Slashdot Mirror
Harald Welte Calls Out Netgear's Open Source Sham
Simon80 writes "Harald Welte, known for his involvement in various open source communities, has pointed out the shortcomings of Netgear's open source router hype. Netgear's own astroturfed community site reveals that the router requires the use of binary-only kernel modules for the wireless and ethernet hardware, which is supplied by Broadcom. Also worth noting are the missing features in third-party firmware versions supplied by Netgear."
One of the open firmware shortcomings is "WPA and WPA2 are not working." That is a pretty big shortcoming.
This post climbed Mt. Washington.
The very same Broadcom blobs are included in dd-wrt. It must also be noted that dd-wrt is supposedly GPL software, yet the evidence in SVN clearly shows that a large portion of the code is Copyright evil corporations such as Intel and Microsoft and that these corporations have NOT given permission to use the code under the GPL. It is in many cases not even clear if they give permission to distribute the code at all.
9/11: Never forget it was a false-flag operation
The WRT54G series all use Broadcom chips pretty much identical to the ones you'd find in Netgear routers. See here:
http://www.dd-wrt.com/wiki/index.php/Supported_Devices#Netgear
I am not proud to admit this, but I took a CCNA years ago, and I've built literally dozens of wifi networks using various combinations of off-the-shelf (or off-the-refurbish-list) routers and stock/modified firmware. I am a minor authority on the subject of cheap-ass consumer routers.
Broadcom is what you'd call a "fabless semiconductor company," which means they design chips but don't actually manufacture them. Almost all consumer routers you can find today use Broadcom-based system-on-a-chips, which consists of basically a CPU, flash and DRAM, ethernet interface, and half a wifi-radio, all crammed onto a single CMOS.
Broadcom designs the chip, someone else leases the design for the chip (and all the accompanying drivers) from Broadcom, then the person that leased the chip pays a third person who owns a CMOS fabrication plant to actually manufacture the chip. Then the chips get sold to yet another party, like Linksys, Netgear, Trendnet, Asus (my pick!), Buffalo, and others. The chip has several dozen wires hanging off the end of it, and someone connects them to various external ports or devices on the router: ten wires make a bank of five Ethernet ports, two or four wires are connected to one or two antennas (more if you have MIMO), more wires are connected to the status LEDs and buttons, et cetera. The end manufacturer is also responsible for providing firmware, which historically they've done by combining Broadcom's drivers with some code they ripped off from the Linux kernel (some manufacturers, like Asus and Buffalo, are reputed to be good about providing source code when they do this). Then they put it in a box with a compatible power adapter, slap a lame warranty on it (because many governments and retailers require them), and sell it.
The end result is that pretty much all the routers you can buy are nearly identical in every way except firmware. Furthermore, almost no manufacturer can actually be bothered to provide long-term support for these routers (why fix a broken routers when they can sell you a new one?), and since firmware development is by far the most difficult and expensive part of what the end manufacturers (eg Linksys) actually DO, this is the area where most consumer routers really fail.
(The other problem is that most Broadcom chips only have about ~100 MB/sec of memory bandwidth on chip, tops, which is obviously less than one gigabit per second (~125MiB/sec). This means that there are no consumer routers you can buy that are actually capable of routing a gigabit of traffic per second- at best they all seem to crap out around 160 megabits per second, in my experience (note: you have even less bandwidth when traversing the NAT gateway, particularly with traffic shaping enabled). This is mostly a limit inherited from the CMOS manufacturing process they use, I think - it's the same process they use to make DRAM and flash, and while it's cheap relative to the number of transistors you get, the resulting chips are rather slow compared to what you get with optical lithography.)
As for your grandfather's router, I suggest you try running BitTorrent on a computer connected to it, and see what happens when you quickly spawn hundreds of new TCP connections. I'm betting it'll choke, because the onboard NAT has to keep track of each individual TCP connection, and your $20 Trendnet router (which is probably quite old indeed, regardless of how recently you purchased it) probably isn't expiring old TCP connections for a good 12 hours. There's probably a way you can set the NAT TCP timeout value to something more reasonable, like 15 minutes (if it's not in the web-based interface, try downloading the config file and editing it with a text editor - I shouldn't have to tell you the risk from doing this). You can also look up DD-WRT,