Will All New Storage Arrays be Hybrid?November 17, 2014
During Storage Field Day 6, I was fortunate enough to get presentation from Andrew Warfield from CohoData about a variety of things. I’ll say a variety mainly because my head is still swimming from all of the concepts that Andy was trying convey. If you don’t believe me, watch the videos and decide for yourself. WARNING!!! BE PREPARED TO PUT YOUR THINKING CAP ON! One of the concepts Andy was talking about was the idea that going forward, all arrays should be hybrid arrays. Immediately, my mind wondered what some of the “All Flash” array vendors would say about this, but he went on to explain this premise in more detail.
Consider what would happen if you could see analytics about your workloads to see cache hit ratios. This is something that several vendors can show now including Coho and Cloud Physics. You can now decide to buy more flash based off of how much a workload will benefit from it. Buying additional flash that isn’t large enough for your working set won’t provide you much value. For an example, take a look at the crude diagram below. You can see that you’ll gain larger benefits from flash if you buy 50 GBs of flash vs only 45 GBs due to the working sets. Alternatively going from 20GBs to 40GBs gives you almost no performance benefit at all due to the size of this working set.
Next consider what happens with multiple workloads on a storage array. We have our heavy readwrite apps like databases next to some file storage that is rarely accessed. If I’m going to put more money into my array to improve performance how should I do this?
Which strategy makes more sense?
1. Continue to add more and more flash to the array up to the point where the entire array is a single tier?
2. Look at the working set of your most important working sets and put even faster media in the array for even faster retrieval?
Since we don’t really care about the data that is rarely accessed, should we really spend the money on making that data more quickly accessible by putting more and more SSDs into the array, or leave it alone and use the money to speed up the important stuff? If we decide to quit adding SSDs and instead add NVMe we can further accelerate the most demanding workloads. Andy’s argument is that hybrid array’s aren’t just going to be around for a while until flash is cheaper, but rather the hybrid model will change to a faster medium than SSDs and solid state.
Hybrid Arrays may very well change from SSDs and spinning disk to NVMe and SSDs or NVDIMM and SSD but they would still be considered hybrid. What do you think? Does this methodology make sense, or will arrays be more like Pure Storage and XtremIO where all the storage is Solid State, much like older arrays were all spinning disks? I’d love to hear thoughts about this idea in the comments.