Summary: Solaris & Veritas SAN question

Thanks to the following members of this list for their prompt replies.

 

Jon Hudson

Jeff Barratt-McCartney

Ddelija

 

I have pasted the body of Jon Hudson's comprehensive reply below.

 

 

In a well tuned SAN on local multimode fiber (62.5, <150m) all scsi requests
should be processed (ECT;End completion Time) in under 500ms. Should be able
to get down to 100ms often, but under 500ms is the target. Sometimes really
low, 

 

Once you go to single mode (long haul) you start having scsi time out
issues. You limitation ends up being between 120-220miles depending on
equipment. This is just due to physics limitation of fibre. Now, you can of
course have devices in there that allow multi-hop synchronous replication,
but it's only for replication. 

 

Now in a test lab I saw a Nishan box in CA allow hosts to mount disk in NY.
But the Nishan had to buffer and cache everything so they could deal with
all the timeout issues. I can't remember what scsi times out at, I'd have to
check. But most solutions I've seen are under 120miles.

 

As far a comparing internal drives to FC disk array, that really depends on
what the internal drives are. Fiber channel drives do really well either
internal or external as long as you are dealing with good size data sets in
that can be carved up into neat 2048 frames. Basic scsi can out run fiber
channel when dealing with lots of really small files. FC ends up having to
pad the 2048 frame causing addition interrupts on the hba. I feel pretty
safe saying in most cases a 2Gb attached array tuned properly with 10k
drives will out perform any internal or scsi attached scsi array. But I
could construct special cases with really small files where this isn't true.
You know how it works, you can always construct a niche test that will prove
your point =)

 

So for example an v880 with internal FC drives can perform well since it's
fiber, but now as well as an external array because it's FCAL

(loop) and not FCSW (switched).

 

But with the right equipment you "should" be able to do several kilometers
and be ok, but I'm not sure of the performance numbers. We

(Finisar) make long haul single mode SFPs that can do 80Km without a
repeater that might work. But you are probably best off getting Brocade and
Nishan (now own by McData) to show you their solutions. Don't talk to Cisco,
their FC stuff sucks.

 

Make sure you spend the time to properly tune the credit buffers, this can
make or kill long haul stuff.

 

 

Thanks,

 

Conor

 

 

-----Original Message-----
From: Conor Svensson 
Sent: Friday, January 14, 2005 3:51 PM
To: 'sunmanagers@sunmanagers.org'
Subject: Solaris & Veritas SAN question

 

Hi, 

 

We are looking to implement a large scale SPARC Solaris system which has two
main priorities. 

1.	Disk latency - ie fastest possible return of any calls to disk IO 
2.	Failover to remove site 

 

I am investigating using a Veritas to manage a SAN mirror of Fibre Channel
disk arrays. The application writes relatively little data. 

 

Does anyone know how attaching a FC mirror several kilometres away will
affect the disk i/o latency?

How does the i/o latency of an internal disk compare to that of a tuned FC
disk array? 

 

I will summarise, thanks in advance,

 

Conor

 

___________________________________________________________
Evolution is the investment banking and venture capital industry's 
first choice for practical advice on strategy, business process and
the application of advanced technology. 

 


Conor Svensson                         Direct: +44 (0) 20 7898 0348
Evolution                                   Tel:      +44 (0) 20 7664 6640
Peninsular House                       Fax:     +44 (0) 20 7664 6641 
30-36 Monument Street
London EC3R 8LJ                      Email:
<mailto:Conor.Svensson@evolution.net> Conor.Svensson@evolution.net
United Kingdom                         URL:  <http://www.evolution.net/>
http://www.evolution.net
___________________________________________________________
 
The information in this Internet e-mail is confidential and is 
intended solely for the addressee.  Access, copying or re-use 
of information in it by anyone else is unauthorised.  Any views
or opinions presented are solely those of the author and do
not necessarily represent those of Evolution or any of its
affiliates.  If you are not the intended recipient please contact 
Evolution, London, +44 (0) 20 7664 6640
___________________________________________________________
_______________________________________________
sunmanagers mailing list
sunmanagers@sunmanagers.org
http://www.sunmanagers.org/mailman/listinfo/sunmanagers