Performance Analysis of dummynet

The Emulab Project makes use of dummynet to provide traffic shaping between end nodes in the form of delay, bandwidth and loss. I am currently exploring pitfalls of dummynet and also propose solutions as to how dummynet can improved.

Acclerating Dynamic Binary Translation with GPUs [Poster]

Dynamic binary translation (DBT) is widely exploited in the areas of performance analysis, program optimization, security enhancement, and hardware virtualization. In contrast to static binary translation, DBT has relatively low startup overhead since it translates code as needed. Because only a small set of basic blocks (hot working set) is executed frequently for most programs, existing translators maintain a code cache for translated instructions, and try to optimize the layout of the cache to minimize execution time. However, a significant amount of time is required before this optimized cache layout is reached. Moreover, when used in a broader range of applications, translator's performance cannot rely on the presence of hot working set.

In this project, we propose a solution to these problems by introducing GPUs to accelerate DBT. To improve the performance, we plan to speculatively translate the instruction blocks that will likely be executed in the near future and store them for later use in a secondary cache memory. The main challenges of this work include how well we can predict the control flow of the target program and adjust the tradeoffs between the amount of computation and the overheads from data transmission. A further contribution of this project is regulating fluctuations in performance occurring due to execution of programs not relying on a hot working set.