Benchmarks: Dask Distributed vs. Ray for Dask Workloads
|Efficient Data Communication Between Nodes
|Worker memory utilization
|Scheduler Fault Tolerance
|5.9GB RAM / vCPU
|7.5GB RAM / vCPU
|Ray is 27% more efficient than Dask Distributed in terms of how much RAM goes towards execution.
|Throughput (# Dask task graphs processed / hour)
|0.4475 graph processed every hour per 1 vCPU
|Ray has 4x higher throughput than Dask Distributed.
|Cost of processing 1.86M Dask task graphs
|Time for processing 1.86M Dask task graphs