Summary: T3 array

From: Ron Olphie (rolphie@lincap.com)
Date: Thu Jul 13 2000 - 10:02:35 CDT

Messages sorted by: [ date ] [ thread ] [ subject ] [ author ] [ attachment ]

Sun Managers,

I received three responses which are listed below.

Thank You:
Kevin Buterbaugh [Kevin.Buterbaugh@lifeway.com]
CIsenburg@ixl.com (Carl)
Tim_Lindgren@enron.net

Regards,
Ron

attached mail follows:

Sun Managers,

I'm considering purchasing a T3ES array. The array would be dual ported,
dual connected to an E3000 and an E4000, both the E3000 and the E4000 run
Solaris 2.6.

The major application would be Sybase, mostly transaction processing. I
currently own an A5000 and although I'm very satisfied with it's
reliability, I've been disappointed in the performance we've experienced.
I believe the following factors contribute to my performance
dissatisfaction:
1) We are using Veritas (Sun's version of Veritas, we don't have the quick
io module). We stripe and mirror Sybase volumes.
I believe we would have better performance if we went with raw partitions.
2) The A5000 doesn't have a write cache.
3) The DMP doesn't appear to load balance.
4) Confusion about the stripe width, Sybase's blocksize has not been
altered, and we are using a 64K stripe width.

We are also looking at several SAN vendors. MTI, Xiotech, and StorageTek.
If we went with one on these vendors, NT and Solaris would share their
array. If we purchased the T3, it would be for Solaris only. Is it even
possible for NT to reasonably use the T3?

The last point of confusion concerns the write cache(or lack of a write
cache). MTI has a mirrored cache, Xiotech has a buffer on the controller
that schedules writes to reduce latency and seek time. This "appears" to be
the case with the T3, since it uses the controller buffer to schedule
writes.

Any experiences with the above products, especially involving Sybase, would
be appreciated. I will summarize.

Ronald Olphie
Lincoln Capital Management Company
200 South Wacker Drive
Suite 2100
Chicago, IL 60606
rolphie@lincap.com
Telephone 312-627-4331
Fax 312-559-1394

S
U BEFORE POSTING please READ the FAQ located at
N ftp://ftp.cs.toronto.edu/pub/jdd/sun-managers/faq
. and the list POLICY statement located at
M ftp://ftp.cs.toronto.edu/pub/jdd/sun-managers/policy
A To submit questions/summaries to this list send your email message to:
N sun-managers@ececs.uc.edu
A To unsubscribe from this list please send an email message to:
G majordomo@sunmanagers.ececs.uc.edu
E and in the BODY type:
R unsubscribe sun-managers
S Or
. unsubscribe sun-managers original@subscription.address
L To view an archive of this list please visit:
I http://www.latech.edu/sunman.html
S
T

attached mail follows:

I would suggest using the STK 9176 controller with the T3 disk arrays.
Why...the T3 controller only allows you to bind 2 LUNS per array, and does
NOT
support Lun masking, Lun partitioning, real mirrored cache, etc. This means
you'd need VM to cut up the luns anyway - causing more bottlenecks once
again.

STK sells the T3's as 9170's so your getting the same disk. The controller
and
associated software is the key. You can bind a 4+1 raidgroup, and cut this
into
rightsized LUNS for your databases. And yes, NT and Unix can share the same
raidgroup without fear of contamination.

Your performance issues with the A5000's is typical. They are JBOD disks
with
nothing to help I/O.

I've pushed A5x00's to the limit, and only got an avg. of 9-15 ms access
times
at best for random write. We moved this off to STK 9145 controllers
(actually
Clariion's) and bumped this to 2-3ms access times for heavy used databases.

DMP is the work of the devil. I would suggest an auto-tresspass feature that
would gain the advantage of load balancing, but be transparent to the OS.
STK,
EMC, etc all offer this.

Managing database raw partitions is a nightmare. I get the simular
performance
using quick I/O.

