Latency Matters for (All-Flash) Storage Arrays

Latency Matters for (All-Flash) Storage Arrays

Chris EvansStorage

Yesterday NetApp announced the latest product in their interim EF-series all-flash array range, the EF560.  This new appliance claims to be faster and offer more performance than the previous model, as demonstrated by NetApp’s SPC-1 performance results and naturally the company seem quite pleased about it.  Although the EF560 isn’t the fastest platform available, it is the second best in terms of price/performance behind Kaminario’s K2 in SPCs testing.

I commented on Twitter that we see very little in terms of latency figures these days from our all-flash vendors and the main focus is on IOPS or throughput.  Most (but not all) vendors like to use the smallest block size as possible when quoting IOPS numbers as it makes their product look better, however application performance isn’t always based on the amount of data you can shovel to and from an external array.  There are plenty of applications that are latency sensitive that can’t also benefit from simply pushing flash into the server because they need the resiliency (or other features) provided by shared storage.  In this instance latency becomes critical and if that’s your requirement/issue, then many flash arrays won’t suit your needs.  As usual knowing your requirements here is the important first start.

So, do we have a two-tier market that means customers are focusing on either latency or IOPS?  Not really.  There are just different sets of requirements including low latency apps, latency sensitive apps (that need consistency) and requirements to deliver QoS (Quality of Service) which is almost impossible to do on disk-based arrays.  Finally there’s the need to generally increase the IOPS density (number of available IOPS per TB of storage) of storage as processor & memory performance also increase.  This is simply a maturing market.

The Architect’s View™

Full disclosure is great and customers should read what they will into all-flash vendor specs that don’t cover both IOPS and latency.  However as all-flash becomes more mainstream and is deployed at scale, then cost becomes a big factor again and this is why NetApp are so pleased with their new storage baby.  It’s a shame more vendors don’t choose to perform testing and disclose their results.  With these figures from NetApp, one has to ask the question, how much better can FlashRay be than the current products?  That remains to be seen, but if NetApp choose not to benchmark FlashRay, we all know what that means, don’t we….

Related Links

Copyright (c) 2009-2021 – Post #3571 – Brookend Ltd, first published on, do not reproduce without permission.