Data Processing Systems, Programming Languages, Computer Architecture
Main-memory database systems waste valuable execution time because the processor waits for data to be fetched from the main memory instead of executing independent work from the many other tasks present in database workloads.
My current work focuses on hiding unavoidable cache misses through cooperative multitasking. I leverage language and compiler support to efficiently interleave the execution of parallel tasks and overlap memory accesses and computation across them. With such interleaved execution, we can eliminate the majority of memory stalls in latency-bound operations like index joins, leading to drastic performance improvements.
This work is done in collaboration with SAP SE.
|Ph.D.||Computer Science||Ecole Polytechnique Fédérale de Lausanne (EPFL)||Ongoing|
|M.Sc.||Electrical and Computer Engineering||National Technical University of Athens (NTUA)||July 2013|