Hope it helps...
Tim

attached mail follows:

Can't offer you anything about Sybase, but I spent some time comparing these
vendors, and I may have some data for you. That said, it has been almost a
year since I have been hands on with these products, so qualify what I
say...

I spent 6 months evaluating SAN stuff, and ended up buying 4 Mags from
Xiotech. This was before they were purchased by Seagate. Their box just
had the best feature set, and the best performance available. The decision
maker for me was the way that it virtualises all of the disks, instead of
trying to push you toward a finite number of drives per volume. They are
pretty proud of their software, which is cool... not as cool as they think,
but cool.

The StorageTek box is just a repackaged Clariion array. It is a pretty
standard disk, but when I last spoke with them, they were using a
proprietary HAB. They may have switched to the Emulex and Q-Logic cards.
If so, it is a decent array, but not as fast as the Xiotech box.

I didn't spend that much time with MTI. I spent some time with their
engineers, and I felt like I knew more about SANs than they did, and I felt
like everybody in their company was more interested in spouting propaganda
than making hardware. Not what I was looking for...

Factors I looked for:
When I thought about it, it is pretty obvious that the virtual disks that
Xiotech creates are going to be considerably faster than any simple,
typical, solution. Once that got me, it became time for the other vendors
to try to show me how their product had some other feature that Xiotech
didn't have. When they failed, I wrote my PO...

Before I left that company, we only had time to do one real load test. We
hooked up 4 servers, that were each 4X450 with 4G (Compaq) and started up
some test processes written by Xiotech. We were performing 20,000 IOs per
second ... per server. If you want it to be fast for small IOs like db
processes, you need to get something like this...

Sorry this response became a testimonial, but ... what the hell!

Carl

-----Original Message-----
From: Ron Olphie [mailto:rolphie@lincap.com]
Sent: Wednesday, July 12, 2000 6:33 PM
To: 'sun-managers@sunmanagers.ececs.uc.edu'
Subject: T3 array

Sun Managers,

I'm considering purchasing a T3ES array. The array would be dual ported,
dual connected to an E3000 and an E4000, both the E3000 and the E4000 run
Solaris 2.6.

Any experiences with the above products, especially involving Sybase, would
be appreciated. I will summarize.

Ronald Olphie
Lincoln Capital Management Company
200 South Wacker Drive
Suite 2100
Chicago, IL 60606
rolphie@lincap.com
Telephone 312-627-4331
Fax 312-559-1394

attached mail follows:

Ron,

I don't have any experience with either the A5000 or the T3. However,
we do use Sybase heavily (on Sun SSA's and / or EMC) and I can tell you
that (as you expect) you will see much better performance using raw
partitions over Unix file systems for your databases. Hope this helps a
little...

Kevin Buterbaugh
LifeWay

"Anyone can build a fast CPU. The trick is to build a fast system." -
Seymour Cray

Sent by: owner-sun-managers@sunmanagers.ececs.uc.edu

To: "'sun-managers@sunmanagers.ececs.uc.edu'"
<sun-managers@sunmanagers.ececs.uc.edu>
cc:

Subject: T3 array

Sun Managers,

I'm considering purchasing a T3ES array. The array would be dual ported,
dual connected to an E3000 and an E4000, both the E3000 and the E4000 run
Solaris 2.6.

The last point of confusion concerns the write cache(or lack of a write
cache). MTI has a mirrored cache, Xiotech has a buffer on the controller
that schedules writes to reduce latency and seek time. This "appears" to
be
the case with the T3, since it uses the controller buffer to schedule
writes.

Any experiences with the above products, especially involving Sybase, would
be appreciated. I will summarize.

Ronald Olphie
Lincoln Capital Management Company
200 South Wacker Drive
Suite 2100
Chicago, IL 60606
rolphie@lincap.com
Telephone 312-627-4331
Fax 312-559-1394

Messages sorted by: [ date ] [ thread ] [ subject ] [ author ] [ attachment ]

This archive was generated by hypermail 2.1.2 : Fri Sep 28 2001 - 23:14:11 CDT