Cheat sheet: cloudbursting
This hybrid approach could be the best way to use the cloud
It’s a sexy term, but what does it mean?
Doesn’t it sound cool? But I’m afraid this particular buzzword has nothing to do with Kate Bush. Cloudbursting is when you run your business mostly on your own kit, but also have a set of cloud accounts sitting idle, ready to take on extra “bursts” of work when demand peaks.
Isn’t that already the idea behind hybrid cloud?
Yes and no. Hybrid cloud is an umbrella term for dividing up your computing resources across local and off-premises servers; cloudbursting is a specific way of using those resources. In practice, a cloudburst setup might use containerised VMs and some form of load orchestration package to shift containers to locations where user sessions can reach them. It will probably require quite a lot of work at the database design level as well, so that this too can be replicated, multi-homed or remotely accessed. In short, cloudbursting isn’t an architecture or a computing philosophy, but a capability of your entire technology estate.
So you might say it’s an agile implementation of hybrid cloud?
That’s a question of semantics. A cloudbursting setup should quickly respond to unforeseen changes in demand, but this isn’t quite what’s conventionally meant by “agile”. Agility is about being able to retool your code quickly to adapt to changing circumstances, whereas cloudbursting requires everything to be in place well before the high-load day comes. You need to have your cloud accounts in place and paid up, you need to be sure that your code platform will run on the cloud, and you need to make sure that it’s actually capable of meeting the demands you want to place on it. Doing this properly involves a great deal of pre-emptive development and testing. I’d be very wary of a business that went into a cloudbursting project with an “agile” mindset.
Presumably a cloudbursting setup is cheaper than regular cloud hosting, because we’re only making occasional use of it?
It might work out that way, but the two models aren’t perfectly comparable. Hybrid cloud usually tends to imply an IAAS model, whereas cloudbursting finds most interest from heavy SAAS users. Cloudbursting also relies on your orchestration software correctly working out when to spin up the offsite services and incur the associated charges – which involves an element of voodoo, as it’s exquisitely difficult to distinguish between blips and booms as they’re happening. A hybrid cloud setup with plenty of slack capacity may or may not work out cheaper, but it’s likely to be more dependable, and have a more predictable cost.
When is cloudbursting the right answer?
There are some such scenarios, but they’re mostly inside the world of IT itself. For example, if you’re an antivirus developer combatting zero-day exploits, you’ll want the ability to scale your download links out into the cloud on bad virus days. Some classes of simulation can also easily parcel up workloads and hand them off to compute nodes with no regard for where those nodes are hosted. Unfortunately this model has become controversial, since it’s currently mostly employed by Bitcoin-mining trojans.
What’s the key downside of a cloudbursting approach?
Finance directors probably aren’t going to love cloudburst projects, because (as we’ve noted) the costs are unpredictable by design. What’s more, since the whole point of cloudbursting is that you don’t use it regularly, it’s only when you really need to fire up those cloud servers that you discover that a recent update has unexpectedly broken your meticulously crafted handover routines. These inherent risks will tend to push most businesses back in the direction of a more traditional hybrid architecture.
So why are vendors pushing cloudbursting as the next big thing?
I suspect that the vendors aren’t trying to get you specifically into cloudbursting. They want to make you think more generally about where your computing resources live. A little research, and my own anecdotal experience, suggests that very few companies have actually committed to a full-on cloudbursting model – which probably tells you everything you need to know.
“Finance directors probably aren’t going to love cloudburst projects, because the costs are unpredictable by design”