As cloud based workloads have started becoming ubiquitous, the need and adoption of storage that are suitable for these workloads, namely, object storage has started becoming prominent. There are various object stores in the market place, but in my humble opinion one that will stand out over a period of time is OpenStack SWIFT. I am saying this not based on my gut feeling but simply looking at the evolution of open source Linux which started as a science project and today has become a prevalent enterprise operating system – SWIFT looks to be on a similar trajectory.
Now, assuming you agree with me that SWIFT is ‘the’ object store going forward, let's ask the question “Is Open Stack SWIFT ‘as is’, good enough?” or will some twist spice it up for the hybrid workloads. To get an answer to this question, we first need to understand and admit the fact that many workloads today have hybrid needs/requirements on the backend infrastructure. Some workloads have strong affinity for file based storage with POSIX based access or distributed access (say via NFS) while others are growing with the affinity towards object storage with REST based access. So now, does ‘OpenStack SWIFT’ in its vanilla form helps me addresses these hybrid workloads out of the box? Well, I guess not.
Lets take my stupid & naive "1/2 chocolate 1/2 cinnamon doughnut" analogy – Let’s say, you love chocolate doughnuts “as much as” you do cinnamon doughnuts. You just have a dollar to spend allowing you to buy only one doughnut. Which of the below will you buy?
With that in mind and if you know what I mean, how about having an object store (SWIFT) hosted over one of the best clustered file system since sliced bread (IBM Elastic Storage). Getting SWIFT over IBM Elastic Storage really adds on with the required chocolate/value making the existing system suitable for consumption by the monolithic as well as hybrid workload – giving you best of both the worlds. It serves like a Swiss Knife giving you benefits of clustered filesystem , availability of standard POSIX interface, ability of using distributed interface (NFS) and power of SWIFT interface. Wow! Additionally, having SWIFT over a clustered filesystem like Elastic Storage gives you unique abilities of leveraging the special clustered filesystem capabilities inside object storage ( I will talk more on this later).
"A Deployment Guide for Elastic Storage Object" is a Redbook that has been published which has the required recipes for the doughnut that's just right for your taste & hybrid workloads. Link : http://www.redbooks.ibm.com/abstracts/redp5113.html?Open
Note: SWIFT on Spectrum Scale Native Raid (ESS) does not require 3 way replication - as erasure coding is inherent implemented in ESS. That is one key differentiator of IBM SWIFT over ESS compared to other SWIFT implementation.
In subsequent scribbling, I will deep dive on some more aspects on SWIFT over Elastic Storage - till then start thinking about your doughnut!
" These are my personal views and do not necessarily reflect that of my employer"