jrohan

Storage Getting You Down? Wipe Away Those Tiers...

Blog Post created by jrohan on Nov 5, 2014

In my role as a Systems Engineer in the UK, I speak to customers and partners on a daily basis about storage requirements and how best to meet them. The storage industry has gone through many cycles of technology change over the last two decades, evolving from the simplest direct attached arrays to today’s powerful hybrid offerings. As you’d expect, two decades of innovation and progress have launched numerous terminologies or “buzz words” into the headlines. All too often, these buzz words drive misunderstanding or misconception, and result in companies buying into inadequate solutions. As customers scramble for the truth, vendors can be all too quick to give their perception of the topics. Technical marketing departments can do a fantastic job of turning a weakness into a “must have” feature! 


It must have tiering capability…!

In almost every storage requirement I see, the word “tiering” now appears. Customers have been duly informed by their incumbent vendors that tiering is required to take advantage of the latest hardware innovations. Without tiering you won’t be able to add new disk technologies, and you’ll soon hit the limits of your SAN. In effect tiering is promoted as a positive feature. 


The shocking truth however, is that tiering is not needed by the customer. It is the storage vendors themselves that need it because, without tiering, there is no way for them to introduce new disk technologies without increasing the management overhead of their aging storage offerings. The problem arises because the perfect disk, one that gives high performance and high capacity at a sensible price, does not exist. Instead at the extremes we have: 

  • SSD drives that deliver high performance but are limited on capacity (poor GB/$)
  • SATA drives that deliver high capacity but are limited on random performance (poor IOPS/$)

Tier.png

Now I’m not going to delve into the deeper mechanics of tiering as this is covered in many other articles, but the point of tiering is to move inactive or “cold” data from the performance layer to the capacity layer. An opposite function also exists to move data that has become active or “hot” from the capacity layer back to the performance layer.



The idea here is great. Until you ask the question “how does the system know what data is hot?” Therein lies the flaw, because it simply doesn’t.



So tiering has been positioned as a clever fix to a limitation in storage arrays. The ground work for most storage filesystems was done in the ‘90s and today’s arrays are by and large running updated versions of old code. Back then, SSDs didn’t exist in the data centre and we operated a more rigid design for our storage needs. There were still multiple disk types, but the answer at the time was to bind multiples of any one disk type together and deliver a volume of defined performance and capacity to a host.

 

Fast forward to 2014 and we now have virtualised environments making huge performance demands of our storage arrays. At the same time, we live in a society that frowns on the deletion of any data and many industries have legislation governing their requirements to maintain stored data for years at a time. In summary we need more power, more capacity, more agility but at a lower cost. This sounds like a job for that perfect disk!


So who has the answer?

In recent years a number of start-ups have tried to attack the previously accepted shortfalls of the storage industry. They have each had their own approach to solving customer requirements, and a few have completely side-stepped the requirement for tiering through some ingenious engineering work. None have accomplished this better than Nimble Storage.  Nimble’s founders started with a blank canvas instead of building upon an aging architecture, and in their development of the CASL filesystem, achieved something that was previously unattainable – the decoupling of performance from capacity in storage arrays. In essence what we call "Adaptive Flash".

Again, the purpose of this article is not to delve into the inner workings of any technology, so if you’d like a deeper understanding of CASL I’d highly recommend checking some of the material referenced at the end of this article. However to see how Nimble completely avoids the need to tier data, it is important to understand what CASL does at a high level.

In summary, CASL translates random writes from workloads into sequential disk IO (the burden of random read IO is taken on by flash). In simpler terms, CASL takes the native data pattern from your application, and turns it into a form that spinning hard drives can work most efficiently with. In doing this, Nimble has been able to completely invert the problem – it is now possible to get outstanding performance from an apparently slow, high capacity disk. The benefits extend even further, as there is now a significant cost saving in building a storage array not from the most expensive and fastest drives, but from the largest and least expensive ones!

Like beauty, performance is measured in the eye of the beholder. With CASL, Nimble can actually make a 7.2K drive out-perform an SSD drive – so should we really call them slow drives?!  
The real bonus in Nimble’s design is that by breaking the tether to spindle-bound performance, future development of our arrays will show remarkable results. Trays of spinning disks are no longer required to deliver the performance demanded by intensive workloads, in fact in Nimble's CASL architecture adding additional drives to an array doesn't change it's performance at all!

Nimble arrays are performance bound by CPU only, and if you consider the rapid growth in capability of CPUs in recent years, you’ll understand the immediate advantage we have in the industry. There really is no such thing as the perfect drive, but Nimble Storage has created an array that behaves like one: performance and capacity at a sensible price.

 

This fresh, new approach to data storage means that customers no longer have to accept a compromise. They can invest in a platform that will scale non-disruptively as their business requires, and negate any concerns that they’ll out-grow their storage. No doubt when IT budget planning comes around each year, many of you will have faced the “what do I need vs. what can I afford?” conundrum – the good news is that now you can have your cake and eat it!


I'd be interested to hear your first-hand experiences of legacy tiering solutions, please insert your comments below!


References

 

CASL Deep Dive (video) by Devin Hamilton

Benefits of CASL: a Customer Perspective (part 1) by Amy Webb

Benefits of CASL: a Customer Perspective (part 2) by Amy Webb

 

Outcomes