Hi Gregory, Thank you for the v3 patch. Gregory Price <gourry.memverge@xxxxxxxxx> write: [snip] > ===================================================================== > Performance tests - MLC > From Ravi Jonnalagadda <ravis.opensrc@xxxxxxxxxx> > > Workload: W2 > Data Signature: 2:1 read:write > DRAM only bandwidth (GBps): 298.8 > DRAM + CXL (default interleave) (GBps): 113.04 > DRAM + CXL (weighted interleave)(GBps): 412.5 > Gain over DRAM only: 1.38x > > Workload: W5 > Data Signature: 1:1 read:write > DRAM only bandwidth (GBps): 273.2 > DRAM + CXL (default interleave) (GBps): 117.23 > DRAM + CXL (weighted interleave)(GBps): 382.7 > Gain over DRAM only: 1.4x I've run XSBench based on the v3 patch and got numbers below. I used your sample numactl extension from here: Link: https://github.com/gmprice/numactl/tree/weighted_interleave_master Performance tests – XSBench NUMA node 0: 56 logical cores, 128 GB memory NUMA node 2: 96 GB CXL memory 1. dram only $ numactl -membind 0 ./XSBench -s XL –p 5000000 Threads: 56 Runtime: 36.235 seconds Lookups: 170,000,000 Lookups/s: 4,691,618 2. default interleave $ numactl –-interleave 0,2 ./XSBench –s XL –p 5000000 Threads: 56 Runtime: 55.243 seconds Lookups: 170,000,000 Lookups/s: 3,077,293 3. weighted interleave $ numactl --weighted --interleave 0,2 ./XSBench –s XL –p 5000000 Threads: 56 Runtime: 29.262 seconds Lookups: 170,000,000 Lookups/s: 5,809,513 In terms of runtime, weighted-interleaving shows 1.19x improvement compared to dram only, and 1.47x compared to default interleave. I’ve repeatedly run XSBench and have not observed any significant variations across the runs. Kind regards, Hyeongtak
