Abstract

Over the last decade a variety of novel server architectures such as Flash and SEDA have emerged from the systems community as alternatives to conventional thread-per-client server designs. These architectures are known to be extremely efficient on single processors, yet relatively little attention has been paid to how they will scale to the next generation of SMT /CMP/SMP machines.

In the first part of this presentation I will introduce Yield, a high-performance, staged application server written in C++ with an embedded Python interpreter for application logic. I will start with an overview of staged server architectures and then focus on some of the specific features of Yield.

The second part of the talk will resume where my last talk left off, with an investigation into the state of the art in stage scheduling algorithms, some proposed improvements to the same, and a discussion of problems that arise in developing new algorithms.