I'm not sure that there's much value in this article.
1) It doesn't really compare benchmarks on OSX vs Linux - you're dealing with different drivers, kernels and filesystems.
2) Then there's the matter of EXT3 - there is little to no value in benchmarking such an old filesystem with modern hardware, not to mention that EXT3 lacks native trim support thus major impacting performance over time (This is assuming the host OS is also using EXT3 and thus not passing through the discard IOCTL to the block device).
3) There's no mention of what Kernel version was being run, modern kernels (3.6+) are significantly more efficient with both disk and network I/O.
4) How much memory is being provided to the host and guest VMs and how much of the 'benchmarks' are being cached? What kind of disk / filesystem caching are they using? What is the IO scheduler for both the guest and the host machines?
5) What 'benchmarks' were actually run? I'd bet that it involves using dd which is by no means a benchmark - nor can it be trusted (especially without what commands were even being run) - if you're going to benchmark disks use fio or bonnie++ (although I think fio is more useful)
1) It doesn't really compare benchmarks on OSX vs Linux - you're dealing with different drivers, kernels and filesystems.
2) Then there's the matter of EXT3 - there is little to no value in benchmarking such an old filesystem with modern hardware, not to mention that EXT3 lacks native trim support thus major impacting performance over time (This is assuming the host OS is also using EXT3 and thus not passing through the discard IOCTL to the block device).
3) There's no mention of what Kernel version was being run, modern kernels (3.6+) are significantly more efficient with both disk and network I/O.
4) How much memory is being provided to the host and guest VMs and how much of the 'benchmarks' are being cached? What kind of disk / filesystem caching are they using? What is the IO scheduler for both the guest and the host machines?
5) What 'benchmarks' were actually run? I'd bet that it involves using dd which is by no means a benchmark - nor can it be trusted (especially without what commands were even being run) - if you're going to benchmark disks use fio or bonnie++ (although I think fio is more useful)
Here's a good article on how to perform a few kinds of useful benchmarks: https://www.binarylane.com.au/support/articles/1000055889-ho...
Beyond that, I'd also look at using PGBench to get an idea of the IOP/s in real world scenarios: http://www.westnet.com/~gsmith/content/postgresql/pgbench-sc...