Many startups have come and gone since the early days of cloud, but when it comes to those that started small and grown organically with the expansion of use cases, Cycle Computing still stands tall.
Tall being relative, of course. As with that initial slew of cloud startups, a lot of investment money has sloshed around as well. As Cycle Computing CEO, Jason Stowe, reminds The Next Platform, the small team started with an $8,000 credit card bill with sights on the burgeoning needs of scientific computing users in need of spare compute capacity and didn’t take funding until this year. They took the $25 million in revenue they’ve gathered since 2008 and pushed it back into the business. Cycle now has broadly expanded its reach with enterprise customers while continuing with large-scale scientific and technical computing customers.
Stowe says Cycle has had three record quarters in a row and while they cannot report what revenue figure their 3X year over year growth represents, he says it is better than he could have dreamed when they were digging in with the first tentative users of cloud—back in the days when managing hybrid resources was still a big challenge. He says that while indeed, the cloud providers are making it easier to onboard and run applications, having an edge with an engineering team that understands the ins, outs, and pitfalls—and translates those into smart middleware that can seamlessly scale across the “infinite” resources of cloud is still necessary, especially for enterprise customers.
“The story hasn’t changed much in some ways since those early days. People are still looking to take advantage of any infrastructure available—both internal and external. They don’t want to wait for compute; they want to ask questions at any scale and get results back immediately. And for end users, there needs to be a toolchain that allows a seamless hybrid environment as well as all the benchmarking and other tools to make sure they’re only getting the best of what is available.”
Cycle already has a number of use cases in research, life sciences, and manufacturing, but one growth area is insurance and enterprise analytics. Most recently, the company worked with NASA’s Center for Climate Simulation and the University of Minnesota on a carbon emissions study using their own in-house middleware to talk to AWS resources. This unique workload used up to 5,000 cores with 43 TB of data pushed through cost less than $2,000. Key to this low cost (relative to the amount of work) was the use of Spot Instances on the Amazon cloud.
In this case, Spot worked out well—as it has for other customers Cycle works with. “Spot is usable for whole classes of workloads. In this case, many were 12-hour jobs. A lot of people say that’s too long to consider using spot with AWS (or pre-emptible instances from Google) but we found we could run the whole workload without worry. In these 12-hour jobs, maybe 25% of the VMs were interrupted and had to be re-run. But the cost savings far outstripped this,” Stowe explains. “The message is that just because a job takes longer than an hour doesn’t mean you don’t want to use spot instances.”
This insight is what makes Cycle interesting to talk to. They are working with diverse, sometimes complex applications that span both the compute-intensive and data-intensive realms. This means they have developed a keen sense, especially with their own internal benchmarking efforts, about which clouds and configurations tend to perform best for broad classes of applications. While Stowe said they all have benefits, there is still no hard and fast rule about which instance type or cloud provider is best. Everything depends on the workload—and that means it’s necessary to run benchmarks and take a close look at applications before getting started. Many of Cycle’s customers come to them with a sense of which provider to use, but Cycle’s teams help refine those decisions based on cost, time to result, and other factors.
Overall, Stowe says that when it comes to the various cloud providers they work with, there are some standout features. Azure’s Infiniband options are attractive for some customers, but he says the networks on Google and AWS are also good. “Azure is good for more traditional simulations; those workloads run well in that environment.” And on the AWS side, there are features the team has rolled out to make cloud more cost-effective, most notably Spot. Still, he says for Google’s part, they are working hard to be just as competitive on both the hardware and price points with a recent 33% price cut for their pre-emptible instances. “With Google, those areas, matched with their monthly usage discounting means they have a good cost impact as well as the performance.”
Even still, it is difficult to make sweeping generalizations. “There are some folks who say that one application should always get a particular instance type but this is not true in our experience. Different users of different applications require different RAM ratios, requirements for performance characteristics, and other factors. You can draw broad strokes by knowing what your application needs, but mileage may vary and it is critical to test before production,” Stowe says.
Stowe says that even as they have watched cloud use cases expand and the story remain mostly unchanged, there’s another element that’s not changing. “People are still thinking about their applications in terms of mirroring their internal and external environments. But shared infrastructure—the cloud—means that’s no longer necessary.”
Even though he says this is a widely known thing about the possibilities cloud opens up, there is still some catching up to do in the user community. “When you move to the cloud, you don’t have to repeat that shared infrastructure metaphor anymore. Budget spend widths and alerts can be set with managed controls so everyone has their own playground and powers it down at the end of the day.” This is still new to some enterprises, and while the operational benefits are clear, new customers are looking to this model for their development and other teams. “This is where the puck is going. Every member of a team having their own supercomputer. There is still an opportunity here.”
Cycle Computing was quick to spot the initial opportunity of cloud and has been out in front of several of the largest public cloud uses to date, particularly with AWS. Even though the various cloud providers continue to build (and acquire) tools to create a similar streamlined onramp for large applications, Cycle has persisted—growing its cloud partners, bottom line, big name use cases, and of course, its legacy as one of the few companies that made it past the first heavy wave of cloud startup cuts.