Re: Best Practices on Writes

You're right... It's only a "cluster in the sense that two servers are
"connected via software and hardware......

The software we are using is a highly rated (and not inexpensive) package
called LifeKeeper by SteelEye. The mirroring (called Life Keeper Data
Replication) is accomplished via a crossover cable on a 10.10.100.x private
network. As I understand it, writes are done simultaneously to both drives.
The target drive is locked and is not accessible on the target machine.

LifeKeeper itself monitors a "heartbeat" and if it detects that the primary
machine is off-line, it will automatically pause the mirror, reverse it and
start all the services as well as switch the IP address. This is a protected
IP address (each machine has it's own "real" one) that is the one used by
all applications to access the server. It accomplishes this in less than 5
minutes.

I have asked them for a "best practices" statement about how they recommend
setting it up.

By the way, this is *far* from a "low cost DR" package. It has been working
extremely well for over a year and has saved our bacon several times.
However, what has changed is that the current servers use standard SCSI
(non-RAID) drives and the new servers have RAID arrays with cached adapters.

--
Regards,
Hank Arnold

"Andrew J. Kelly" <sqlmvpnooospam@xxxxxxxxxxxx> wrote in message
news:uUOzSmNpFHA.1416@xxxxxxxxxxxxxxxxxxxxxxx
> Be careful of the term Cluster if it is not an actual Windows Cluster<g>.
> Not sure what software you are using to do this mirroring but I would have
> my doubts as to how effective that is if it is asynchronous. And yes this
> would affect the use of Write caching if the software is getting the data
> directly from disk when it does it's updating to the standby server. If
> the server fails over to the standby you don't have any ability to recover
> those writes in cache. If the process is asynchronous and is at the block
> level I have my doubts it is a viable solution at all. But I have no idea
> what you have and might be missing something. If you turn off write back
> caching altogether you minimize this risk but you can loose a great deal
> of performance as well depending on your load and the capability of the
> disks. This sounds like a low cost DR solution that may end up costing you
> more if it doesn't work<g>.
>
>
> --
> Andrew J. Kelly SQL MVP
>
>
> "Hank Arnold" <rasilon@xxxxxxx> wrote in message
> news:exZ6xNNpFHA.4088@xxxxxxxxxxxxxxxxxxxxxxx
>> No. Machine is not clustered. It *will* be part of a High Availability
>> cluster where the backup machine has all the SQL services in manual mode.
>> The primary machine is actively mirroring (asynchronously) its D: and E:
>> drives to the backup machine's equivalent drives. If the primary machine
>> goes off line, the HA software detects this and puts the services into
>> service and switches the mirror roles.
>>
>> In that case, do you think it would make a difference?
>>
>> TIA,
>> Hank Arnold
>>
>> Andrew J. Kelly wrote:
>>> You might want to have a look at this and the associated links within:
>>>
>>> http://www.microsoft.com/technet/prodtechnol/sql/2000/maintain/sqlIObasics.mspx
>>>
>>> Is this machine clustered? I assume not. If not and the cache is
>>> battery backed it is usually safe to enable write back caching. Most
>>> modern controllers will hold the writes until the drive is available
>>> again and write them once it comes back up. The log writes are flushed
>>> thru the cache when written and are the key to ensuring you can recover
>>> in the case of a disk failure. Even if all the data changes were not
>>> written from the cache the database should be able to redo or undo the
>>> changes when it comes back up and goes through recovery. If the disks
>>> are really toast and the whole array gets fried then you have to restore
>>> from last good backups anyway and the cache doesn't even come into play.
>>> Backups are always your best defense for hardware failures. So make
>>> sure you never put your backups on the same array as the data or logs.
>>>
>
>


.

Relevant Pages

  • Re: Best Practices on Writes
    ... Be careful of the term Cluster if it is not an actual Windows Cluster. ... Not sure what software you are using to do this mirroring but I would have ... If not and the cache is ... >> last good backups anyway and the cache doesn't even come into play. ...
    (microsoft.public.sqlserver.setup)
  • Re: Best Practices on Writes
    ... It *will* be part of a High Availability cluster where the backup machine has all the SQL services in manual mode. ... The primary machine is actively mirroring its D: and E: drives to the backup machine's equivalent drives. ... The log writes are flushed thru the cache when written and are the key to ensuring you can recover in the case of a disk failure. ... If the disks are really toast and the whole array gets fried then you have to restore from last good backups anyway and the cache doesn't even come into play. ...
    (microsoft.public.sqlserver.setup)
  • A Cautionary Tale - About Backups
    ... Unfortunately in playing with it I had fiddled with drive letters not realising the backups were scheduled. ... What it had done was mirrored the 64MB or so on its HD first to RAID Array 1 and then to RAID Array 2, i.e. it had done exactly what it was supposed to, deleted anything from the RAID arrays that was not on its own HD. ... There must be a lesson to be learnt, I suppose a backup rather than mirroring the drives would have prevented this happening. ...
    (uk.people.silversurfers)
  • Re: Expert opinions on SATA, IDE drives, etc.
    ... also have several USB external drives, 100GB and 200GB, for backups. ... I collect a few opinions and facts on the old vs. new drives (sure, ... 2a) Is it that 2 SATA drives can be used to parallel-record files, ... I think what you are talking about is a RAID 1 (data mirroring) ...
    (microsoft.public.windowsxp.help_and_support)
  • Re: Server Advice Wanted.
    ... > RAID 0+1 will be more efficient (stripping and mirroring), ... which is why you see it less than RAID 5. ... > could still restore from backups because it takes day to do so, ... a network backup solution. ...
    (borland.public.delphi.non-technical)