Re: StorEdge 3310, disk sets, 2 hosts

• Previous message: David J Dachtera: "DNS / Resolution question"
• In reply to: noone: "Re: StorEdge 3310, disk sets, 2 hosts"
• Next in thread: noone: "Re: StorEdge 3310, disk sets, 2 hosts"
• Reply: noone: "Re: StorEdge 3310, disk sets, 2 hosts"
• Reply: Scott Howard: "Re: StorEdge 3310, disk sets, 2 hosts"
• Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]

Date: Fri, 24 Sep 2004 20:25:43 -0700

noone wrote:
> Rytis Sileika wrote:
>
>> noone wrote:
>>
>>> I need to evaluate ( since I have not done this before ) and test
>>> having a "shared" disk array via the following setup:
>>>
>>> * 2 x StorEdge 3310 ( 1 3310 is a mirror of the other via DiskSuite
>>> mirroring )
>>> * Each StorEdge 3310 currently has only 1 controller
>>> * Hosts to be attached to the 3310 are a SunFire 280R and a
>>> SunEnterprise 220R
>>>
>>> The idea is to have one host actively running a database ( sybase,
>>> oracle, etc ) and reading/writing to disksuite raw partitions on the
>>> array, while the other host is a ( very ) cold standby. So if hosts A
>>> fails, host B takes ownership of the disksets on the array, and the
>>> database is started on host B.
>>>
>>> Sun clustering and HA are __NOT__ involved in this setup ... only
>>> disksuite disksets.
>>>
>>>
>>>
>>> Now to the setup questions:
>>>
>>>
>>> 1) One or two controllers on the 3310 ?
>>>
>>> The 3310, even with just one controller, allows you to have just one
>>> controller and 2 hosts connected to the same controller:
>>> ( See Figure 4-6 below: )
>>>
>>> http://docs.sun.com/source/816-7290-15/ch04_cable.html#48586
>>>
>>> Do I need 1 or 2 controllers ( one controller on 3310 per host ) ?
>>>
>>>
>>> 2) If I really need 2 controllers, what I plan to do is take the
>>> controller from one 3310 and add it to the other 3310. Any issues
>>> with this ?
>>>
>>>
>>> 3) Finally, has any done this before and is this setup ( diskset, 2
>>> hosts, NO clustering and NO HA ) supported by Sun ?
>>
>>
>>
>>
>> Why would you ever possibly want to mirror disks using software
>> mirroring tools when you have hardware controllers available to
>> perform the same task with a far greater efficiency??
>>
>> What I would propose is to put both raid controllers into one disk
>> array (if you don't like the idea of throwing the second one away, you
>> can jatt#ach it to the first controller as a jbod) and have two scsi
>> cables running from each node down to 3310. this wday you won't need
>> sds. If you want some automatic failower without using sun cluster
>> just write some scripts arroud your dbases to check the availability.
>> and of course manual switchover is always possible.
>>
>> Rytis
>
>
>
> Forget about the mirroring of the arrays via disksuite for the mean time.
>
> Assume for the moment that there is just 1 array, the question still
> stands:
>
>
> 1) One or two controllers on the 3310
>
> Do I need 1 or 2 controllers ( one controller on 3310 per host ) ?
>
>
> 2) Any issues of taking the controller from another 3310 and placing it
> on my existing 3310.
>
>
> 3) Finally, has any done this before and is this setup ( diskset, 2
> hosts, NO clustering and NO HA ) supported by Sun ?
>
>

The second controller does not add functionality, it only adds
redundancy. So, there is no need to get a second controller for
either array.

As for earlier comments about array based vs host based mirroring, in
your current case host based mirroring is a better way to go. Any
time you have things in one physical enclosure you have risk. By
spreading the mirror across multiple arrays you reduce risk. There
are plenty of instances where a single array (this is vendor neutral)
that didn't have a single point of failure goes down. This is usually
caused by bugs in the firmware.

The type of DB failover you want to do is probably the most risky in
terms of having a successful failover. One of the value add features
of cluster software is the cluster file system. If you have locks or
reservations that don't get properly released the second server may
refuse to import the volume.

There are other options though. Most of the options that have more
reliable failovers involve replicating the data from server to server
instead of through a mirror. This would mean that you put one 3310 on
your 280 and one on the 220, with no array doing any multi-hosting. A
list of things that are known to work for this are:
- Sun Network Data Replicator
- Oracle DataGuard
- Veritas Volume Replicator
- Scripted Log Shipping

You should be seeing a theme here. Replicate the updates over the
network to the standby server while it's in recovery mode and have it
read in all the logs, or replicate the volume over the network. By
having each server use independent storage you don't have to worry
about importing and exporting volumes. You do have to make some
decisions on synchronous or asynchronous replication. i.e. does the
commit happen when both servers write the data or only when the
primary server writes it. Your risk is transactions that haven't been
sent and the ones that are in flight. It's all a matter of how much
risk you want to take.

Elias

• Previous message: David J Dachtera: "DNS / Resolution question"
• In reply to: noone: "Re: StorEdge 3310, disk sets, 2 hosts"
• Next in thread: noone: "Re: StorEdge 3310, disk sets, 2 hosts"
• Reply: noone: "Re: StorEdge 3310, disk sets, 2 hosts"
• Reply: Scott Howard: "Re: StorEdge 3310, disk sets, 2 hosts"
• Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]