Companies Promise a Cloud that Won't Crash
Some say that recent high-profile cloud outages are just a matter of bad design.
Whenever Amazon’s cloud service goes down, the sites and services that rely on it, including Reddit, Quora, and Foursquare, go with it. After a couple of major outages early this year, competitors are aiming for the jugular by promising different approaches that, they say, provide better reliability and performance.
Joyent, a San Francisco-based cloud-computing company, announced a new product on Thursday that promises 99.99 percent uptime, with “downtime measured in seconds,” and no data loss (Amazon promises 99.95 percent). Joyent says it can achieve these improvements by using a different architecture than Amazon does. It also offers analytics tools that give customers more insight into the state of its system.
Concerns about reliability have long been a stumbling block for the cloud-computing industry. Businesses have felt financial pressure to switch to the cloud model, which saves money by shifting storage and processing onto shared computing resources, but many worry about losing control over their data, or wonder what would happen if their chosen provider went down.
These concerns came to the fore after high-profile outages of Amazon’s Elastic Compute Cloud in April and August of this year. This service allows users to rent virtual computing resources, which can be scaled up or down as their needs fluctuate. Some customers lost data during the outages. Google Docs and Microsoft Office 365, two cloud-based office suites, also suffered outages in September.
Joyent has been operating a cloud service for more than six years, and its customers include LinkedIn and Shopzilla. The company says the new service, called Joyent Cloud, adds greater reliability and flexibility to its existing one. Customers can also, for the first time, use any operating system in combination with it, and Joyent now offers an application programming interface that makes it easier for software developers to build programs that use its cloud services.
To allow customers to better understand what’s happening in the event of a slowdown, Joyent has added features that let customers see statistics showing, for example, how long it takes for information to travel from one point to another in the system.
A key problem for Amazon’s Elastic Compute Cloud service, critics say, is that it relies on technical wizardry that’s difficult to pull off reliably. It uses a technology called Elastic Block Store to record and retrieve data that’s being actively used by applications running on the service. It’s easy for cloud systems to store data reliably if there’s plenty of time to distribute the data across all of the machines that make up the cloud. In that case, data might not be entirely consistent at all times, but it will get that way eventually. But Elastic Block Store is built to have it both ways—to be reliable, but also to record changes throughout the system very quickly.
Steve Tuck, general manager of Joyent Cloud, says that such a balancing act is too precarious to be reliable. “You are setting yourself up for this type of [outage],” he says.
Joyent’s cloud avoids taking data that’s being actively used and distributing it over the network. Instead, it distributes that data over a number of computers locally to keep it safe and accessible, and stores less frequently used resources more distantly in the cloud.
Nasuni, a startup based in Natick, Massachusetts, also boasts a different architecture that, it says, can make cloud computing services more reliable. Nasuni believes companies will be more comfortable trusting the cloud as part of a reliable system that includes local storage. According to CEO Andres Rodriguez, a system such as Elastic Block Store can be “a ticking time bomb” within the Elastic Compute Cloud. He says that because the system needs to be responsive, it can’t safely distribute changes with enough time to guarantee that it won’t lose data.
Nasuni offers a solution that keeps active data local, and it’s careful about what it moves to the cloud. It might move backup information to the cloud, for example, but store frequently used documents locally.
Rodriguez argues that making these architectural changes won’t diminish the power of the cloud. “What almost no one understands is the capability of that massive data structure in the cloud,” he says. “It’s a commonly shared, unlimited, immutable, self-protected data structure.”
Companies such as Nasuni and Joyent hope that businesses will be more comfortable using the cloud if their active data is kept close, with cloud storage reserved for data that doesn’t need to be as consistent and immediately accessible. “We’re trying to take everything that’s revolutionary about the cloud and keep it under covers,” says Rodriguez.
Become an MIT Technology Review Insider for in-depth analysis and unparalleled perspective.Subscribe today