Common Sense about Commodity SSDs

November 8, 2018

When we call something a “commodity,” we all know what that means: it’s cheap. Widely available. Possibly even disposable.

Storage hardware vendors have long considered SSDs a commodity, and with good reason. Falling prices have allowed arrays to be over provisioned to boost performance and endurance specs. When a customer needs those numbers to be high, and who doesn’t, simply pack in a lot more drives, and let them pay the inflated price.

On the surface, this seems to make economic sense…for the vendor. For the customer, it incurs waste, it’s inefficient, and it’s unsustainable.

More importantly, different data center applications have different requirements. Heavy database or transaction processing needs write latency that’s as close to zero as possible. A provider of hosted/cloud apps needs consistent, reliable uptime for a positive user experience. Each workload poses its own challenge, and throwing more drives in the boxes does not address the challenge, nor does throwing more and more boxes at it.

These are not one-size-fits-all scenarios, and as we already discussed, today’s users need to stay agile to be able to meet different challenges tomorrow. Commercial SSDs intended for the mass market are not designed to do this, or even to meet the requirements of each workload in use today. Instead, they’re designed to be fairly disposable.

Because we use programmable controllers with commodity Flash, we have the ability to address the waste, inefficiency, and unsustainability that is proliferating in storage. We can tune or optimize the SSD profiles for different environments and workloads. For example, tuning for traffic management with multi-stream QoS gives more consistent performance. Tuning for capacity or density gives us more than 100 TB with a single controller.

If we tune for better read/write leveling, we save significant wear and tear on the drives, and we get longer lifespans and reliability. This can be a big savings. Because programmable controllers are also reprogrammable, those drives can be re-optimized and repurposed in the future instead of being crushed before their useful life is over.

Another advantage is we can source different Flash with different inherent characteristics or pricing – and exploit the benefits of each flavor even further.

Yes, that includes low-cost, consumer Flash even for Enterprise applications, because our logic and firmware can be tuned to get the performance from that Flash well beyond its specified capabilities. In fact, we’re big fans of using the lowest-cost flash possible for a specific application, because programmable controllers let us do that.

Hyperscale infrastructures in particular need to realize these savings. Keep in mind we have users whose flash storage investment is greater than a billion dollars, so to save tens or even hundreds of millions a year over the life of the storage is a significant business advantage.

For array and hyperconverged appliance vendors, the value proposition is different, but still meaningful, because new systems can be brought to market much faster, bespoke systems for key customers can be developed easily, and all at a lower cost.

Relying on commodity SSDs because they’re cheap and available is tempting, but it’s not the smartest way to build flash storage infrastructures, nor is it the smartest way to get the benefits SSDs have always promised. But with programmable controllers we are getting there today.

Tags: , ,