Slashdot Mirror
End Of the Line for SpeakFreely: NATed to Death
Arun writes "John Walker (of AutoDesk and Fourmilab fame), primary author of SpeakFreely, has decided to EOL the program (a pioneering network telephony effort), come January 15th, 2004. He cites difficulty in maintaining a decade-old code base, lack of appropriate developer support and a fundamental change in the peer-to-peer nature of the Internet upon which SF is dependent as motivating factors behind his decision. While the last release of the program will continue to be available from SourceForge, the main web site, mailing list, and web forum will be shut down on the aforementioned date." He's got some good points too, like how once IPv6 is more common, most users probably won't go back to one address per machine. I know I enjoy the added security of a NATed firewall, and without a really good reason, I won't be quick to give it up.
Why did I discover this cool application in a discontinuation announcement?
I wish I had discovered it earlier.
Oh well, I can only hope that I can repent this mistake in my next life.
It is not a matter of (just) static port mapping, it is more a fundamental problem in the way DNS works with Internet addressing -- or more specifically, the way way applications interact with Internet addressing. (This will no doubt invite flames from those outraged at the idea that there might be a fundamental problem/mistake in the Internet.)
More specifically, what happens when you have multiple machines behind the NAT device? How do you map the ports statically to multiple machines *and* also communicate this information to devices on the outside of the NAT device? (That is, port 80 on the NAT device maps to server1, port 81 on the NAT device maps to server2, etc.)
The key issue is that applications are using network level addressing (IP addresses) rather than application level addresses (URLs) to establish the network connection -- we have network specific information far too embedded in the applications, which is why the transition from IPv4 to IPv6 is such a nuisance. At the moment, the DNS SRV record could help with some of these matters by specifying a port number to use for a specific service and host/domain.
A better design for applications would be for them to be completely unaware of 'IP addresses' and function purely on URLs or hostnames + service name, and link to libraries or network drivers on the machine that handle the network aspects. Really -- excepting network mangement tools, what application bothers about the MAC addresses of machines or PPP negotiation details? IP addresses should not matter to the applications, either -- at that point, much of the arguments against NAT go away.
Honestly, the fact that NAT causes applications to break is more a reflection on mistakes in the architecture/application. IP packets themselves don't fall over and die just because they transition from a PPP link to wireless to ethernet to SONET to etc. The differing layers are independent of one another -- the applications have not yet been weaned off directly diddling with the IP layer.