Here’s part 2 of our DPM 2010 launch week overview
For the full set:
DPM 2010 launch week @ MMS 2010: Part 1: Technical Introduction
DPM 2010 launch week @ MMS 2010: Part 2: Protection Applications
DPM 2010 launch week @ MMS 2010: Part 3: Protecting Windows Clients
DPM 2010 launch week @ MMS 2010: Part 4: Virtualization and Data Protection, better together
DPM 2010 launch week @ MMS 2010: Part 5: Disaster recovery and advanced scenarios
DPM 2010 launch week @ MMS 2010: Part 6: Partner announcements
Let’s continue today with Session 2
Session 2: Protecting Application Servers with DPM 2010
Session two of an entire DPM day (first four sessions were all on the same day so for me it was like being a child in the candy store for the entire day! Not to mention that I had the change to have speak to Jason himself in the evening… :-))
Again this was presented by Jason Buffington. This session explained the entire VSS process and the differences when protecting exchange, sql or sharepoint. So here are my notes:
How does this VSS writer thingy works?
To start, when you decide to protect a workload, DPM will create a replica. This means that it will make an exact copy of the original sources, whether this is sharepoint, sql databases or files. After that, DPM will never ever again make an entire copy of the data.
So what does it do? DPM works with Express Full backups which is block level based and synchronizations which is byte level based. The express full backup is the latest version. All previous versions are the so-called layers.
So after the replica, DPM will create a volume map of the data. Is it large? No, a 0 or 1 for each 120 kb so the footprint is small. Here’s an example of a volume map
So let’s say after one hour, an express full backup will be taken and this is how the volume map looks like
This is what happens:
1.VSS Snapshot taken on production volume to ensure consistent data
2.Cache of changed blocks is sent to DPM server
Important to know here is that the file IO continues, the VSS writer will only “freeze” the blocks that have changed so that the server can continue normal operation! So no more placing databases offline, bringing solutions in maintenance mode… If it has a vss writer, it is all online.
Finally, after the blocks are sent to the DPM server, the VSS writer will release the frozen blocks
In a nutshell, this is how the express full backups work.
But how does the synchronization works? Again, this was explained with an example so here goes:
We assume in our example that we are working with a database
Every xx minutes (depending on your settings) you synchronize the closed transaction logs
In the case that you need to recover, you return to the database express full backup from 0:00 and roll forward the transaction logs till the point in time that you want.
That’s it. So with DPM 2010 you can go to about any point back in time when you want.
Now how many points can you create?
If you perform the following schedule:
you want to be able to return 512 weeks times 7 days (one express full per day) times 24 times 4 (24 * 4 for 15 minutes synch) means 344.000 points in time. This is the maximum but would mean point in time recovery for the last 9 years!
Now here is the joke:
- MS doesn’t want you to recover a SQL 2005 database in 9 years
- It will cost a lot of disk
- It will cost A LOT of disk
(For your information, these are not my words :-))
If you want more information about this mechanism, make sure you check out http://edge.technet.com/Media/DPM-2007-SP1-How-does-DPM-really-work/ or one of our SCUG offline DPM events
You can imagine of course that there are some differences in protecting the different workloads. So here is an overview of the differences
Exchange 2007 LCR (Local Continuous Replication)
What is it? One exchange server with a redundant copy of the database. It can failover to the redundant copy in case of database corruption or when the drive is lost where the active database stands.
DPM will backup the database from the Active Database drive
Exchange 2007 CCR (Cluster Continuous Replication)
What is it? Redundant exchange servers and redundant databases. These can be geo-diverse and the database logs are replicated.
DPM can now backup the active or passive database which you prefer.
You can choose this on a role preferred base:
- Active – most current data
- Passive – least production impact
Or you can choose this on a node preferred base when you are working geo-diverse, then you choose the node closest to the DPM server.
Exchange 2007 SCR ( Standby Continuous Replication)
This is even more intelligent. Suppose you have a DPM server on your main site and exchange SCR. The first DPM server will protect from the passive node. SCR means that it will replicate to a standby node and if you see this picture this means that it need to replicate over the WAN. Suppose that you have a secondary DPM on that other site. Instead of replicating twice over the WAN, DPM is smart enough to do the second protection from the standby node, thus no additional bandwidth necessary.
Exchange 2010 DAG
And finally there is dag, where DPM works with a copy instead of a full backup. This lowers the resources necessary for protecting your exchange environment. See the screenshot
SQL Server Mirrored database
- Mirrors feature redundant SQL servers and redundant databases
- Databases logs are replicated
- Database Failover is automatically recovered
SQL Server Log Shipping
This features one SQL server with redundant databases.
- Each copy is treated as a unique drive by DPM
- Redundant backups require that both drives be protected
- Express Full’s only – no T-Logs
If you are wondering why there are no transaction log backups with this kind of solution, the reason is pretty simple… Never but never let a protection application work with the transaction logs when the system is doing it himself. It would be asking for trouble.
Sharepoint uses a lookup to determine what kind of data is necessary to protect the sharepoint farm, including the content databases, web front end servers and so on.
For 2007 you still need a recovery farm if you want to do a item level recovery but with sharepoint 2010 you can actually do item-level recovery WITHOUT a recovery farm.
In the end, one final question…
Each time you deploy something… How are you going to back it up?