NSF Award
2403253
Traditional data centers are organized with a cluster of computer servers. Like a personal computer, each computer server in a data center includes a set of computing hardware resources such as CPU, memory, and disks. In recent years, a new data-center architecture called resource disaggregation has arisen. With resource disaggregation, different types of hardware resources are divided into separate pools (e.g., a CPU pool and a memory pool), and an application can run with any available resources in a pool, thereby improving the resource utilization of a data center. Prior resource-disaggregation research has taken two main approaches: (1) asking application developers to port their software to a resource-disaggregation-specific model and (2) changing the operating system to add support for resource disaggregation. The former requires manual work, while the latter incurs significant performance overhead because of its generality. To solve these shortcomings, this project proposes to leverage application features and behaviors in building resource-disaggregation solutions. The project's novelties are a new direction in resource-disaggregation research, new computing layers explored when building resource-disaggregation systems, and the study of data-center applications from the perspective of resource disaggregation. The project's broader significance and importance is to render resource disaggregation a cost-efficient and performance-efficient option for production data-center and cloud environments that often host a range of applications with diverse behaviors and performance requirements launched by multiple users.<br/><br/>More specifically, the project aims to extract and leverage application-inherent semantics, including static and dynamic semantics such as memory access patterns, data object ownership, type-based data structures, execution profiles, and language runtime activities. This project lays the groundwork for constructing programming languages, compilers, and system support that can integrate a comprehensive range of program semantics into diverse disaggregation choices. This project aims to enable disaggregation systems to customize, without manual intervention, these multiple choices based on application-specific behaviors rather than relying on generic decisions.<br/><br/>This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.
