that's icky. You can do it over the serial port, too, tho.
> The converse is that it's SNMP managable, plugs into Sun Management Center, etc.

You're reaching here. Any decent array is SNMP manageable through an agent
on an attached host.

> > Very, very poor performance
> 
> Debatable-- what's the benchmark?

Load generation through vxbench, measurement through SBF/TNF. In fact, the
write performance was so bad, we gave them three weeks to fix it. They
brought it their I/O performance gurus, and even their network storage
engineering team. They simply could not figure it out, threw up their hands,
and backed an A5200 solution with fast-write-cache cards (which, BTW, are
not supported under Sun Cluster :).

David