Over the past few years, cloud computing services like Amazon's EC2 have commoditized computing resources. Despite this success, cloud computing primarily targets users whose virtual machines (VMs) are rarely idle, as cloud users are typically billed for the amount of time the VM is awake, not how much work it does. Thus, services that are long-lived but mostly idle a significant fraction of the time are prohibitively expensive for many potential users of cloud computing. In this project, the PIs are developing a finer-grained model of cloud computation based around lightweight instances. Lightweight instances are more akin to processes than virtual machines; they allow clients to only pay for their actual usage of cloud resources, and to not have the complexity and overhead of running an entire operating system. The lightweight instance approach has the potential to bring the benefits of cloud computing to a variety of users for whom it is a poor fit today. The PIs are exploring a new, finer-granularity model for cloud computing, wherein the cloud makes it appear that all client instances are running all the time, but may actually swap them out while idle to permit statistical multiplexing. Compared to today's VM-based model and container-based model, the proposed process-based model provides a much higher level of abstraction, and lets clients run long-lived, mostly-idle services much more cheaply than is possible today. If successful, the research would open up a wide variety of new applications and architectures: processes could cheaply be run in the cloud on behalf of end users, providing them with the ability to run long-lived services cheaply.

National Science Foundation (NSF)
Division of Computer and Network Systems (CNS)
Application #
Program Officer
Marilyn McClure
Project Start
Project End
Budget Start
Budget End
Support Year
Fiscal Year
Total Cost
Indirect Cost
University of Maryland College Park
College Park
United States
Zip Code