Peloton-benchmark performs worse than "weaker" machines

[lmwnshn]

Summary Benchmark machine has faster CPUs, more CPUs, more memory, but still performs worse on YCSB.

Details

• Benchmark machine
  • 32 cores (Intel(R) Xeon(R) CPU E5-2660 0 @ 2.20GHz)
  • YCSB (default): 4-6k
• Comparison machine
  • 8 cores (Intel(R) Core(TM) i7-8550U CPU @ 1.80 GHz)
  • YCSB (default): 8k

Disk space doesn't appear to be an issue.

Tried without success

1. Disabling cores on the benchmark machine to also have 8 cores.

This is with a custom branch from @pervazea , however, there's generally been a trend of the benchmark machine reporting poorer numbers.

[tli2]

Maybe this has to do with high contention somewhere in our code?

[lmwnshn]

That's part of why we tried disabling the cores, but no luck.

[mbutrovich]

If this is single terminal (1 thread), I'm not surprised to see a 6 year-old Sandy Bridge @ 2.2GHz (turbo up to 3.0GHz) getting beaten by a brand new Kaby Lake @ 1.8GHz (turbo up to 4.0GHz) in single-threaded performance, even if the former is a server CPU and the latter is mobile.

The turbo frequency difference alone should explain the difference, but on top of that Intel makes regular IPC improvements with each microarchitecture.