On Mon, 01 Feb 2016 00:49:56 GMT, Johnny B Good
wrote:

On Wed, 27 Jan 2016 15:10:41 +0000, T i m wrote:
snip

I'll have to test mine but being yours is a 'real' server (focused on
i/o and not economy like mine) is likely to be much better an ant
generic PC hardware running as a server.


That won't necessarily be true. For several years, I tried just about
every trick I could to get the data transfer rates between my NAS4Free
box and my win2k desktop machine (connected via 2 or 3 metres worth of
CAT5 in total using an 8 port Netgear GBit switch above 60MB/s (circa
500Mbps). Both machines were using 2010 vintage MoBos with built in GBit
lan ports and dual core CPUs.

The CrystalDiskMark results were interesting in that sustained large
sequential transfer rates hovered around the 75MB/s mark for any of the
four disks in the NAS box (mapped to local drive letters) almost without
regard to any real world stop watch timed benchmarked improvements I was
able to make.

The biggest improvement arose out of replacing the single core Semperon
in the NAS box with a dual core Athlon 64 chip

snip more interesting stuff for brevity

That would reinforce what I was thinking regarding the poor i/o of a
'std' (onboard NIC) compared with one focused on efficient / low CPU
involvement / server orientated NIC?

Transferring data is a very I/O based task and therefore shouldn't
require much in the way of CPU. So, as long as the hardware involved
was self sufficient (could use DMA etc) then it should offload much of
the CPU load onto the Ethernet card itself (and why the sell such
cards for 'servers' presumably)?

Network *and* any hard disk controllers may help.

http://www.intel.com/content/www/us/en/ethernet-products/gigabit-server-adapters/overview.html

Cheers, T i m