Atom D510 1.6GHz tests & results

[Spy Alelo]

I know there isn't much research or results about this kind of setup, so let me share what I was able to see with mine. As of note, I didn't do any IPsec tests. The hardware is the following:

Intel D510MO@1.6GHz
Intel Pro/1000 MT Dual Port, PCI-X
Intel 82574 GbE NIC, mini PCIe
Realtek 8168, onboard, PCIe bus

The CPU is dual core, and has HyperThreading as well, which gives you four logical processors total. Some tests were done with and without HT and PF scrubbing, all other options are set to defaults. I am using pfSense 2.0.1-RELEASE (amd64). The tests reflect the total bandwidth I was able to achieve and the CPU load as a result:

**Test 1 without PF Scrubbing:
498Mbps Realtek to Intel, PCI-X, HT disabled, CPU 80%
470Mbps Realtek to Intel PCI-X, HT enabled, CPU 40%

Test 2 without PF Scrubbing:
570Mbps Intel PCIe to Intel PCI-X, HT disabled, CPU 80%
570Mbps Intel PCIe to Intel PCI-X, HT enabled, CPU 57%

Test 3 with PF Scrubbing:
392Mbps Realtek to Intel PCI-X, HT disabled, CPU 82%
376Mbps Realtek to Intel PCI-X, HT enabled, CPU 40%

Test 3 with PF Scrubbing:
523Mbps Intel PCIe to Intel PCI-X, HT disabled, CPU 92%
570Mbps Intel PCIe to Intel PCI-X, HT enabled, CPU 68%**

Enabling PF scrubbing has a hit on bandwidth and CPU, and is even worse with the Realtek chip. But if you use just the Intel cards, enabling HT actually helps and brings back the maximum bandwidth this setup can offer. Not too bad for a little device that uses an average of 14W. I know there's a lot of people that is against HyperThreading on FreeBSD, but if you use the right hardware, it seems like it actually helps. Anyone that have ways of squeezing even more performance out of this setup, or has any cool ideas on how to make it better, feel free to share.

[tirsojrp]

Clearly Intel beats Realtek but those CPU % seems to be wrong.

[Spy Alelo]

Really? Well, I think is the fact that I use the CPU load bar from the GUI. If you have a more accurate way of representing these numbers, do share! I'll do it all over again.

[stephenw10]

CPU seems about right to me. You could use top at the console though. Having the dashboard up uses some processing power.
Although enabling HT appears to reduce CPU usage in reality you are adding two virtual cores that pfSense cannot usefully exploit. Hence the average usage across all 4 cores is reduced by two that are hardly working. You will see that from top.
E.g. in test1:
Core1 100%, core2 60%. Average 80%
With HT
Core1 100%, core2 60%, core3 0%, core4 0%. Average 40%.
Obviously those figures will be slightly different!

Steve