At HDS, we’re pretty bullish on the future of flash and we think we’re very well positioned for the coming transformation. That's why we launched our new all-flash array family, our Virtual Storage Platform (VSP) F series, as well as our next generation flash module drives (FMD). Our flash solutions have proven themselves in the largest flash data centers in the world. Continue to follow us as we build out both our flash systems portfolio and, just as important, our software defined approach to optimizing the platforms.
But we hear a lot things stated about flash that need to be called into question. Just because something is repeated over and over again, doesn’t necessarily make it true. So with apologies to Adam Savage and Jamie Hyneman, I want to offer a couple of unsolicited ideas should they want to produce one more MythBusters episode.
Myth #1 – Our IOPS prove that we’re the fastest
Yes, IOPS can be fairly easily measured on synthetically generated workloads in a lab environment and all vendors, including HDS, will publish IOPS results on their data sheets. It makes for an easy way to compare different systems. But the fact is that few applications will come anywhere close to requiring the IOPS performance that even the “slowest” enterprise flash systems can achieve.
We see fast response times, measured in latency, as far more important. When transactions are executed faster, when boot times are reduced, or when analytics can be done in real-time speed, the value of flash is realized with tangible business benefits. And while flash storage has far faster response times than hdd storage, it can be subject to random and unpredictable spikes in those response times. Unpredictable spikes in wait times can be super annoying or even the cause of a catastrophic failure. Imagine an IT system designed to track a transportation network. If the flash storage in that IT system were to lose track of the geospatial location of all of the transport vehicles it monitors due to response time spikes, then truly bad things could happen.
While every all-flash array is likely to be more than “good enough” in its ability to deliver IOPS, the same can’t be said about delivering consistently fast response times. Our VSP F and G series with flash modules will off-load flash functions like garbage collection to the flash devices so the storage controllers won’t struggle to respond to IOs quickly while also handling garbage collection activities. The flash modules take this several steps further by off-loading functions from the storage controller like block/page mapping and contiguous block write avoidance. The result of this is that all of our flash systems can deliver consistent sub-millisecond response times to IOs even when capacities fill up or IOPS loads approach their max limits. You can read much more about what makes our flash modules so unique here.
Myth #2 – Everything should be on flash
Let’s face it. Data ages but it doesn’t get deleted even when it’s well past its useful life. That power point presentation you created in 2011 for a quarterly business review is likely still sitting on a network filer somewhere. Some data needs to legitimately be kept for legal or compliance reasons. But other data lives on and on because it’s nobody’s job to delete it.
As rapidly as the price of flash has been dropping, it isn’t nearly as inexpensive as nl-SAS. Furthermore, data on enterprise all-flash arrays are typically protected with periodic back ups so there may be multiple copies of that old power point in existence. It’s far more economical to move aging data that hasn’t been accessed in months from flash to nl-SAS. It might be even better to move that data to a back-up free content platform with an automatic retention policy.
Myth #3 – Legacy architectures weren’t designed for flash
At HDS, we’re all about innovating for the future. Our portfolio of 350+ flash storage patents is just one example of our commitment to innovation. We’ve worked for decades to develop storage technologies that are suitable for a modern data center. While newer flash architectures have come to market in the last few years, we see that they lack the resiliency, security or scalability to move much beyond tactical deployments.
We know that customers want more and not less from their enterprise storage in these regards. Here are some of the questions we have about the emerging all-flash array designs and their ability to enable the digital transformation objectives that customers are working to achieve:
- Can data be migrated without an outage to the operations?
- How will the data be protected? Is it application aware?
- What happens to data that’s aged and no longer needs to be on flash?
- Does the management platform scale to handle petabytes of flash? How well is it automated?
- Is the platform capable of a cloud services delivery model?
- Can the data be securely and completely erased, with verification, from devices in a system that are: a) at the end of their useful life, b) suffered a failure, or c) completed a proof of concept evaluation?
So with these three common myths about enterprise flash storage officially BUSTED, we can get back to working on building the next generation data center that’s real-time responsive, exceptionally scalable, fully automated and always-on for cloud delivered services.