[Systems talk]: Accelerating Data Analytics in the Post-Moore Era
Abstract:
Hardware-software codesign is necessary for data analytics performance to keep up with the demand of processing the ever growing size of collected data. Traditional data processing performance optimization techniques offer diminishing returns as hardware performance scaling is slowing down. Towards this goal, I will first present Castle, a data analytics system codesigned with an associative processor (AP). Castle shows that we can effectively use the large data parallelism of emerging APs to accelerate performance by an order of magnitude. Then, I will briefly discuss Dynamic Interpolation (DIP), a set of new algorithms for searching sorted data, a core data processing operation. DIP can improve state-of-the-art performance up to 4.8X by taking into account the diverging memory and processor speeds, also known as the Memory Wall.
Bio:
Yannis is currently a Systems Researcher at Google. He received his Ph.D. degree from the University of Wisconsin-Madison, under the supervision of Jignesh M. Patel. He is broadly interested in hardware-software codesign for data processing and adaptive, robust and cost efficient query processing. His Ph.D. thesis focused on accelerating data analytics performance using modern and emerging hardware. Yanni's research has been supported by a Facebook Fellowship. He also received a B.S. and M.S. on Computer Science from the University of Athens, Greece.
Hardware-software codesign is necessary for data analytics performance to keep up with the demand of processing the ever growing size of collected data. Traditional data processing performance optimization techniques offer diminishing returns as hardware performance scaling is slowing down. Towards this goal, I will first present Castle, a data analytics system codesigned with an associative processor (AP). Castle shows that we can effectively use the large data parallelism of emerging APs to accelerate performance by an order of magnitude. Then, I will briefly discuss Dynamic Interpolation (DIP), a set of new algorithms for searching sorted data, a core data processing operation. DIP can improve state-of-the-art performance up to 4.8X by taking into account the diverging memory and processor speeds, also known as the Memory Wall.
Bio:
Yannis is currently a Systems Researcher at Google. He received his Ph.D. degree from the University of Wisconsin-Madison, under the supervision of Jignesh M. Patel. He is broadly interested in hardware-software codesign for data processing and adaptive, robust and cost efficient query processing. His Ph.D. thesis focused on accelerating data analytics performance using modern and emerging hardware. Yanni's research has been supported by a Facebook Fellowship. He also received a B.S. and M.S. on Computer Science from the University of Athens, Greece.
Practical information
- Informed public
- Free
- This event is internal