Modern computer systems are complex, and
designing systems with good performance and high reliability is a major
challenge. In this talk, I will show how a measurement driven
"methodological" approach to system design can create better systems.
The approach combines real-world measurements with techniques from
statistical workload and failure analysis, user behavior
characterization, analytical modeling, performance evaluation, and
scheduling and queuing theory to gain insights into system behavior that
lead to better design ideas.
Specific applications we consider in this talk include:
-
How to schedule connections in a web server to combat transient
overload;
-
How to provide QoS for database transactions;
-
How to exploit client behavior patterns to maximize system
performance;
-
How to improve reliability of large-scale clusters and storage
systems.
Bianca Schroeder is currently a postdoctoral researcher in the Computer
Science Department at Carnegie Mellon University working with Garth
Gibson. She received her doctorate from the Computer
Science Department at Carnegie Mellon University under the direction of
Mor Harchol-Balter in 2005. She is a two-time winner of the IBM PhD
fellowship and her work has won two best paper awards. Her recent work
on system reliability has been featured in articles at a number of news
sites, including Computerworld, Slashdot, StorageMojo and eWEEK.
Bianca's research focuses on the design and implementation of computer
systems. The methods she is using in her work are inspired by a broad
array of disciplines, including performance modeling and analysis,
workload and fault characterization, machine learning, and scheduling
and queuing theory. Her work spans a number of different areas in
computer systems, including high-performance computing systems, web
servers, computer networks, database systems and storage systems.