ZFS vs EXT4 vs XFS

[Ben Scott]

On 8/31/07, Dan Miller <rambi.dev at gmail.com> wrote:
> The bottle neck has and always will be I/O. I doubt the day will ever
> come that a fetch out to the disk will take the same time as a fetch out
> to memory.

  That assumes the constraints are not cumulative, and that workload
is a fungible thing, equally affected by both I/O and CPU wait.

  A process which is I/O bound and light on the CPU will likely behave
in the manner you describe.  Say, copying a file.  The CPU isn't doing
anything hard anyway.  The system is just pushing bytes through
buffers.  If the CPU is put to work doing storage management, so much
the better.

  On the other hand, something which is keeping CPU busy while also
doing some I/O (say, processing of a dataset), may well find that
latency stacks up, as the throughput is delayed first by an I/O wait,
and then a processor wait.

-- Ben