For some latency-intensive workloads, the developer can stream only part of the application and keep other parts on the local client. "It's not all or nothing," Frazzini said. "The parts that have no tolerance for latency can be run locally, but other parts, such as background scenery or advanced effects, can be streamed in."
This hybrid approach could help the size of downloads for very large applications, Frazzini said.
The service uses NVidia G2 graphics processors, running on Microsoft Windows Servers on EC2 (Elastic Cloud Compute).
To write to the service, the developer only needs to know how to work with Microsoft's DirectX rendering library. The local program interacts with the service through a set of APIs (application programming interfaces) that the developer can work with by way of the SDK AWS provides.
Over time, AWS will also offer a set of plug-ins for popular IDEs, such as Eclipse.
To stream the imagery, AppStream uses Amazon's recently developed SDX (Streaming Experience) broadcast protocol (which is also used by Amazon's new WorkSpaces virtual desktop infrastructure as well).
SDX uses forward error correction and other techniques to smooth performance even when the network bandwidth is choppy.
Like other AWS services, AppStream will be priced on a pay-per-use model, depending on the amount of processing done on EC2, bandwidth usage, licensing for the Windows Server, and the use of SDX.
Those interested in trying the service before it becomes commercially available can apply to for an account.