Soekris net5501 performance
This morning I did a simple test I wanted to share with you, I connected on PC to an additional interface of my router based on a Soekris net5501-70. I wanted to see how much "juice" this puppy can throw… What I was interested in knowing is the speed it can handle with QoS enabled...
Here is the setup. Nothing is bridged everything is normally routed.
vr0 = WAN (dhcp)
vr1 = LAN (10.77.2.0/24)
vr2 = OPT1 (10.6.9.0/24) (My lonely PC)
From the OPT1 to the LAN I connected to my SMB server which is connected to a Gigabit switch with a gigabit link (pfSense is also connected in the same switch).
Without QoS Windows 7 reported 10 MB/s of file transfer, pretty much the whole 100 mbits available... I then struggle to make a quick Traffic Shaping rules and finally got it working and to my surprise I could see in the Queue monitor more than 7000 pps and a bandwidth of 100 Mbps. Using SSH and "top" the CPU was maxed out though... Ping time to the router from the LAN where in <20 ms range. The WebGUI was responsive and SSH too. The CPU was used at more than 80% by interrupts. Windows 7 reported the same speed, then I started a file transfer from WAN to LAN (12 mbits) then bandwidth from LAN to OPT1 began to lower a little bit, my WAN to LAN transfer was not affected at all by the file transfer.
This router board is truly a beast...
I know this test is not really what you guys would call a "good" test, but this is enough for me, I know now that my router can handle the 50 mbits my ISP is offering... Maybe one day I'll get that speed, the price tag didn't got approved by the WAF.
That's good to hear. Those boards, and the ALIX (same CPU) pack a lot of punch for 500 MHz. But yeah you're going to drag them down at around 100 Mb. That's very solid performance for the price point.
I have tried to enable polling but that cause the WebGUI almost impossible to work with… I understant why the CPU get to 100% all the time when polling is enabled, but something must be wrong is slow down everything else...
Yes the polling does lower the interrupts usage, but then without any traffic at all the WebGUI is too slow to be workable.