Transactional consistency - snapmirror sync / semi sync

[nsitps1976]

Is it fair to say that SME / SMSQL etc would not be needed with snapmirror if using snapmirror sync or semi-sync as replicas are not async point-in-time snaps. With sync or semi sync writes to disk are affectively captured on the destination disk... ???? 

This is assuming SME etc is only being used to provide consistent replicas and is not used as a backup product...

[radek_kubka]

Hi,

I would say this assumption is a bit risky - it follows the logic of other storage vendors who say preserving write order is good enough to be able to recover from a remote replica.

To me it means that in the event of an actual sudden outage mirrored volumes are crash-consistent, as (presumably) none of the application at the primary site has been put in a hot-backup mode prior to outage. Depending on your luck you may, or may not be able to restore services. If you are unlucky & things go pear-shaped, there are no application-consistent recovery points.

For that reason I think SnapManager products (or scripted equivalents) are still useful, regardless of replication mode.

Regards,
Radek

[nsitps1976]

So how can you protect databases real time if they will always be crash consistent? Scripts or SM products are executed at scheduled times so seconds  after this time the data has exceeded the RPOs?

Also, does sync and semi sync also transfer snapshots like  async?

Thanks

[radek_kubka]

Well, it is a very good question.

I often argue that Continuous Data Protection (CDP) for databases is a bit of a fiction, as you need some checkpoints / markers which represent consistent state of a database.

You may get zero RPO with synchronous replication if you are lucky - if you are unlucky, you need to roll back to the latest checkpoint.

It also varies between different database technologies, e.g. Oracle should come back no matter what (there is a TR describing 'dirty' NetApp snapshots as valid Oracle backups).

For MS SQL synchronous replication of logs with async replication for databases is probably as close as you can get to a near-zero RPO solution (another TR describes this in detail).

Regards,

Radek

[nsitps1976]

Looking for an RPO of 30 / 60 seconds for SQL - What would you do? RTO is not as tight as RPO so this is not an issue and we have this covered. Do you have the TR# re SQL Sync as I can't seem to find it...

[radek_kubka]

This is the TR:

http://www.netapp.com/us/library/technical-reports/tr-3604.html

It actually proved me wrong , as the solution utilising sync SnapMirror does not involve SMSQL.

It does not ignore potential problems with consistency though & proposes a workaround in Appendix B:

In case of problems attaching the database at the DR site caused by checkpointing activity, please follow the procedures below. More about checkpoints may be found at the following link: http://msdn2.microsoft.com/en-us/library/ms189573.aspx.

And the final comment:

The above-mentioned workaround may result in data loss subject to the recovery.

Regards,

Radek

[nsitps1976]

Great, I will take a look over this.