Zoltan Modly, you are my hero :-)
From the real world, from a customer, and from a great SE. I love the initiative, and the “I really want to dig in” attitude.
There’s a lot of debate that plays out in the industry, analysts and everyone around “AFA vs. Hybrid”. Yup – for some workloads, AFAs are the way to go (where latency consistency is critical). For some, Hyper-converged. For other customers that need a swiss-army knife to do it all (across mixed protocols, workloads, and use cases), a VNX or VNXe is often the tool.
While today, most AFAs don’t “Tier” it will be interesting to see how the hybrid core tech of “auto-tiering” makes it way into AFAs (if it does). It strikes me that there are potential benefits of using even some of the “coldest” write-cycle NAND in AFAs for some use cases…
Zoltan’s own words (this post is his, but I pushed to get him to ask his customer to share):
“Looking back over the past years, we have been very successful with the VNX Tiering (FAST VP) story/technology at a number of customers, but we have also seen lot of customers who simply did not believe (or did not want to believe) the benefits of tiering.
Many of them claimed that their data does not exhibit skew, or is too dynamic for tiering, and many argued that it is kind of a religious question.
This usually became a theoretical discussion, as nobody had the tools to look at this in reality (at least not on the sub-LUN level). Then came the “VNX Skew” Mitrend analysis, which provided quite a bit of insight into the sub-LUN slice level skew and tiering statistics. On the other hand, there is a lot more in this data than Mitrend currently shows, and Mitrend only looks at a single point in time, so it cannot be used to counter concerns around dynamics.
This gave me the inspiration to start digging into it with the goal of creating not only various point-in-time visualization of slice-level data, but also graphs and charts showing tiering behavior over time, and even time-lapse video visualization of tiering over time – as it happens, in reality, in the wild.
Thanks to Alex and the Mitrend folks, I got in touch with engineering and was given a preprocessor tool that allowed me to access the FAST VP data from the VNX array.
Once I had a way to access the data, I approached a few friendly customers who were kind enough to do the data collection on their systems. As a result, we had all the FAST VP slice statistics for every hour over 1-2 or even 3-4 weeks. Multiple GBs of data, literally millions of records in flat files.
I love that in the latest VNXe we have brought this goodness down to everyone – even the smallest customers.
How cool is Zoltan’s “search for the truth”? Read on for the data from a real customer VNX 5300, presented in beautiful video time-lapse. For EMCers and EMC partners that want to use Mitrend, PLEASE DO. Other than our investments in the SEs themselves (Zoltan – this is a great example of why!), tools like vLab and Mitrend are the best $ I spend in my SE leader capacity.