Hardware

Hello People!

I'd like to hear your educated guess (or not) on the slowest x86 cpu that can handle 100mbit down or up.
And also how things change if we let it happen through many connections - like through bittorrent.

I'm running m0n0wall on a 500mhz amd k6/2. And have done limited tests. A normal http get would give me 100mbit - at some 40% cpu usage - peaks at around 70%.

I'm concerned this would not be enough for multiple torrents.

This is using fxp and xl based netcards. Avoiding rl-based ones.

Let me know your thoughts

fxp NIC's would likely be your best bet. em NIC's would probably be better, even just using them for 100 Mb. Possibly not a measurable difference between the two.

I'm surprised you can get that much through an AMD K6/2. Personally I would go with a 800 MHz P3 or better, but it sounds like what you have might be OK with the hardware you have.

Things vary so much from install to install that it's hard to give you any specific recommendation. CPU will be your first bottleneck, so watch your CPU usage and see what happens.

Thank you for your input. I do have an 800 P3, 1.3 ghz celeron and a duron 700 on hand. But it's also a matter of watts The 500 AMD uses just 19 watt according to volt/amp meter. And I'm sure I saw twice that on the p3 800mhz.

I'm running m0n0wall on a 500mhz amd k6/2. And have done limited tests. A normal http get would give me 100mbit - at some 40% cpu usage - peaks at around 70%.

That's somewhat a disconnect in light of Manuel's test results on the new PC Engines board. Is the via rhine that much worse than the xl? Your result makes the k6/2 look pretty good next to the geode LX if you're truly getting 100 mbps throughput.

http://forum.m0n0.ch/index.php/topic,369.0.html

db

That's somewhat a disconnect in light of Manuel's test results on the new PC Engines board. Is the via rhine that much worse than the xl? Your
result makes the k6/2 look pretty good next to the geode LX if you're truly getting 100 mbps throughput.

... Well the Geode LX @ 500mhz has a TDP of just 3.8 Watt (according to wikipedia), which makes it far superior in my eyes. My test was a real life wget from inet test with a look at cpu usage reported by m0n0wall. I did not measure the exact throughput. And so i'm not sure it's valid to compare it to his test.

wget isn't quite as good of a test for testing network throughput as iperf, what Manuel used. If unlotto is seeing 40-70% CPU with wget @ 100 Mb wire speed (the ~94 Mb Manuel got is wire speed of 100 Mb), that sounds very comparable with what Manuel was seeing on the Geode 500 MHz. I would expect the Geode and K6/2 to perform roughly the same, with some difference depending on the NIC's used.

The 500 AMD uses just 19 watt according to volt/amp meter.

How did you check that?

The 500 AMD uses just 19 watt according to volt/amp meter.

How did you check that?

Google Kill-a-watt, you can pick up a meter for ~$20 USD that'll tell you that.