For work, I read a lot of blogs about storage, virtualization, software defined whatever, and other technologies. When I get home from work, I often times turn on the news to see what’s happening in the world. Recently, I felt a little Déjà vu between my work and home experiences. During the day, I read Matt Kixmoeller’s post “PureTruth #1: Is consumer-grade MLC flash reliable enough for the enterprise?” The post is an interesting look at why Pure Storage believes MLC flash drives are the way to go for high performance, high reliability and low cost storage. Later that day, I watched Jay Carney attempt to explain his way through the government’s Benghazi nightmare. I felt like I was experiencing the same tap dance for the second time.
The conundrum that the PureTruth post dances around is “Why does a Pure Storage array using the lowest cost cMLC flash media cost 5-10 times that of a hybrid array from Tegile Systems?” Leveraging the canary in the coal mine metaphor used in Pure’s post, I believe they’ve got the canary in their mouth like an old Sylvester the cat cartoon. It seems odd to me that one would boast about leveraging the lowest cost technology and having the most elegant software to maximize its value, yet be so much more expensive than other solutions.
I will walk through each of the main value points mentioned above:
I have written posts about this point before. There is a corner case set of applications that need the absolute bleeding fastest storage out there. For that segment, all-flash arrays like Tegile’s HA2800 and Pure’s FA series of arrays are great solutions. For the other 97% of the market, a hybrid array delivers far and away more performance than is typically needed. Take for example, Tegile customer Voonami. By going from traditional storage to lower cost hybrid storage, the cloud service provider was able to reduce their time managing storage performance by almost 90%!
The storage industry has been through this one for a long time. If you rewind about 15 years, storage systems vendors dabbled with using desktop-class drives in arrays in place of more expensive SCSI and FC interfaced drives that also had higher spindle RPMs. Desktop drive pundits used the same shell game of math to attempt to prove that high volume components yielded lower failure rates and thus were more reliable. What happened? Due to looser testing and less robust firmware, drives had to get swapped out of these arrays quite frequently. The drive vendors responded with an in-between solution called “Enterprise SATA” or “Enterprise SAS” drives. These drives used lower cost components than their big iron 15K RPM counterparts, but featured lower failure rates due to their more robust error handling in these drives’ firmware. Does this sound familiar? We at Tegile have evaluated SLC, cMLC and eMLC drives, and have found that eMLC provides the best balance of performance, reliability and cost.
This one is the easiest to unravel. All-flash arrays are very good at addressing $/IOP for great performance economics, but are horribly expensive when it comes to $/GB. To address capacity, a purely all-flash array vendor must point to a high capacity “bit bucket” from another vendor to address high capacity. This introduces all the pains of having separate silos of storage to manage and how to move data across tiers in the high velocity virtualized world we live in. Remember the promises of ILM? Yuck. By integrating flash and spinning disk into the same array architecture, the hybrid array’s caching algorithms resolve the data management issue to very effectively put the right data at the right place at the right time at five to times less than an all-flash array. Why not use all of the cool data reduction software found in most all-flash arrays on spinning disk too? We do, and our customers love getting all-flash class performance at less than a dollar a GB.
So, let the tap dancing continue; be it on the floor at VMworld in a few weeks or on the cable news channels. Only time will tell the real truth.