Flashvols – Latency Predictability in a Hybrid World

Working in the enterprise storage space is a lot of fun.  It is a highly competitive industry, where there is a very low margin for error when building solutions for customers.  The challenges we face are not only technical in nature.  Yes, we have to look at the technical variables involved; however, we also have to make sure our technical solution aligns with our customer’s business from an operational and budgetary perspective as well – all while differentiating ourselves within a very noisy market place.  It’s like a fun version of multi-variable calculus, minus the mind-numbing root-canal style pain. One of the key reasons I came to Tegile two years ago was because of the robust feature set built into the foundation of the product.  This foundation allows me to focus on my customer’s problem, instead of polishing up my Ben Kenobi impression to convince them that “these aren’t the feature you’re looking for”.


One area that this comes up a lot amongst our competitors is in the very publicized battle between hybrid and all flash technologies.  The “hybrid” players spend their marketing dollars trying to convince you that they can give you all flash performance while utilizing rotating disk for capacity, while the “all-flash” players spend their marketing dollars trying to convince you that you can actually have flash for the price of spinning disk.  The reality of the situation is that both camps are right – at least when a use case follows their very specific data reduction criteria or very specific IO patterns – and you land somewhere within a neat and tidy bell curve.   Now, if there’s one thing I’ve learned in IT – it’s that what you plan for on paper works out 100% of the time, 20% of the time.   So what happens when you have one workload that fits one curve, and another that fits, well, the other curve?  Let me give you some facts and some examples that I run across often to help set the stage:

  1. For all current flash technologies, raw flash is more expensive than class equivalent 7.2k RPM large capacity magnetic disk on a per gigabyte basis.  Period.  Yes, you can change the game if you have the ability to dedupe the flash and you compare yourself with a technology that does not dedupe the magnetic disk – but even then that only works out if you’re talking about datasets that reduce really, really well.  One of the large players in the space bases their marketing math on a 6:1 data reduction in order for their pricing claims to pan out.  That might be OK when you’re talking VDI – but not so much for your production databases and general application data.
  2. The entire storage industry built their second-generation storage arrays on the fact that not all data is active in the data center.  The basis for “storage tiering” back in the day was that there is a subset of data that is busy and can be kept on 15k storage media, while the inactive stuff, which typically accounts for huge percentages of the whole, can be serviced just fine on 7.2k drives.  These guys were solving for the fact that you deal with the $/IOP differently than you deal with the $/GB problem.   This premise is still true today – only now there is a better approach through caching.  Hybrid plays can essentially service the IO requests through flash (best $/IOP) and capacity through 7.2k RPM deep disk (best $/GB).  We also have the side effect of being able to adapt to changes in workload in real-time verses having to wait for a tiering job to execute, so we can deliver the performance without having to jam the spindle counts into each tier.
  3. If I have a large SQL dataset, say, one larger than the size of all the cache on any hybrid system, and I create reporting jobs that aggressively access that data in a 100% random fashion, my job will run faster on all flash.   This may or may not be meaningful to every business – but I can assure you that the hunger for instantaneous business intelligence will not diminish as kids who grew up with Netflix, Spotify, YouTube and other instant gratification platforms that permeate our daily lives take leadership roles within businesses.

So here is the over-arching problem:  Hybrid is fantastic!  All flash – equally fantastic!  The problem is they are not a best fit for the same workloads.   What if you have all of the above? Traditionally, storage engineering has been an exercise in optimizing for the convergence of compromises that least impact each other.  But that doesn’t have to be the case here.


That’s why I couldn’t be more excited about our FlashVol capability.  FlashVol’s allow you to pin certain workloads into an all SSD pool, which run right along side your hybrid storage pools on the same storage controllers. And, it can be right-sized for the need so you don’t have to over buy just to address the workloads within your environment that would actually benefit from it.  It may sound cliché, but you can actually have your beer and drink it to.

Leave a Reply

Your email address will not be published. Required fields are marked *