FalconStor, NetApp lead the way on data protection

If there's any easy way to tell products apart in the iSCSI SAN server space, it's in the support for advanced data protection features like snapshots and replication.

If there's any easy way to tell products apart in the iSCSI storage-area network server space, it's in the support for advanced data protection features such as snapshots and replication.

In theory, snapshots encompass a simple idea: you instruct the storage array to make a point-in-time copy of the state of a virtual disk. You can continue reading and writing to the virtual disk, but if you so desire, you can also go back to the snapshot to see what things looked like at the moment you captured it. That said, every vendor whose product ships with this capability (only D-Link's DSN-3200-10 and Nexsan's SATAbeast don't) takes great pains to document that their products aren't actually making a copy of the virtual disk, but rather, are simply tracking the delta between data versions. If you actually do want to keep a snapshot forever, that process may require some additional work to either copy it to a new volume or it from the parent volume.

One easy differentiator among the products tested regarding snapshots is operating system and application support. It is dangerous to have the storage system simply pick a moment in time to create the snapshot because the iSCSI storage protocol has no sense of files or directories. Depending on what is happening at that moment, the file system may or may not be fully self-consistent. There are lots of bits and pieces that point to each other on a disk, and if you catch a snapshot when one pointer is updated, but not the other, then the disk may not be 'legal' anymore. The problem compounds itself with certain applications, such as e-mail and databases, where consistency and a true or full backup may require multiple virtual disks to all be in synch at the same moment.

If you plan to make snapshots of disks that aren't in active use, then this will not be an issue. However, if you want to guarantee that a snapshot is fully self-consistent and can always be used in place of the original disk, you need to verify that your iSCSI SAN server has some sort of agent that can communicate between the operating system or application and the array to ensure consistency at the moment of snapshot.

The leader in claimed agent support for snapshots is FalconStor's NSS-S12, with specific support for more than a dozen databases and e-mail servers, along with the most common Unix and Windows operating systems. Unfortunately, our testing of FalconStor's snapshot capability with Windows 2008 showed a particularly pernicious bug in the product: it said that we were getting consistent snapshots, when we actually weren't. Eventually, technical support weighed in that Windows 2008 Server wasn't supported quite yet. While FalconStor's technology worked great with all of our other tests, this highlighted one of the difficulties of agent support.

Close behind FalconStor is NetApp with agents for Windows and Unix operating systems, as well as some application support. We tested NetApp's Windows 2008 agent and were able to get a consistent snapshot.

Every other vendor participating in this test was less ambitious about tackling the snapshot consistency issue with agents as they offered no real agent support, or support for a specific Microsoft operating system feature (introduced in Windows 2003) called Volume Shadow Services (VSS). Compellent, Dell, HP, LeftHand Networks and StoneFly simply all point specifically at Microsoft for operating system level consistency, and are silent in their documentation when it comes to Unix operating systems or VMware's ESX server (a predicted heavy user of iSCSI resources in the future). We tested snapshots to be sure that we could indeed make one with all of these products, advertise it as a new iSCSI volume, and retrieve files from it. We did not run into any problems with these simple requests.

One common snapshot feature across products was the ability to schedule regular snapshots to occur. SAN servers from Celeros, Compellent, Dell, FalconStor, HP, LeftHand Networks, NetApp and Stonefly all support scheduling of periodic snapshots. We also found some differences in what you can do with snapshots, such as how easy or hard it was to split off snapshots as separate virtual disks (Compellent's StorageCenter was particularly good at this); revert a virtual disk to a snapshot image (the NSS-S12, HP StorageWorks 2012i, and LeftHand Networks NSM 2120 made this easy to do); or, use a snapshot automatically as part of a backup strategy (only the Kano Technologies NetCOR 7500 made this difficult).

Snapshot functions offered by Compellent, FalconStor, HP, Kano and NetApp require an extra license. (Kano said they were going to stop charging for snapshot licenses in August 2008.)

Most iSCSI servers require you to make some allocation of space or prediction of how much space the snapshot will use; a few including Compellent's StorageCenter and LeftHand's NSM 2120, do not. Whether this matters to you depends on why and how you're making snapshots. Similarly, most have limitations on the number of snapshots you can have for a particular virtual disk. The HP StorageWorks 2012i actually licenses the number of snapshots you can have. For others such as the NetApp FAS2050 with limits, the number is fixed, often to a very high number (such as 255 per system).

Replication variations

Replication is another data availability feature that you can use to differentiate iSCSI storage systems. Most often, replication is used to keep a copy of a virtual disk on another storage system, with the idea that you'd physically locate the second server in another building or, possibly, in another country. As with snapshots, basic functionality is present in many products, but there are many variations, such as data de-duplication and bandwidth-limiting capabilities.

We divided replication into two categories: synchronous and asynchronous replication. With synchronous replication, every write to a virtual disk is mirrored to the replicated volume, which means the replicated volume is guaranteed to be always consistent with the original volume. It's called "synchronous," because the actual write operation is not signaled as completed until both copies are in synch. Synchronous replication is a touchy subject because it requires high bandwidth and low latency between the two storage systems. Otherwise, performance will be heavily impacted.

Celeros, Compellent, FalconStor, LeftHand Networks, NetApp, Reldata and StoneFly all support synchronous replication. FalconStor and NetApp offered the greatest options in supporting synchronous replication to both the same storage subsystem and to a remote subsystem. Reldata had the most interesting option: its product will replicate traffic using iSCSI, leaving the possibility that you could use a non-Reldata system for the remote storage server.

Asynchronous replication is a much looser form of replication and is often tied in with snapshots. The idea is that you take regularly scheduled snapshots of a virtual disk, and those snapshots are used for replication. The replication process sends only what changed between the last snapshot and the current one. Asynchronous replication doesn't significantly affect performance, because it doesn't get in the way of every single write. Thus, if you snapshot once an hour, then your replicated virtual disk will be about an hour out of date (depending on how long it takes to transfer the updates). Asynchronous replication is supported by Celeros (remote), Compellent (local/remote), Dell (remote), HP (local), LeftHand Networks (local/remote), NetApp (local/remote), Reldata (remote) and Stonefly (local).

As with snapshots, replication is often a separately licensed feature. Because we didn't test replication (we only had a single one of each storage system in our test), our pricing does not include any costs for replication.

Although we focused on iSCSI in our testing, three of the systems from Celeros, NetApp and Reldata, also include network-attached storage (NAS) protocols (such as NFS or CIFS/SMB0 in the same systems). Celeros caught our eye on the data recovery and data protection area because it has pre-installed agents for popular enterprise backup applications from Veritas, Dantz, and CA for use with the NAS part of their system. This lets the Celeros server itself handle backups, a slightly more efficient approach than having a server which has NAS volumes mounted do the backups.

< Return to test intro: NetApp, Compellent, HP, Dell top the field in 12-product test >