Posted by
Sharing a resource, such as a computer processor or disk drive, requires the system to make decisions about which user’s task gets to use the resource and for how long. When my task is running, yours may be idling in a queue, awaiting its turn on a processor. The process of making these resource-allocation decisions is known as scheduling, and it has long been one of the most important aspects of system design, as a poor scheduler not only wastes resources but also can ruin a user’s experience.
Scheduling is a well-studied area, but cloud computing raises new challenges. The focus of our paper Zeta: Scheduling Interactive Services with Partial Execution, being presented this week in San Jose, Calif., during the Association for Computing Machinery’s Symposium on Cloud Computing, is an important type of cloud service: time-bounded interactive services. These are websites—such as Microsoft’s Bing and Office 365 or Facebook—in which a user interacts directly with a service running in the cloud, as opposed to an application running on a local machine. To provide a pleasing user experience, the service’s response to a user’s click must occur near instantaneously, typically within a couple of hundred milliseconds, around the limits of human perception.
