Re-imagined Architectures: Towards more Efficient Data-Intensive Systems
Today’s datacenters host data processing and data management systems that routinely span hundreds of computers. At such scale, data movement and coordination is becoming a bottleneck at various levels, limiting further scalability. One way of reducing data movement bottlenecks is by pushing parts of the computation closer to the data source and, thanks to the changing hardware landscape, this can be done using emerging specialized hardware in an efficient manner. Another way of increasing the scalability of such large-scale systems is by reducing the cost of coordination across nodes, which is both a latency sensitive operation and potentially requires large bandwidth. Balancing these two requirements can be made possible by re-imagining existing architectures, both software and hardware.
In this talk I will showcase several examples of successfully tackling the above challenges. Some of these examples involve a specific type of specialized hardware, namely Field Programmable Gate Arrays (FPGAs). These are versatile reprogrammable chips that can implement algorithms in ways that are fundamentally different from CPUs or GPUs. FPGAs are, however, still an evolving platform with significant future work remaining related to programming, integrating and managing them within applications.
Zsolt Istvan is an Assistant Research Professor at the IMDEA Software Institute in Madrid, Spain. He earned his PhD in the Systems Group at ETH Zurich, Switzerland, working on distributed storage built with FPGAs and database acceleration. In his research, he explores ideas around specialization as a way of lifting bottlenecks in distributed systems and databases.