Comparison of chunkstore and version store performance

[TomTaylorLondon]

See attached. results.xlsx

[bmoscon]

What is the issue @TomTaylorLondon ? Which of those numbers need to be improved?

[TomTaylorLondon]

@bmoscon @jamesblackburn in summary:

• CS faster for large writes. Up to 20% in some cases (median 7%)
• Read performance in-line.
• CS twice as fast when appending a single day.

[TomTaylorLondon]

@bmoscon - we've been talking recently about combining both CS and VS into a single versioned chunk store. I'm surprised that CS is quicker than VS - the one variable I didn't test is changing the number of observations per day, my prior was that VS would be quicker. I can re-run with fewer observations per day and see if VS pulls ahead.

[jamesblackburn]

What’s the test code?

[TomTaylorLondon]

I'll clean it up and add to a branch.

[TomTaylorLondon]

https://github.com/manahl/arctic/blob/582-performance-bench/benchmarks/cs_vs_vs.py

[bmoscon]

@TomTaylorLondon for chunkstore are those appends or updates?

[TomTaylorLondon]

Writes and appends. (no updates)