Last month, Juan wrote a blog on the five fingers of backup administration. One of those fingers was storage. Storage means several things when it comes to data protection: the backup of the client’s disk, the disk being utilized by the master server to hold the metadata (catalog), the place you're holding your backups (disk storage pools in NetBackup parlance), and several other things. In this post, we'll look at measuring the performance of the disk you use to conduct backups.
You can use several interconnects to talk to your disk: local disk, fast fibre channel links, fast FCoE/Ethernet links, and even slow network links to a remote location. You can also have a combination of fast (SSD, SAS, FC) disks and slow disks (SATA). On top of that, you have the device those disks connect to: fast array controllers, dumb array controllers, dedupe controllers, virtual tape library devices, hosts, etc. Understanding how these interact with each other and determining the optimal configuration for your environment can be a significant challenge. Fortunately, there are tools that can help.
One generic open source tool is Iometer. Originally developed by Intel in the late 1990s, this free tool can create workloads to storage devices. To eliminate variables on the source computer that runs the tool, the data generated to send to a target storage device is stored in memory. Iometer is available for Linux, Solaris, Windows, Mac, and NetWare.
Another tool that is backup product-specific is GEN_DATA, which has been available in NetBackup since version 6.5. This little known testing and problem isolation tool can be used to test disk or tape performance for varying workloads, data size, number of concurrent streams, and as of NetBackup version 7, deduplication for both backup and restore. It uses a standard NetBackup policy, where you add directives to "backup selections" to generate the data stream(s) as seen in the example below.
Other features of the tool include variables for testing deduplication. I would not use this feature to test dedupe ratios between vendors’ products since everyone’s data is different and the results would be difficult to relate to actual data testing. Only your data provide an accurate measure of deduplication ratios. Reference this technote for additional detail on different directives in the backup policies.
These two tools are useful for fine tuning your backup environment or isolating a problem. When you use these tools on the master/media server, you are effectively measuring the backup target performance based on the connection speed to this device, as well as the ingest ability of the target backup device. If you have a target backup device capable of terabyte per hour of backup performance, but your connection to this device is capable of gigabytes per hour, the slowest component dictates overall performance. The number of concurrent backups run with these tools will best stress the connections.
In general, Iometer gives you the performance of your devices independent of the backup software. However, it’s limited to testing disk only. NetBackup’s GEN_DATA facility is cool in that it can be used to measure disk, VTL, dedupe, and even tape performance. But, it also includes the overhead of the backup software – which is hard to isolate. The combination of these tools should give you a really compelling tool set to evaluate your environment’s disk devices.