Cross Over vs. Switched Network performance

Howdy, so I just changed my setup from having a single ethernet cable running from my mac pro to my file server (using a Netgear gigabit NIC – GA311NAR) to connecting over a DLink Gigabit switch (DGS-1005D).

I have had the understanding that network transfer rates through switching fabric will constrain the transfer rates (I would have thought by means of bandwidth or propagation time) and perhaps they do, but in short my little change over here has shown me that the difference is relatively irrelevant.

How I compared transfer times: On my mac I just ran the time command at the terminal and ran it for a copy of a 3.7GB file. I just copied the file twice for the cross-over configuration and twice for the switched configuration to compare.

For the cross-over arrangement, the times were:

real 2m1.398s
user 0m0.010s
sys 0m9.157s

and

real 2m0.478s
user 0m0.010s
sys 0m9.106s

For the switched configuration, the times were:

real 2m1.191s
user 0m0.011s
sys 0m9.108s

and

real 2m1.205s
user 0m0.010s
sys 0m9.082s

This is hardly enough of a test to really pattern the difference between the two. Both machines were idle while conducting the test. The performance differences of the switch would probably become much more obvious if there was a bunch of other traffic, but as it stands, there were only the two machines even connected to the switch. I read somewhere that the performance of the switch changes when non gigabit nics are live, though I haven’t tested that.

As I said before – there’s no obvious performance difference in this case. Perhaps I’ll submit another set of test results once I get my secondary workstation alive again.

Have fun

1 comment to Cross Over vs. Switched Network performance

  • Okay, so next I tried connecting a third NIC to the switch. It’s another GA311NAR NIC, however it’s not running at gigabit speeds, but rather at 100Mbps speeds.

    The transfer times for the same conditions were:

    real 2m5.934s
    user 0m0.009s
    sys 0m8.858s

    real 2m3.082s
    user 0m0.009s
    sys 0m8.779s

    So, obviously there seems to have been a slowdown with another NIC on the switch, though the slow down is pretty tiny it is measurable.