NetBackup 7.6 and VMware Accelerator

By Juan Orlandini
2/25/2014

In my last blog, I touched on the benefits of the newly released Symantec NetBackup 7.6. Here, I want to get specific and describe the VMware Accelerator feature.  
Reducing the need for full backups
By now you know that we essentially can do a full backup once and incremental forever after that, yet still create periodic full backups. What's happening under the cover involves several key technologies. First, NetBackup figures out what needs to be protected via the VMware VADP API. That API is what allows a backup product to protect VMware hosted virtual machines. As part of those APIs, there is a feature called Change Block Tracking (CBT). This is designed so that the backup application informs VMware that a full backup was run and for VMware to keep track of what blocks change until the next backup. In traditional backup approaches, this makes incremental backups much more efficient. The backup application does not need to enumerate what's changed. It merely has to request from VMware for a list of the blocks that have changed and then capture those. What's cool in NetBackup 7.6 is that although it is only capturing changed blocks, it can catalog the individual objects (files, emails in Exchange, objects in SharePoint, and database elements in MS-SQL). This is true even without Accelerator. With Accelerator, the difference is that the need for full backups is greatly reduced or even eliminated. (I've said that before, but it bears repeating).
Intelligent deduplication
The way that works is that NetBackup leverages the ability of intelligent dedupe arrays to create Optimized Synthetic Backups (OpSyn). This is a feature present in NetBackup's own dedupe technology (MSDP and PDDO) as well as some third party vendors (DataDomain, Quantum, etc.). Prior to this technology being available, synthetic backups relied on a very cumbersome and I/O intensive process. To create a synthetic backup, the last full backup would be copied in total, and then each and every incremental moving forward would be laid on top of it to recreate what a full backup would have looked like. As you can imagine, this could take quite a while and potentially consume a ton of space. Fortunately many vendors developed the ability to leverage their dedupe technology to optimize this process (and hence the name). What happens in these arrays is that the dedupe engines need to maintain a database of pointers to where the data really sits. To do this, they need to be very intimately aware of how NetBackup writes information (TAR and other things). Because they already know this, it's relatively simple for them to simply manipulate that database and recreate the equivalent of a full without moving any of the data. Because there is no movement, this also happens extremely fast (the equivalent of many GB/s of throughput).

Accelerator is thus the fusion of VMware API's, NetBackup's ability to protect and restore VMs at various granularity levels, and the back end storage's ability to efficiently create synthetic backups. Notice I said “restore” in that sentence. This is a key point. Accelerator backups give you the exact same restore functionality as traditional VMware backups -- individual files, individual objects, and entire VMs. Oh and you can also use VMware Instant Recovery with those as well. What's that you say? That's my next blog post.
Using Accelerator today
And before I conclude this one, a few words on where you can and cannot do Accelerator today. I mentioned above that you can do Accelerator to both MSDP/PDDO and third party arrays. That's mostly theoretical right now. Symantec has released the APIs to the third party vendors, but none have made it to the Software Compatibility List yet. In the case of Data Domain, that appears to be merely a function of finalizing some testing. The other vendors are a bit behind, but will be releasing this functionality over the next … ahh … soon. Check with the SCL, your device manufacturer, or ask Datalink. We have most of them in our Datalink on Demand Labs and can give you the results of our own testing. In the meantime, if you are eager to play with this, focus on MSDP/PDDO. We'd love to hear what your experience is like.