Getting the real dope on your cloud deployment

Lots of companies can perform cost analysis of cloud-based instances; Krystallize Technologies promises to do more. The Austin, Texas–based startup said its technology delves deeper into what’s going on with cloud workloads and compares how a given job will do across Amazon, IBM SoftLayer, vCloud Air, Microsoft Azure, CenturyLink, etc. to help you decide where to run it.

If it works as advertised, it could be a big leg up for cloud deployers (or would-be cloud deployers) who are discovering that there is no one-size-fits-all cloud. There will be times when a private cloud running large instances is more cost-effective than a public cloud churning a ton of itty-bitty instances. The problem is that most of that is discovered now by trial and error — if it is discovered at all.

Krystallize CloudQOS, on the other hand, enables real capacity planning, according to founder and CEO Clinton France. To get to its CloudQOS index, Krystallize buys the instances in question and then applies its own technology.

“We’re the first to put a workload simulation engine and a performance statistic to measure what’s going on in the cloud,” France said in an interview.

I’d been hearing about Krystallize a bit already from people in the industry who were impressed with the pilot, which has been running for about six months. It also got some early press. One data center specialist was particularly impressed because Krystallize takes a lot of factors into account — the cloud resources used, how the components are integrated and how oversubscribed (or not) the hypervisor is in any given case.

Krystallize measures cloud environments down to a level that gets to what is the true performance of a cloud instance, this expert said. One example: “Most people don’t know that the internal clocks in cloud instances are not necessarily reality,” said the specialist, who did not want to be named because he works with a lot of the cloud providers.

“You can think of it as the clock on a Star Trek holodeck.  The clock in the holodeck can be slower or faster than real time.  To solve this, Krystallize adds its own clocking mechanism that runs in the cloud instance to get apples-to-apples comparison of what gets done over a period of time.”

The company can measure a given workload, with calculations per second on one axis and variability along the other to give a better representation of what a customer can expect to get for its money. Great if a given cloud can claim a zillion calculations per second; not so great if it only hits that high only once in awhile.

In one analysis,  Krystallize ran the same workload on [company]VMware[/company] vCloud Air private cloud, [company]Amazon[/company] Web Services and [company]Google[/company] public clouds. Given the specifics of this application, vCloud Air private showed the best performance for a given number of transactions (about 78,000 calculations per second) with fairly good variability. Google delivered just over 40K calculations with more variability and AWS performed just under 30K calculations but with less variability.

Krystallize private v public

In this case VMware looks best if the application requires all those transactions, but if the application only had to deliver 27,000 transactions or less, VMware as configured would use just a third of the resources allocated. This is why to properly gauge performance, you must understand both the application workload requirements and the cloud capabitilies private or public, France said.

The company has been self-funded to date but just landed a $1.2 million seed round from several unidentified angel investors.

France said the beauty is that Krystallize can look at service performance over time and monitor them to make sure the customer gets what she’s paying for. One value is to provide a price/performance index, and another is that Krystallize can help customers do “cloud pruning,” jettison resources that aren’t up to the task and redeploying workloads to resources that will handle them better.

That could help companies rid themselves of “shelfware” or cloud instances that are still up there but not really doing anything.

“This is like going to the vegetable stand and instead of taking the stuff off the top, going in to find the better, fresher fruit that may be at the bottom,” France said.