The case for network-accelerated query processing
The case for network-accelerated query processing Lerner et al., CIDR’19
Datastores continue to advance on a number of fronts. Some of those that come to mind are adapting to faster networks (e.g. ‘FARM: Fast Remote Memory’) and persistent memory (see e.g. ‘Let’s talk about storage and recovery methods for non-volatile memory database systems’), deeply integrating approximate query processing (e.g. ‘ApproxHadoop: Bringing approximations to MapReduce frameworks’ and ‘BlinkDB’), embedding machine learning in the core of the system (e.g. ‘SageDB’), and offloading processing into the network (e.g KV-Direct) — one particular example of exploiting hardware accelerators. Today’s paper gives us an exciting look at the untapped potential for network-accelerated query processing. We’re going to need all that data structure synthesis and cost-model based exploration coupled with self-learning to unlock the potential that arises from all of these advances in tandem!
NetAccel uses programmable network devices to offload some query patterns for MPP databases into the switch.
Thus, for the first time, moving data through networking equipment can contributed to query execution. Our preliminary results show that we can improve response times on even the best Continue reading