With an Intel Pro/1000MT Card in one of the G5’s ;)

2.3->2.3, same conditions.

FTP Push: 34-42mb/sec. 50% Cpu on Recv, 22% Cpu on Sending
FTP Pull: 41-47mb/sec. 30% Cpu on Send, 28% Cpu on Recv

Note: On a pull, the card is much less reliable - it shows up to 50mb/sec, but ends at much lower rates. With the Mac Card, this behavior didn’t happen - it was basically steady the whole way.

Dual 2.3: (250Gb Sata) - All rates are actually mb = 1000000b/s
Harddisk: WDC WD2500JD

Gig Zeros: 53mb/sec write
Gig Random: 11.8 mb/sec write

Read:

Gig Zeros @ 64k: 53mb/sec
Gig Zeros @ 16k: 55mb/sec
Gig Rand @ 16k: 56mb/sec (could be due to drive position)

All network tests done with a gigarand.bin - 1Gb, all random numbers.

Network: Dual 2.3 to Single 1.6:
FTP Push - 2.3->1.6: 22Mb/sec - 22% system CPU on 2.3, 100% CPU on 1.6!
FTP Pull - 1.6->2.3: 36Mb/sec - 28% system CPU on 2.3, 65% CPU on 1.6.

Note: 1.6 is using IDE - that could contribute to very high CPU overhead of FTP transfers - 90% or more of the CPU is contained in system (kernel) space, not user.

Network: Dual 2.3 to Dual 2.3:
FTP Push - 38-42Mb/sec - 20-30% system CPU
FTP Pull - 50-53Mb/sec - 30-45% system CPU

Note: Pulling via FTP, regardless of which direction between 2.3s, is 10Mb/sec faster. This could be due to the way ftp is optimized (for pulling). It’s unlikely it’s the disk or network card, because the results are completely symmetrical.