search

zarr-developers / zarr-benchmark

Benchmarking the performance of various Zarr implementations, using our perfcapture framework.
MIT License
1 stars 2 forks source link

Re-implement `SlowMemcpyWorkload` & `SlowMemcpyDataset` & rename #8

Open JackKelly opened 1 year ago

JackKelly commented 1 year ago

Vincent suggests:

  • one-pass_out-of-core_batchwise_low-latency <-- this is what I proposed, locally from SSD
    • access profile 1, profile 2, profile 3, ... (various advanced indexing profiles while doing the one-pass)
  • one-pass_out_of_core_batchwise_high-latency <-- this would be the same, over the network
    • access profile 1, profle 2, profile 3, .... (various advanced indexing profiles while doing the one-pass)
  • parallel-map-reduce_local_low-latency <-- this would be map-reduce but on a single computer
  • parallel-map-reduce_clustered_high-latency <-- this would be map-reduce but in a clustered environment

xref #5 #6 #7