Rumen Hristov
MIT EECS — Actifio Undergraduate Research and Innovation Scholar
Dynamic Planning for Distributed Computations
2015–2016
Matei Zaharia
Distributed systems play a huge role in modern day computing. Recently people have focused on developing distributed data processing engines. Computing time is directly connected to cost that the user pays, so efficiency in these engines is a primary focus. The user specifies a list of commands, which he wants to be executed on the data, but then the engine needs to translate it to its internal representation and create a plan for execution. Distributed data processing engines select an execution plan up-front for each query, but suboptimal plans can lead to significant slowdowns. Instead of this approach, we propose adjusting the query plan dynamically while a query is executing, and modifying data processing engines to make these kinds of adjustments possible.
I think the SuperUROP program is a great way to thoroughly engage in a research project in MIT. I am really looking forward to learning new things and seeing my project applied in the real world.