[X-HW] Ethernet Transfer Rates

[Ronald Chmara]

On Tuesday, December 17, 2002, at 04:15  AM, Vard Nelson wrote:
> We transfer large files (>> 2 GB) across the network.  At "100 mb/sec"
> it takes 13 minutes/2GB, much longer than the theoretical 100 mb/sec 
> (12
> MB/sec), and a Gigibit switch only decreases the time to 11 min / 2GB
> file.  What goes on?  Are there software products or whatever that help
> achieve closer to theoretical speeds?

Things that make a difference:
1. Cable quality.
2. Switch quality/setup.
3. Client hardware speeds (CPU/Disk/NIC/RAM).
4. Server hardware speeds (CPU/Disk/NIC/RAM).
5. Client OS software.
6. Server OS software.
7. Transfer protocol.

A long time ago I gathered a bunch of results and put together a table 
with some relative comparisons, benchmarking different AppleshareIP 
server solutions:
http://www.opus1.com/ron/asipstats.html

It's quite a bit outdated for what's currently on the market 
hardware/software wise, but it does a good job, in at least this case, 
of showing the *extremely* wide range of performance in the various 
products that were on the market at a given point in time. For a 100Mb 
link, it was possible to see everything from 2 to 65 mb/sec. with some 
pretty ancient hardware... (If anyone wants to add numbers for newer 
hardware, send it on in! I hear that current speeds are hitting up to 
80mb/sec with current OS X boxen.)

Typical bottlenecks usually include slow server OS's (*cough*Microsoft 
SFM*cough*), cheap "switches" that were designed for "workgroups" (not 
core infrastructure), cheapie NIC's, slowish IDE drives, and 
misconfigured switches.

-Bop