When you’re building a product or solution then you have to start with the reason why you are building it and this has to encompass the value that it brings to the people you’re going to sell it to and the value that it’ll bring to you as the company selling it.
And this is what I don’t really see when I look at Pure’s Cloud Block Store (CBS).
It’s a neat solution, but as with their FlashBlade product it feels that whilst it’s technically clever it’s not really sure what problem it’s actually trying to solve, if you remember with FlashBlade it started life as ‘The Isilon killer’ for unstructured data, then it became a Flash backup target and now it’s the solution ‘built for AI’. I like innovation and seeing companies try new things, but history is littered with good solutions that never really worked out what problem they were trying to solve, or more importantly, a problem to solve that enough people would care about to want to buy it.
When you take a look under the hood of the Cloud Block Store (CBS) solution there are some quite nice ways that they’re looking to ensure data protection and also performance, no surprise really as Pure has always been more focused on the underlying storage media than the actual data services.
In an on-premises array you have a pair of controllers and under these are the Flash devices. At a high level the way that the CBS works is that you setup AWS EC2 compute instances to simulate the Storage controllers and then have groups of 7 x EC2 storage optimised instances to simulate the disks. It really is a virtual manifestation of the physical device. This provides some good capabilities for resilience from the possible failure of any individual instance and good performance.
Back to what would you actually use it for? The ‘why’ if you will
Firstly, it’s block only and in my experience, when I talk to companies about their plans to move applications to the Cloud then part of this discussion is also a move away from using traditional block protocols. For example, there are a large number of companies moving SAP Hana to Azure right now, this is the epitome of a Tier 1 Enterprise Application, and many of these moving to Azure are using file based solutions to underpin it. There are so many benefits, it’s simpler, it performs and it’s taking a more Cloud appropriate direction rather than trying to force fit your traditional block-based approach into a modern Cloud environment.
Taking a file based approach with the Cloud also opens up a lot of extremely interesting possibilities, I talked in a previous blog about ‘Data Amplification’, if you can snapshot or clone file based data then it’s incredibly simple to be able to attach this data to any application that’s available in the market place of the Cloud that you’re working in. You also get the opportunity to support technologies such as Global File caching in order to deliver file-based data services backed by the Cloud but with local performance across multiple different locations.
There are some edge cases for block services across multiple clouds though, but I would say that most frequently this is for DR, Test and Development types of activities, it’s the easiest way to replicate the traditional on-premises way of doing things into a traditional way of doing something inside a hyperscale cloud provider.
Putting aside the use cases above then is there a reason why they’ve taken this approach?
It feels to me like this was designed by a company with a very traditional hardware product, ie dual controller, block only storage, one that was never designed for the Cloud and that the sudden need to be able to claim cloud integration has forced them to take this traditional approach and kind of force fit it into the cloud in order to try to build a relevant story. Now these traditional hardware focused block storage products might fit but it feels like your forcing the cloud to emulate the way that you work on-premises. It’s a legacy approach to Cloud data services, and as Pure have always been very vocal about this, it goes completely against their argument as to how you can’t embrace modern ways of doing things with legacy technologies or approaches.
When I think about the Cloud, I’m thinking about it from a different direction, providing NFS, SMB or S3 services to support any data for pretty much any application sets up a hugely diverse range of use cases that the solution can support. If you can also support block then you can also deal with the edge cases as well. It’s much more about the data services you can deliver along with data protection and efficiency and much less to do with the underlying media. This of course means that if your array was built from the ground up for Flash and little else then it pretty much pulls the rug out from under your solution.
Is it down to performance then? Yes the CBS solution has good performance, but then you have to define good. You can get performance that’s good, even excellent from much simpler Cloud based data storage services.
My feeling is that Pure have engineered a solution that under the covers has some neat engineering features, it also gives them the opportunity to claim they have a cloud story, but it doesn’t feel like the questions of “why would customers want this”, and “would enough want it to make it a significant business” were actually asked. It feels like this was developed to demonstrate innovation to investors and customers but could well end up having very little uptake, hopefully Pure will break this number out in their earnings and maybe I’ll be proved wrong.
I suspect that as with their FlashBlade, CBS will go through a number of reincarnations till it actually finds a problem to solve, which I will watch with interest.