The "Intel Accelerators Ecosystem: An SoC-Oriented Perspective" paper published in ISCA 2024 mentions achieving 2x higher throughput when running db_bench over RocksDB with the IAA plugin. However, the paper doesn't mention the exact workload or parameters used. Can you please share a command I can run to replicate these results?
The "Intel Accelerators Ecosystem: An SoC-Oriented Perspective" paper published in ISCA 2024 mentions achieving 2x higher throughput when running db_bench over RocksDB with the IAA plugin. However, the paper doesn't mention the exact workload or parameters used. Can you please share a command I can run to replicate these results?