The death knell for RAID?

Ingenuity in storage is evolutionary — but clustered storage could be the revolution that replaces RAID

Defining the single-most significant change in storage I have seen throughout the years could easily lead to multiple discussion threads. After all, technological achievements are plentiful. Take the emergence of SATA and SAS, for example; or the staggering increase in capacity of a wide range of storage media; or the ongoing acceptance of Ethernet as an alternative to Fibre Channel for block storage connectivity.

The list could go on almost indefinitely.

Yet examining the individual tiles of the increasingly complicated storage puzzle more often than not gives undeserved prominence to technical details that are usually short-lived. Top-of-the-line not too long ago, 30MB disk drives are best used today as paperweights. Today’s 1TB drives will soon become equally obsolete.

The same can likely be said for just about any other cutting-edge product du jour: None will likely be a watershed for storage, just a step along its evolutionary path.

After all, saying, for example, that the increase in FC transport speed from 1Gbps to 8Gbps is the hallmark of storage evolution in the past few years is a lot like describing the progress of human ingenuity by the speed of our automobiles. Not the best indicator — unless you work for Ford, that is.

For my money, the most significant change is one that hasn’t happened just yet: The rise of a bona fide alternative to RAID.

Clustered storage holds promise. But there’s still a ways to go.

Why do we need an alternative to RAID? Because this pillar of data reliability can’t keep up with today’s demands for performance and reliability.

Steve Todd raises some intriguing points about RAID in a recent blog post:

I discovered it was about two things: (1) Performance, and (2) Data Integrity in the face of disk failures …

He is absolutely right.

Yet, RAID 5 can’t survive a simultaneous failure of two drives, which does not provide a sufficient reliability guarantee for arrays using large media, for example. RAID 6, aka dual parity, can handle that scenario but still may not provide sufficient protection.

Recently, I ran a series of tests of SUN ST5800, alias Honeycomb. Details will be available in an upcoming review, but for the purposes of this discussion, I’ll just say that after I abruptly pulled out eight of its 64 drives, Honeycomb survived without losing data. No RAID system can compete with that.

Moreover, multi-node solutions such as Honeycomb speed up normal access taking advantage of multiple pathways, and they shrink rebuild times with a cooperative effort in the event of hardware failure.

Honeycomb is meant to store fixed data content, but products such as Isilon’s or EqualLogic’s clustered solutions offer a similar multi-node resilience and speed for other requirements.

Is RAID dead, or does it have any future? You can bet that vendors will keep pushing traditional RAID solutions for many years, but depending on your requirements you may want to start thinking clustered, multi-node storage for your next replacement cycle.

What do you see as the most significant change on the storage horizon?