If you thought hyper-convergence is all about putting everything in one box, think again: some users are now asking for disaggregated hyper-convergence that sees the storage put back out on the network.
So says Michael Hay, Hitachi Data Systems' (HDS') veep and chief engineer, who tells The Register he's seeing customers who like the idea of hyper-convergence but strike situations where it makes no sense to have storage inside the server.
Hay says disaggregated hyper-convergence isn't for mainstream applications – he's seen it used for things like Hadoop when certain jobs are so processor-intensive it's useful to “move the storage process off the server.” In such situations, a pool of data and a storage fabric can suffice, he feels, even though users like to retain the software-defined view of their infrastructure.
How's this different from private cloud? The distinctions are wafer-thin but lie in disaggregated hyper-convergence being happy to work with an appliance model. It's just that sometimes you choose not to use the appliance for everything. “Classic” private clouds assume heterogeneous servers and networked storage from the get-go.
Vendors of hyper-converged appliances aren't going to take this road: Hay reckons it's an exotic option for those who need very high database performance, but also thinks there's enough interest in it that hardware-accelerated database companies are currently sucking in serious quantities of venture capital.
HDS is also moving on its own hyper-converged wars. Hay said “you'll see us do more” with the company's Hyper Scale-Out Platform before too much time has passed.
You'll also see HDS chase what Hay calls “co-investment” opportunities. These gigs see HDS work with customers doing things so novel that they can only be accomplished when getting down and dirty with HDS' research and development wonks, so that low-level hardware nuances can be brought to bear on a problem. Hay said the HDS has been working with a client on how to comply with Europe's <MiFID 2 regulations that, among other things, regulate automated high-frequency share trading. Complying with MiFID 2 will require the ability to monitor and analyse a great many transactions in close to real time, an effort that Hay feels creates saleable intellectual property. HDS hopes to offer customers the ability to create and then sell that IP, with the customer scoring a royalty stream as HDS finds more users with the same requirements. ®