Wi-Fi роутер Linksys WRT54G2 (Отчет) (2/5) / Інформатика та побутова електроніка

Benchmarking WRT Firmware... Some Surprises!

This is a benchmark I've been wanting to do for ages.

Results

Sorted by LAN-to-WAN speed performance, wired only.
All numbers in Megabits per second.

Linksys 4.71.4.001 (current from Linksys)
MIN: 47.8 AVG: 50.86 MAX: 52.5

Tomato 1.22
MIN: 40.8 AVG: 45.21 MAX: 47.4

OpenWrt Kamikaze 7.09
MIN: 35.5 AVG: 35.69 MAX: 35.9

OpenWrt Whiterussian 0.9
MIN: 35.0 AVG: 35.21 MAX: 35.4

DD-WRT 2.4 std (not sp1)
MIN: 31.4 AVG: 31.63 MAX: 31.9

OpenWrt Kamikaze 8.09RC1 (2.6 kernel)
MIN: 23.7 AVG: 24.01 MAX: 24.2

Test Details

The device is a WRT54GS version 3. This is essentially identical to the WRT54Gv4 and WRT54GL except it has more memory. No firmware used the extra memory to my knowledge and standard firmware was used.

Benchmarking was done by iperf version 2.0.2 using default settings except for window size (which I elevated on the windows client to the Linux default of 85.3k). Server on the WAN is my Ubuntu 8.04 server and client is my Win2K/SP4 main machine. The two units average around 81Mbps when directly connected on the same LAN. They're not screamers by any means but this is a 100mbps LAN and I was quite impressed with iperf. This reference speed is important since it should be significantly higher than the devices being tested and is.

Exactly 10 tests, each 10 seconds long, were performed and MIN, AVG, and MAX reported.

Every firmware defaults to a WAN getting connection via DHCP and a LAN on address 192.168.1.1. The benchmark was run exactly like this WITH NO CHANGES. Wireless was never used or even looked at (my next project, perhaps?). I'm sure I had open wireless several times.

To be extra cautious, every firmware initialized its own NVRAM because I manually cleared it first. Those familiar with NVRAM know this is, essentially, the same as "factory defaults". Different firmware calls it different things but the important thing was to negate ANY interaction between settings from one firmware to another. They all started with a clean slate (and clean NVRAM)

I wanted to use the latest of the popular ones but ran into a snag with DD-WRT 2.4-SP1. When flashing from OpenWrt I strip the *.bin header creating a flash memory image known as a "trx" file. In the case of DD-WRT 2.4-SP1, OpenWrt didn't like the format. The exact filename of both 2.4 and 2.4-SP1 is "dd-wrt.v24_std_generic.bin" (the same!!). There may be something I don't know about DD-WRT versions but the trx check scared me - I don't need to be de-bricking.

Be assured I closed all web pages and SSH sessions to the device when I tested it. What can I say... I'm a thorough engineer.

I started wanting to only benchmark the new OpenWrt 2.6 kernel. Notice it sucks? BTW, *ALL* these firmwares are a 2.4 Linux kernel except the bleeding edge OpenWrt Kamikaze 8.09RC1. Keep in mind the 2.6 kernel, itself, probably has more overhead than 2.4. But one thing I really like is the Broadcom stuff is all open source - no others can say that at this time!

Conclusions

Some may consider going back to Linksys firmware!

Tomato users should be happy but remember, this is DEFAULT (no bells and whistles turned on).

OpenWrt users may think twice before venturing into Linux 2.6. I still love my OpenWrt Whiterussian 0.9.

DD-WRT didn't show as low as I thought it would, even at factory defaults. I've never been a fan of DD-WRT for several reasons - one is slow config page response but the penalty isn't seen in default throughput.

Everyone should be aware there are trade-offs between features and speed. Add more features, like a QoS or logging, and things may slow down quite a bit.

In other words... all bets are off if you turn on things above default settings. Wish I had the time to explore all these but the data would be HUGE.

Of course, anyone with a 6mbps ISP connection will see very little difference between these firmwares running default since their throughput exceeds the ISP speed by quite a bit. As ISP speeds go up this hardware will look very old. Aging hardware is probably the biggest strike against a WRT54GL these days.