Summary: SAN questions

From: jgormley (jgormley@scu.edu.au)
Date: Sun Aug 10 2003 - 22:25:13 EDT


Thanks to the quick responses from:
Nasyr Yilmaz
Nelson (Kawumin) from Uganda
Colin Bull
Anthony Bento
Alan Rollow - great detail!
Raul Sossa
And in particular Greg Rudd from Uni Sydney who was kind enough to
follow through with a phone call and heaps of info.

Original Question:
We are currently looking at getting into the world of SANs with an
initial setup of about 3 terabytes. We want to connect Tru64, linux,
Windows 2000 and Novell boxes.
I have read through some 600 postings in the archives of problems and
information people have shared.
Is there anyone that can offer any fairly recent information (not
already posted) about any pros or cons of any particular brand of SAN
they are using, any gotchas or good feelings etc. I am particularly
interested in info relating to implementation of the SANs, backups
(software and procedures) and ease of use of SAN related management
tools.

I followed up to most of the respondents with further questions about
backups and virtualisation.

Responses:
Yasir uses Legato for backup software and brocade switches on an HSG80
- some initial problems with snapshot but O.K. now. Advises EMC as a
possibility but likes HP. He also suggested that when doing snapshot of
a database the database needs to be quiesced.
                        -----------------------------------------
Nelson suggest we insist on a fully redundant system. He is using hsg80
in a model 2200 enclosures connected to 4354 disk enclosures. OS's are
TRU64, Windows 2000 and Solaris. He says SAN is the way to go.
                        -----------------------------------------
Colin - The easiest way to back up is to do a snapshot, which freezes
an image of the data at a moment in time. The DEC/Compaq/HP Storageworks
with HSx controllers suggest you can do this. you MUST have the
controller with version x.xxS, for snapshot which is approx USD5k extra
per controller. He uses Legato with Informix database and does not have
the snapshot facility.
                        -----------------------------------------
Anthony provided the reference to
http://h18006.www1.hp.com/products/storageworks/san/documentation.html
and suggested chapters 3 and 4 of the SAN Design Reference Guide.
                        ----------------------------------------
Allan Rollow - My bias has to suggest the EVA, but since I'm currently
contracted to HP, it is bias. I think the drive backend on the EMC
subsystems may still be parallel SCSI, but isn't necessarily a bad
thing; well understood, good performance, good price, etc. The Fibre
Channel back-end of the EVA has a frequent source of problems. The
copper interconnect (I think the backend of the EVA 3000 uses copper),
instead of optical and the limited
configuration may help the EVA 3000.

I don't have a SAN and do my backups to a local TZ87 tape drive...

A suitably functional version of backup software from a single vendor
will tend to be able to coordinate access to multiple drives that have
equal access from multiple hosts.

Operating systems have been doing virtualization of memory for a very
long time. Moving those concepts to Storage Management is very useful.
Generally when virtual memory is discussed, two very separate things
tend to be merged:
o The apparently address of a page (block) as opposed to its physical
address (LBA).
o The ability to have more virtual memory promised than physical memory
available by using some other type of storage as a backing store
(page/swap space).

Doing the first in Storage virtualization, seems to have gone pretty
well in the EVA family. If there were problems it was tested out of the
product long ago. It seems to
offer considerable advantages, compared to the disadvantages.One thing
you do lose, is knowing where your data really is, if that matters. A
bad block seen by the host is very
hard to track down to a back-end disk. On the other hand, proper choice
of RAID levels makes it unlikely that you need to care. Read errors
will simply regenerate the right
data and find a new block for it. Write errors will be hidden. Failing
disks will be noticed by the firmware, the data migrated to spare space,
the disk dropped and a notification sent to you or your service vendor
that the disk needs to be replaced (*).

Configuration is more complicated at some stages, but all the relevant
vendors, can help deal with that. Other parts of the configuration are
very easy, at least from my experience with the EVA family. Creating a
new virtual disk is very easy; pick the size, disk group, RAID level and
which host should see it.

I don't know the extent to which the various vendors try to use the
other characteristic of virtualization; over-commit. The user interface
for the EVA family won't let
you overcommit when creating an initial unit. If space disappears as a
device is being created or expanded, then the allocation will stall and
the storage system will
deliver an event to say "add a disk". Snapshot functionality seems to
just be like AdvFS cloning. There's always a risk there that the
changes to the original will overwhelm the
available space, so some care is needed.
 
(*) With the right level of support contract or setup of the relevant
tools.
                                --------------------------------
Raul suggested a configuration of
1.Enterprise Virtual Array 5000.
2. hp Business Copy Software for EVA with HP Dataprotector 5.1.
3. Snapshots and Clones Backups.
4. SDLT 160/320 SAN Backup technology.
5. Zoned SAN Implementation.
6. FC Switches upper than 16 ports (from hp obviously).
7. Tru64UNI/TruCluster 5.1APK4/51BPK1.
8. Windows 2000 SP2/SP3 with hp Secure Path 4.0A/B

If you can not buy a EVA3000, you might buy two MSA1000 with Virtual
Replicator Software. Dataprotector is OK, might use veritas or
ARCServeIT.
I use virtualization, it's pretty cool to see how you can shutdown your
oracle db, take a one second snapshot, startup your database, continue
your business and then put the Snapshot backup into the tape drive,
without worry about the OFFLINE backup for a 6TB DB size that last 7
hours.
                -------------------------------------------
Greg-
HSG80 a good old Dec product. The advantages of this system is that it
can
be managed by cli and in operation is similar to a hsz70. The
disadvantages
of this system are the disks can only go up to 20mb/s and the product
is
about to be no longer sold as a new product.

HSV110 HPs virtual storage array. These are a relatively quick array.
As a
virtual storage array you don't know where the data is stored unlike the

HSG80. These use fibre channel disks have a backplane speed of about 2
GB/s
The model ones used FC AL as the interconnect between drive shelves but
latter models used a switch based architecture. While this system is
very
fast ( the alphas work like a charm on them) we had a failure in Feb
with our
unit and the disks got shiped back to the US with HP providing loan
HSG80
storage and by the way we are still waiting on a staisfactory anwser
from HP
as to what had happened.

We use legato networker backing up to a dlt (soon to be retired) and lto
silo
with about 240 tape capacity
What OS's do you have attached to your SAN?
Tru64 5.1x
Solaris
Linux
Win2k

What I will warn you about is that with all of the above OSes listed
with the
exception of Tru64 5.Xx you will need to purchase secure path software
if you
are using more than one HBA per machine so you can have failover on each
of
the paths. This software in the case of HP costs $$
Virtualisation works ok but when something goes wrong like what happened
to us you dont
really know where the data is.
 

Thanks again for all of the response.

Regards, John
 
John Gormley
 
Senior Unix Systems Administrator
Information Technology and Telecommunications Services
Southern Cross University
Lismore NSW 2480
AUSTRALIA
Office :(02)6620 3365
Mobile: 0409 150060
 



This archive was generated by hypermail 2.1.7 : Sat Apr 12 2008 - 10:49:30 EDT