Launching a Beta Powered by Arm
Supergiant has always practiced what it preaches—or more precisely, used what it offers.
The company was born out of an internal project at QBox, a hosted Elasticsearch provider. The project’s goal was to make QBox’s infrastructure more efficient by using Kubernetes container orchestration and its own packing algorithm for resource management. The pilot resulted in a staggering 50% savings in QBox’s monthly AWS bill, totaling in the mid-five figures—and a whole new business.
The team open-sourced the Supergiant project in 2016, and began taking on customers who needed enterprise-grade Kubernetes support. Last year the company rolled out its own fully-managed, auto-scaling Kubernetes as a service beta under the name Supergiant Cloud.
The biggest challenge to launching Supergiant Cloud was changing the user interface to make the platform more approachable, no matter a user’s level of technological expertise. “We have a solid API, it works really well for technical people who integrate it with the tooling they have,” says Mike Johnston, a co-creator of Supergiant and the Qbox Chief Technical Officer. “We brought in a UX consultant, and spent a fair bit of time reworking how we present everything to someone who isn’t technical. We have both, very technical users, and non-technical users—for example, in enterprise teams an Ops person might be using the API but you might also have an end user that’s using the UI. But they’re both getting and consuming their data in a way that is more natural for their workflow. I call it an Ops Forward platform, meaning that it was built by our Ops folks, but the point of the platform is to make it accessible to people, to regular folks.”
Johnston was determined that the beta of the platform incorporating the new UI should be free. But that wouldn’t be possible using AWS. “Our alpha, which was actually on AWS, cost us over $30 per user per month,” says Johnston. “Part of that was because it was an alpha, so it wasn’t optimized. We had a lot of cleaning up to do, but a lot of it was infrastructure cost. The cost per user on AWS was unsustainable for a free beta.”
Looking for a New Provider
After doing some research, Johnston came across Packet, and the price point was right. The infrastructure cost on Packet for the beta would be less than 20% of what it would have cost on AWS. “It’s actually less than a dollar per user per month,” he says.
Before he found Packet, Johnston had been contemplating what to charge for the beta and how to get people to sign up. Would he have to offer a trial period? “We just added all this complexity,” he says. “With Packet, we didn’t have to do any of that. We could just spend a couple of weeks in engineering and simplify all of our stuff and just pack it on there. What Packet really gave us was the ability to democratize our beta by making it free.”
What was equally intriguing was the fact that Packet offered Arm servers, and was bringing in new versions too. “One of the reasons we chose Arm was because no one else is doing it,” Johnston says with a laugh. “But that’s just because the support and all that stuff haven’t caught up. As the support and community grows we’re seeing a lot more interest in Arm. It’s a selling point for us to be on a platform that was already on Arm. Grab Supergiant, and you can be on Packet, and you can be on Arm. And we can tell people, ‘Hey, you should try this Arm stuff that we support because we use it too.’”
After doing some research, Johnston came across Packet, and the price point was right. The infrastructure cost on Packet for the beta would be less than 20% of what it would have cost on AWS. 'It’s actually less than a dollar per user per month,' says Johnston.
The Arm platform was a good fit for the Supergiant beta’s particular needs. “Comparing an Arm server to an x86 server, the per dollar resources that are available on Arm are 8 times higher by purely core count, and on memory count it’s twice as high,” says Johnston. “My performance per core is a little bit slower, but we make a CPU reservation on a container, and we’re doing a beta, so we have like 500 containers running on a single server. Because the core counts are so high we’re able to just basically pack, pack, pack things on there. The performance is a little worse but I’m not impacting anything else and I’m not fighting for resource attention.”
A Dynamic Working Relationship
Working with Packet’s team, Supergiant took just three weeks to “go from ‘Hey, we should do this,’ to ‘Hey, it works,’” says Johnston.
While getting ready to launch the alpha, “we accidentally used up all of their Arm servers that were available,” he adds. “It was before Packet had gotten a big shipment of servers. They were really cool because they brought on some hardware that was only for test and development. They basically popped in new resources for us to get us by until they actually rolled out new systems. They’re really supportive.”
In addition to the beta, Supergiant is looking to move some customers onto Packet. They are currently collaborating with Packet on some pre-baked solutions to make it easier for Supergiant’s customers to use.
Beta and Beyond
The beta launched in early December 2017, and though it’s labeled beta, Johnston says, the product is very stable. The Supergiant API hasn’t had significant changes in almost two years; it’s the UI component that’s in active development. “Being in beta lets us collect a bunch of usage and metrics and figure out where we are,” he says.
With the UI revamped, Supergiant is planning to add new features to the product this year. “We’re focused on the things that people don’t realize they care about it until it’s too late, and that is cost management: setting up budgets for forward or back billing, scaling up and scaling down, certain types of resource scheduling,” Johnston says.
The company is also looking to ramp up the consulting, management and onsite engineering part of the business in conjunction with increasing adoption of the Supergiant platform. “We want to get the product into more people’s hands, but that seems to happen naturally as people find it and are able to use it,” says Johnston. “The more useful features that we put in there that are differentiators, the more people seem to like it.”