XORP 1.0 Released

Mark Handley writes "XORP is the eXtensible Open Router Platform - an open-source router software stack for FreeBSD and Linux. It's designed from scratch to be extensible, so you can write your own router applications that play nicely with the existing routing protocols. We just released XORP 1.0! There's also a Live CD if you want to try it out without reinstalling your machine. More details in this CNET article."

Comments

License vs Proprietary forks

[Gopal.V]

For the time I've looked at routers (briefly) , I've already noticed the BusyBox Hall Of Shame - where router vendors have refused to comply to the license. But I sincerely fear that all this work might get "embrace and extend and sell" by a company - like what happened for the BSD TCP/IP stacks (ok, do an nmap -O on your favourite MS box).

But this is good for colleges and other places where the concentration of "guys who can stop by and fix the router" is high. Also not to mention the tinfoil factor of a readonly-livecd router (but does it have remote logging).

Re:License vs Proprietary forks

[ffsnjb]

But I sincerely fear that all this work might get "embrace and extend and sell" by a company - like what happened for the BSD TCP/IP stacks (ok, do an nmap -O on your favourite MS box).

The BSD folks look at this differently than you do. As long as Microsoft complies with the license for the BSD tools they use in Windows (tcp/ip stack, ftp.exe, etc.), and they have, everything is just peachy. BSD folks just love to see their code used (superiority complex?), and generally don't care as long as the copyright remains; even if every penguin-worshipping-codemonkey's most hated evil empire uses it.

A welcome development

[Tandoori Haggis]

The cost of traditional routers has been a problem for commercial and non commercial organisations with tight budgets. Even in larger organisations with considerable routing investment there can be a real lack of routers for spares and training.

Add to this the fact that some old but worthy routers may not run the latest IOS and hence may have some unpatched vulnerability, which will not be addressed by the manufacturer, who understandably wants you to buy their latest kit.

As the CNET article points out the perceived disadvantage of this open source router software is the performance of the physical platform.

Perhaps it is the hardware implimentation that made people unsure of the Open Router project?

Perhaps a respectable router can be built more easily and for less outlay now than 5 years ago?

Gated skirted over

[Anonymous Coward]

Gated got a mention in the CNET article. It was actually a very good alternative to the commercial vendor hardware - argueably many ISPs (that are around today, eg Demon Internet in the UK) would never have been able to start up if it were not for the existance of Gated.

Unfortunately this was an example of a kinda BSD-style licence causing problems. The Gated community (which cost a couple of thousand pounds per year to join, providing access to all code/updates) added a great deal of functionality to the code, and benefited from this.

The whole code was subsequently taken and "owned" by NextHop, meaning most/all of these code benefits were lost. Take a look at the gated.org website to see what happened :(

- Ivan

Performance is pretty reasonable

[Fzz]

I don't have results for a new machine with PCI-Express, but a regular 1GHz-class x86 PCs with 32 bit PCI tops out at about 400K minimum-size packets per second. This is limited by PCI saturation - you get fairly low PCI utilization with small packets. But even so, a $300 PC compares favourably with something like a Cisco 7206VXR (which cost ~$30K about 3-4 years ago). This is assuming you are smart about using interface polling rather that being interrupt-driven. Otherwise you die from interrupt livelock.

This is plenty fast enough for most edge routers, but clearly not going to compete with a Cisco CRS-1 or Juniper core router.

But most of the software in a router is control-plane (routing protocols and the like) and this is what XORP has focussed on to-date. As more people get involved with the project, we'll be able to do more things.

A decade ago no-one thought we'd be running Linux on a supercomputer. But we are. If we can get to the point where XORP is stable enough and fully featured enough for carrier-grade routers, who knows what hardware people will run it on in a few years time.

We are however very committed to keeping XORP as an open-source platform. No matter who uses it commercially, in the long run the only way to open up the router software market is for many boxes from many vendors to run a common open base software platform. With luck and with a lot of help, maybe that can be XORP.

- Mark Handley, XORP Project

Re:Performance is pretty reasonable

[!ramirez]

The problem, IMHO, is that ALL high end routers use HARDWARE routing (see: flow/fast switching in 7500/12000s) instead of software routing. Unless you 're building ASICs to handle stuff in the data plane (VIPs or whatever the 12ks use for dCEF and the like), you're not really in any danger of becoming used by the higher end routing equipment manufacturers.

Also, software can be written all day long to run on a general purpose PC - but when was the last time you saw a multimode ATM DS3 interface, or a multichannel T3 interface for a PC?

Finally, I'm not convinced that opening the router software market is a 'good thing' - Linux advocates continually talk about 'software monoculture' being a bad, bad thing, as evidenced by Microsoft, yet you speak in your last paragraph of wanting precisely this. Imagine a flaw in the way you handled PIM (a la Cisco about a year ago), which would effectively DoS the router. Cisco at least notified their large carriers before specific details leaked onto the net - I shudder to think of someone posting 0day exploit code for something like this on Full-Disclosure.

(incidentally, I find the project fairly interesting, with the noted lack of redistribution ability. :( Hopefully, this is something that will come along in the next version (maybe?).)

Anatomy of a Slashdotting

[Fzz]

In case you wondered about the wisdom of linking to an ISO, here are our traffic stats.

www.xorp.org is in California, www2.xorp.org is in London. Both are 6-year old dual 450MHz Xeon machines with 768MBytes of RAM and SCSI disks, running FreeBSD and Apache 1.3.x. Both machines have 100Mb/s access to the Internet.

In 5 hours:

www.xorp.org: transfered ~30 GBytes peaked at around 175 simultaneous httpd processes 15 min load average peaked at 0.7. www2.xorp.org: transfered ~20 GBytes peaked at around 75 simultaneous httpd processes 15 min load average peaked at 0.4. Aggregate bandwidth was ~25Mbit/sec average. I won't know the peak bandwidth without some more analysis, but it's obviously quite a bit more than 25Mb/s. I didn't notice any obvious slowdown on either machine.

I've no idea how typical this is, but I'm always curious about how easily sites seem to die due to slashdotting.

- Mark