Thanks for the pointer to iperf Joris!
Below are results using iperf, which look more promising.
* The vm is the iperf client
* Each iperf test was 10 minutes with X parallel threads.
* I ran each test 3 times.
* Performance in Mbits/S is the sum of all parallel threads.
This is on a core i5 iMac running virtualbox.
Parallel | Virtio: | Emu: em(4)
Threads | Mbits/S | Mbit/S
隖隖隖隖隖隖隖隖隖隖隖隖隖隖隖隖隖隖隖阊1 | 168, 168, 169 | 127, 128, 128
5 | 329, 326, 326 | 166, 169, 166
10 | 359, 355, 358 | 179, 171, 172
15 | 364, 360, 362 | 176, 175, 175
20 | 361, 368, 364 | 179, 179, 179
25 | 358, 354, 369 | 182, 184, 184
30 | 341, 350, 354 | 185, 187, 180
35 | 350, 352, 362 | 184, 184, 182
40 | 349, 364, 355 | 183, 182, 183
On a Xeon Linux/kvm box (same box w/ original scp results I mentioned),
the results are all over the place:
Parallel | Virtio: | Emu: re(4)
Threads | Mbits/S | Mbit/S
隖隖隖隖隖隖隖隖隖隖隖隖隖隖隖隖隖隖隖阊1 | 94.8, 66.3,72.6| 65.8, 67.1, 65.8
5 | 144, 144, 151 | 168, 116, 168
10 | 134, 149, 133 | 153, 134, 140
15 | 152, 137, 153 | 132, 135, 149
20 | 148, 151, 154 | 111, 149, 61.8
25 | 131, 150, 149 | 135, 105, 140
30 | 151, 132, 149 | 120, 138, 94.5
35 | 147, 149, 155 | 128, 93.1, 63.1
40 | 143, 177, 149 | 110, 124, 131
45 | 154, 142, 149 | 133, 131, 133
Tim