Between the earth and the sky

Abstract landscape and cloud illustration

The term ‘Hybrid Cloud’ is much bandied around, especially by on premise vendors repositioning their product offerings. I’ve heard it used to describe being in migration to the cloud e.g. ‘we’ve migrated some of our stuff, but not all, so we’re hybrid’. A more reasonable definition, to me, is a single control plane managing resources running in public cloud and private data centers.

Hybrid cloud should be a single control plane managing provisioning, configuration, deploy, security, access and more across storage, compute and network. It’s a compelling, even seductive, proposition to the average CIO: a single plane of glass to manage all aspects of your infrastructure, agnostic of location or vendor. The ability to move workload and data to where cost and privacy can be best managed.

Yet, like all abstractions, this one comes at a cost:

The reality is that you’re going to have all sorts of terrors running on premise that can’t be easily updated to meet these requirements. Until you do, you don’t have the single control plane.

Worse, you’re not really getting the best benefit from the cloud. The really valuable cloud services are those that abstract away hard problems your team ideally shouldn’t worry about like ACID compatible geo replication and geo load balancing. Databases like Google Spanner and Microsoft Cosmos require tight vertical integration from the network up. They’re not going to just ‘run’ in your data center.

The real benefit in cloud comes from these higher level abstractions that save the team wasting time on low level problems. If you’re treating cloud as infrastructure only, you’re paying a lot for compute that you could be doing cheaper on premise.

So does hybrid cloud have no value? For most of us, no: you’re better off spending to migrate to platform as a service offerings. Where there is value is if you’re big enough to have homogenous compute workload that justifies the arbitrage benefit.

Just ignore the marketing.