The proclivity of vendors to emphasize jumps in performance is almost as commonplace as the proclivity of IT veterans to view such claims in performance gains with high degrees of cynicism. So when a system administrator recently went on the record during a recent round table and said that he experienced a 4 – 5x improvement in testing and a 2x improvement in his production environment using Symantec’s OpenStorage (OST) API for backup, it caught my attention.
Unlike other testimonials I often hear, this one particularly piqued my interest because the user in question, High Point Regional Health System’s System Administrator Richard Nosal, was already using disk as a backup target in both in his testing and production environments before he began using OST. So Nosal saw these performance improvements not going just from tape to disk, as is so often the case in other stories I hear, but going from one type of disk target (NAS in testing and VTL in production) to another disk target that was using OST.
To try to get a better understanding of why OST delivered such substantial performance gains for this user, I first spoke to Symantec’s Paul Sustman, Veritas NetBackup’s Product Manager. He explained that the main difference between OST and other types of interfaces is that other interfaces are only for data movement. As such, they do not allow for the type of integration of other features that OST provides as they have no control path. OST added a control path so Symantec’s partners can leverage OST to do both control and data transport when sending and receiving data.
Another big difference is that other interfaces (NAS, VTL) used for disk-based backup targets are not ideally suited for disk-based backup. In the case of disk devices that act as a VTL, they appear as a tape device so data is fed to them sequentially even though the disk sitting behind them is still a random access device. In the case of primary storage devices that handle NAS-type data, they are well suited to handle small files. However backup data contains large amounts of data in large blocks that require a tight window of time to move it. Sustman says, “The data environments between primary and backups environments when using disk are much different, so the requirements are much different. This is what OST was designed to address.”
Using OST, partners can write a piece of software that is installed on the NetBackup Media Server which is used by NetBackup to talk to the partner’s device. The upside here is that the OST partners can tell NetBackup the optimal method to talk to their device using its new control path. Since there are a lot of different interfaces and protocols supported by NetBackup, it is really up to the OpenStorage partner as to what interface and protocol they use. This flexibility allows NetBackup to work best with almost any device since the partner is in control of what protocol is used between NetBackup and its device.
Symantec’s partners can tell NetBackup to use the protocol that will deliver the best throughput for their device as well as perform tasks such as selecting the appropriate size packets for the buffers on their devices so they perform optimally. As a result, users should see substantial increases in backup performance on the devices of Symantec’s partners versus using native VTL or NAS interfaces.
After talking to Paul at Symantec, I then circled back with Data Domain and FalconStor who also participated in the recent round table to get a better understanding of how they were leveraging OST to improve backup performances with their environments. In the case of FalconStor, it is using block protocols such as Fibre Channel and iSCSI in conjunction with OST to optimize performance between the NetBackup Media Server and its VTL.
Data Domain took a slightly different tact. It adopted its own proprietary protocol for communications between the NetBackup Media Server and it disk solutions to accomplish two objectives. First, it uses its own proprietary protocol to eliminate the protocol handling that is necessary on other interfaces such as NAS and VTLs. Second, Data Domain was able to introduce parallelism into the process so it could both use more threads and package the data faster on the NetBackup Media Server and then unwrap the data more quickly on its solutions. Data Domain’s VP of Product Management, Brian Biles, says, “Because we own the client (plug-in on NetBackup Media Server), we can optimize client behavior according to what we have in place.”
The 2 – 5x improvements in performance that System Admin Nosal saw in his backup environment is likely not to become isolated incident because of the freedom that Symantec gives its partners to select the interface and protocol that is best suited for their storage device. But as further investigation revealed, not all of Symantec’s partners are implementing OST in the same way and organizations such as Data Domain have taken some extra steps that enable it to optimize the performance of NetBackup OST for its platforms. So as users consider the use of OST and the benefits it provides for expediting backups, they are advised to look beyond just if Symantec’s partners provide support for OST but also how they implement it as how Symantec’s partners implement it could result in significant variances in the performance of backups in their environment.