HA: A3500 vs. FC/AL

[Dave Garrison - Systems Engineer]

A couple points of clarification:

The fast write cache card(s) does have a battery backup and should
be installed in pairs (as mentioned).  The dual card config protects
against a card failure.  The battery backup protects against data
corruption in the event of a *standalone* server failure as the
cache will be flushed once the server is rebooted.

The FW cache is not supported in clustered configs because there
is no reliable way for the secondary server in a clustered pair to
know the state of data contained in a failed server's cache.  The
potential for data corruption does exist in this case.  If you
need cache and clusters, then storage based cache is preferred
(and the only way Sun will support it).

As for on-board cache for the A5x00 products, there is no cache.
What you've probably heard about is the cache that is on each
of the FC disks themselves.  The disk cache is disabled by
default (at least when Sun ships them it is).  I have personally
seen significant performance increases in some cases by turning
this cache on.  However, once again there is a risk of data
corruption if you turn on this cache, which is why Sun doesn't
support using it.  In theory the disk based cache will survive
a server crash, but just may cause corruption if the disk itself
fails with data in it's cache.  

-Dave



> X-Authentication-Warning: dns.eng.auburn.edu: majordom set sender to 
owner-ssa-managers@Eng.Auburn.Edu using -f
> Date: Mon, 10 Jan 2000 12:30:24 -0500
> From: "Steve Bulmer" <bulmer@theALLIEDgroup.com>
> X-Accept-Language: en
> MIME-Version: 1.0
> To: "Beck, Joe" <JBeck@dolsun.dol.state.nj.us>
> CC: ssa-managers@Eng.Auburn.EDU
> Subject: Re: HA: A3500 vs. FC/AL
> X-Info: To unsubscribe, send 'unsubscribe ssa-managers' to 
majordomo@Eng.Auburn.EDU in message body
> 
> Just a forewarning; since the card and thus the
> cached data lives on the server (not the storage) a
> system crash could potentially mean that your
> application thought the write was complete but it
> never actually made it to disk.  I was *told* by
> a Sun engineer that the fast write card is not
> supported in Sun clusters for this reason.
> 
> You should note that the fast write cache card
> when ordered actually comes with 2 cards so
> the cache is mirrored and protected in case of
> just a cache card failure.
> 
> In terms of fast write cache in the A5x00 I do not
> know of any, I would have to check my books. Your
> document sounds like it has more info than I do. I
> know there was some limited cache in the older SSAs
> like 100's 110's 200's etc.
> 
> And yes, the fwcadm comes as part of the software you
> get with the fast write cache cards.
> 
> Steve
> 
> "Beck, Joe" wrote:
> 
> > Was this potential for lost/corrupted data just a forewarning or did 
you
> > experience problems?
> > Very curious as we are suppossed to start running performance tests 
in our
> > environment (large OPS, read intensive app)--trying to decide 
whether it
> > behooves us to look into fast write cache for our A5000's & A5200's.
> > Also, I was told that even w/out a fast write cache controllers 
there was a
> > way to turn on fast write cache on the Array itself.
> > Do you know if this is true? or is this capability availability at 
both
> > levels?
> > I recently received a document sun storedge fast write cache best 
practices,
> > so i guess that's a good starting point.
> > one thing i noticed is a fwcadm command which must be part of a 
package that
> > comes with the controller?
> > Any feedback on this topic is appreciated
> >
> > ______________________________________________________
> > Joe Beck        Unix Administrator/Tax Redesign Project
> > jbeck@dol.state.nj.us
> > voice:  (609)292-5785
> >
> > The roots of education are bitter, but the fruit is sweet. 
-Aristotle
> >
> > > -----Original Message-----
> > > From: Steve Bulmer [SMTP:bulmer@theALLIEDgroup.com]
> > > Sent: Monday, January 10, 2000 11:12 AM
> > > To:   Seth Rothenberg
> > > Cc:   ssa-managers@Eng.Auburn.EDU
> > > Subject:      Re: HA: A3500 vs. FC/AL
> > >
> > > The only supported way to have 3 Sun hosts connected
> > > to 2 Sun storage arrays is to use Sun A5200 technology and
> > > direct connect the computers to the arrays WITHOUT hubs.
> > > This is the only way Sun will support more than 2 servers 
connected
> > > to fibre channel disks. Their 3 and 4 node Sun clusters must be 
con-
> > > figured this way.
> > >
> > > You *might* consider using the fast write cache controllers for 
the
> > > Suns (they are SBUS cards) if you go to the A5200 solution but be
> > > aware that this is not a good idea for failover or Oracle Parallel 
Svr
> > > as data can be lost/corrupted in cases of host failure.
> > >
> > > As for the need for cache, it helps most in small I/O, random, 
mostly
> > > write
> > > intensive applications and less as you move toward larger, 
sequential read
> > > intensive applications.
> > >
> > > Steve
> > >
> > > Seth Rothenberg wrote:
> > >
> > > > SSA Managers, Help !
> > > > I have spent the past month with my Sun Sales rep
> > > > designing an HA system using two Enterprise 4500's and
> > > > one A3500, 2x2 "Lite+" configuration.  Now, my boss says
> > > > we need a third host.   So, I asked about the A3500FC.
> > > > My Sales/Tech said the 3500FC control module only has
> > > > one FC (is that per controller?), and therefore, it is not a 
"loop",
> > > > but really a "hub".  The Sales/Tech did not know if we can
> > > > install 3 hosts and 2 controllers using 2 hubs.
> > > >
> > > > IF A3500 does not support 3 hosts, another idea is to just
> > > > get two A5200's, which really can do a loop.
> > > > However, we give up the Controller Module.
> > > > We need to evaluate whether the Read/Write Cache in the
> > > > controller can be sacrificed.  My impression is yes.  I suspect
> > > > my boss won't agree.
> > > >
> > > > The A5200's also have hubs.  I would be interested in
> > > > whether anyone has one of these configurations.
> > > >
> > > > Thanks
> > > > Seth
> > >  << File: bulmer.vcf >>