MIT Department of Electrical Engineering & Computer Science

E E C S

System Software Support for I/O-intensive Applications

Greg Ganger
MIT Laboratory for Computer Science

Monday, April 7, 1997
4:00 PM (3:45 refreshments)
Room NE43-518
EECS Special Seminar

Abstract

I/O is one of the main limiting factors in computer systems. Network and disk activity often dictates the overall performance and complexity of both individual applications and the system as a whole. In this talk, I will discuss ways of restructuring system software to solve the I/O problem, thereby realizing substantial improvements in both functionality and performance of real applications.

The solutions fall into two broad categories: (1) improvements to the implementations of existing I/O abstractions, such as file systems, so as to address I/O limitations for unmodified applications, and (2) a substrate that allows aggressive applications to, with little complexity and no replication of functionality, tightly integrate and control their I/O activity. This latter approach is critical for applications, such as networked data servers, that involve substantial amounts of device-to-device data movement.

We have implemented our proposed solutions and measured factor of 2-10 increases in end-to-end performance for important applications, such as HTTP servers and software development tools. In addition, our substrate has enabled the straightforward construction of applications that can not easily be built on conventional systems.

URL of this page: http://www-eecs.mit.edu/AY96-97/events/34.html
Created: Apr 2, 1997 | Modified: Jun 24, 1997
This announcement is from the MIT EECS 1996-97 archive. | Current events
To MIT EECS home page | Your comments and inquiries are welcome.