Enterprise Storage: Management Goals

Scenario 1 - today's solution

Figure 1: classic recovery avoidance through basic local mirroring with disaster recovery via remote snapshot

The storage management example in Figure 1 looks at what might be expected of today's storage management. The basic assumption is that a mission-critical application suite, Application A, is running on the processor, accessing and updating data held on storage Device D. As the data is mission-critical it would be reasonable to have it mirrored within D. In the (unlikely) event of a disc in D failing and taking some of the prime data with it, the device's own management will switch to the mirror and rebuild the other copy of the data using a spare disc and enabling the failed unit to be replaced with no loss of service levels.

That's fine and, apart from a very short period during the switching and rebuilding, the data is not exposed to a single point of failure.

To give the required degree of disaster recovery (DR) resilience, the system takes a periodic snapshot (by whatever means is supported on Device D) on the remote device, R. Each instance of the remote snapshot can overwrite the previous instance as, by definition, its purpose is to allow recovery to the previous checkpoint/snapshot. This is asynchronous but today's technology means that the lag between real-time and snapshot currency is very short.

Figure 2: protection against hardware-induced single point of failure

Now, let's suppose that D suffers a failure in one of its twinned power supplies (Figure 2). Hardware failures are rare now, but this type of failure is probably the most likely to occur in the field. All the data on D is now exposed to a single point of failure - the remaining power supply. Mirroring within D no longer gives any recovery avoidance cover, so the mirror needs to be switched to Device E until the offending power supply has been replaced, when the mirroring within box D can be reinstated. DR cover is as before.

A SAN would obviously make it simpler to switch mirroring to another device with no loss of performance. In an ideal world, it would be possible for Devices D and E to come from different vendors. Apart from the latter, trapping the power supply failure and using that to trigger the rest of the operations is not difficult. There is nothing to stop today's technology from doing this completely automatically - at least within a homogeneous storage environment - though none of the major storage vendors would appear to take this level of function for granted, let alone include it as part of their current offering.

See the next page, or use the left hand frame to navigate between the three timeframes.

back to top

By entering and using this site, you accept the conditions and limitations of use