Last time I casually installed a soft router, but now I had to spend a lot to set up another one. The previous article: casually installed a LEDE soft router, testing whether the WAN port can run at 10G (Part 1).
Some issues were unclear, and I felt uncomfortable psychologically, causing sleepless nights. I’ve been tinkering with it for several nights, and my sinusitis has flared up again, really affecting my health. This machine, apart from the case and memory being new, is all second-hand. I won’t count the hard drive since I used a USB drive to install the system for convenience, totaling 2200 yuan. This price is enough to buy a second-hand two-port 10G switch (Asus XG-U2008 and Netgear GS110MX), and for an additional thousand, I could buy a four-port QNAP QSW-804-4C. However, my machine isn’t lacking in the number of ports either, with 4 10G ports and 5 1G ports.
The platform is Z270 + G4560 + 4G memory, with an Intel X520-T2 and X540-T2 installed, as well as a four-port 1G Intel 82576, plus one 1G port from the motherboard. The motherboard also has 3 PCIe 1x slots, which can accommodate dual or even quad-port network cards (just for fun). Let’s first talk about the power consumption of this setup, generally around 70W, and over 80W when the 10G network card is working. Hmm! It’s not 18W, it’s over 80W. Compared to the 10G switch (around 50W), it’s a bit higher.
The back of the three network cards looks quite nice. (PCIe slots: 8x + 8x + 4x)
The X540 and X520 chips are quite old and require cooling fans.
Back network ports:
There are also three 10G network cards installed on other test machines, the same as the ones shown in the previous article: X520-T2, BCM57710, and another TN9712P (this is the model recognized by the computer, essentially Tehuti TN4010). Before testing this 10G soft router, I need to solve the direct connection speed issue between the test machines. Whether it can reach full 10G depends on two main factors: whether the CPU is powerful enough and whether jumbo frames are enabled.
The data in the table below took hours to test:
<<>> represents bi-directional transmission.
The G1840 CPU struggles to meet the demands of the X520/X540 network cards; using it with the G4560 shows a significant improvement. Previously, the bi-directional speed was only around 13G, but now the X520 or X540 with BCM57710 achieves a bi-directional transmission of 19G. After disabling jumbo frames, the bi-directional performance drops significantly, and the upload speed of the X520/540 is noticeably lower; jumbo frames need to be enabled to achieve an upload speed of 9800m.
Regarding the earlier tests of transmission speeds using a 10G network card on category 5e cables, see: using 10G network cards to test the transmission speed of category 5e cables, overturning your understanding. Now I can provide a supplement because I bought two Samsung SM961 256G SSDs from Taobao, which might fail at any time, but they cost less than 250 each, probably MLC. I transferred a 19G file back and forth without any noticeable speed drop. However, I discovered that the length of the previous category 5e cable was miscalculated; it is not 69 meters but 59 meters long, as seen from the numbers on both ends: 660m and 601m.
Using a WIN7 machine with a B150 motherboard and an I3 6100 CPU, I installed the SM961, and this new assembled computer runs WIN10. Both computers use the X520-T2 for transfer.
First, I used the 59-meter long category 5e cable to directly connect the two computers to transfer a single 19G file:
The speed was stable at around 1.05GB/second.
Then I switched to a 99-meter long category 6 cable, and the speed was as follows:
During the transfer to the latter section, there was a significant drop in speed, then it recovered. This cable’s length exceeds the technical specification of 55 meters by a lot, yet it still achieves such transmission speed, which surprised me. I wonder if it’s related to being bundled together. The earlier category 5e cable also exceeded 55 meters and still maintained full speed.
Using IxChariot to test the 99-meter category 6 cable, let’s see how the curve looks:
It feels like it’s breaking intermittently.
The 59-meter category 5e cable shows this curve:
Here, it’s not to say that category 6 is worse than category 5e, but there are distance limitations. Both category 5e and category 6 specify in the parameters for 10G network cards that they cannot exceed 55 meters; otherwise, category 6a cables are needed, but they cannot exceed 100 meters in length.
OK. Back to testing the soft router, LEDE allows you to easily set which port you want to use as WAN and LAN, just a few clicks of the mouse. All ports are as follows:
First, test WAN to LAN:
LAN to WAN speed was a bit slower; I guess I need to switch the two computers’ positions to achieve full speed, but I forgot to do that. Just posting the status diagram of the soft router:
Then is the LAN to LAN test (acting as a switch)
With two computers, if the direction is switched, the speed also decreases significantly, failing to meet the requirements. The above tests were conducted without enabling jumbo frames; if the MTU is set to 9000 in the LAN settings, the speed is even slower.
Next, I tested a 10G port against four 1G ports (four computers connected to four network cards).
Speed test results:
10G to 4x1G
4x1G to 10G
Each 1G network card computer achieved over 900m speed, regardless of upload/download.
I’m too tired; I didn’t conduct more tests, and the upload speed issue remains unresolved. If the soft router can enable jumbo frames, this problem could be solved. Or maybe switch to a network card that consumes less CPU resources? This 10G soft router will take a break for now. I still have several purchased routers that need testing. This machine has temporarily achieved the desired effect.
I really can’t stay up late anymore; health is important!