Very different performance between Unix and Windows for EOO.computing

[thiagosfsilva]

I was running an analysis where I bootstrap alpha hull calculations for thousands of species, and after investigating the reasons for the unmanageably long processing times I was getting, I happened to find out that performance of EOO.computing on Unix systems seems to be at least an order of magnitude faster than on Windows, Here are some results using benchmark() (script and data attached).

Manjaro Linux kernel 4.19.60-1; i7-8750H 2.20Ghz, 32 GB RAM

replications elapsed relative user.self sys.self user.child sys.child
1           10  13.176        1    13.153    0.019          0         0

Ubuntu linux kernel 4.15.0-54; AMD Ryzen 1800x 3.6Ghz, 64 GB RAM

  replications elapsed relative user.self sys.self user.child sys.child
1           10  13.687        1    13.642    0.072          0         0

Windows 10 Pro 10.0.17135; i7-5820K 3.30Ghz, 64 RAM

  replications elapsed relative user.self sys.self user.child sys.child
1           10  327.89        1    1.65    0.065          NA         NA

I'm not sure if there is something I am doing wrong, but I thought it was worth raising the issue. I also benchmarked it on a an old Mac Pro and didn't save the results, but speed was overall on par with the Linux systems. Another thing I noticed was that reducing the number of cores (I was initially running it in parallel, but the above timings are all single core) decreased Windows processing time slightly.

Unfortunately I don't have immediate access to a second Windows machine to test it.

testing.zip