Slashdot Mirror

Hmm I could have sworn that I have seen several companies who are doing exactly that. Offering open source routing platforms to take on Cisco and other high end routers.

Vyatta http://www.vyatta.com/
XORP http://www.xorp.org/

Plus there was Alteon who was bought out by Nortel that was doing a lot of open source stuff. I worked with them before they got bought out. I also know of a few other companies around the same period who were open sourcing the software that ran on top of their hardware as well.

You might be surprised at how much stuff out there is open source because they are selling you the custom hardware (ASICs/whatever) not the software that runs on top of it.

I believe you may be onto something that would suit my needs for this Wireless ISP perfectly.

Do you know of any commodity TCAM hardware?

Maybe i can run RTLinuxFree on it?

http://www.rtlinuxfree.com/

and then i could put XORP on top of it

http://www.xorp.org/

or... i answered my own question...

intel xscale cpus and powerpc based routing platforms (most single board computers) are all i really need.

http://gateworks.com/
http://routerboard.com/
http://www.adiengineering.com/

RB1000 from Mikrotik supports 400,000 pps and 4 gigabit ports for $700

Take that Cisco

I'm thinking along the lines of a Nvidia Tesla running http://www.xorp.org/ - any ideas?

Maybe Vyatta @ http://www.vyatta.org/ does what you want. I really don't have any idea what that is from the actual post, tho. You need some routing for thousands of users, and can't afford a Cisco UBR. I'm not sure exactly if you wanted to use the UBR for DOCSIS type support for some reason (a la cable modem) but the fact it'll be wireless leads me to believe it won't be. I'm assuming you don't need a lot of physical ports, just something to manage your VLANS, some routed subnets, a bit of BGP, etc. Maybe XORP is what you want, tho @ http://www.xorp.org/ so you may want to look there. IHeck, 'm not even sure if you want to take a server with a bunch of PCIe ports and slam multiport switchable fabric cards in there like the ones DSS @ http://www.dssnetworks.com/v3/gigabit_pcie_6468.asp makes, or do something else. Maybe these links will help, and hopefully there'll be a detailed followup so we can aim at the real target :)

I don't hate Google like I do Microsoft. I staunchly disagree with Google's censorship of information in China, but, Yahoo does it too so that is not reason alone to hate either of them. I hear people grousing about Google's "monopoly." No, you have a number of choices: Yahoo, Altavista, Lycos, and Webcrawler (note: I am not endorsing any of these.) This is quite unlike the Microsoft of the 1990s. Linux was still quite immature and you really needed a stronger compsci and UNIX background. BSD was and still is a viable choice but it really took more advanced users. As much as I hate to admit, Microsoft was unfortunately, the only real choice for the non technically savvy until recently.

So, why do I hate Microsoft? They stifle innovation under a pretext of encouraging it. As other Slashdotters have noted, Microsoft takes the embrace, extend, and patent attitude towards open source. This is what happened with Kerberos and the infamous PAC. They extended the olive branch to MIT then effectively changed Kerberos enough to make it their own. If that wasn't IP theft, it damn well should have been. Beware of any project sponsored by Microsoft as, "the appearance differs from reality." My eye is presently on the XORP Extensible Open Source Router Project as Microsoft has taken a keen interest. Fortunately, there exists an implementation of BGP and OSPF that has been around longer than XORP and already outperforms it. See the OpenBSD project. Google, thus far, hasn't behaved quite like Microsoft; the coming years remain to be seen.

Eddie Kohler, whose PhD thesis at MIT was the Click modular router (which from what I understand turned into the "engine" behind XORP), is one of the principal designers and developers of XORP. They published a paper at NSDI last year, which you can read here (Warning: PDF). It states very clearly what the goal of XORP is, and how well it performs. Quite interesting.

and then there is XORP

From the XORP FAQ:

Software forwarding plane limits forwarding rate

For small workplace settings / lan routing where a little added latency and limited bandwidth is not an issue this is a great solution and companies like cisco will probably see their market share eroded a bit.

However, backbone routing, especially anything with packet shaping, is not going to be software only driven for a while simply because the customized ICUs in routers are so much better at this then commoditiy hardware.

check the faq :)
http://www.xorp.org/faq.html#pronounce

How about reading at xorp.org, instead of just news stories that talk about xorp?

If they're smart, they are. Microsoft or Sun could use some really top-notch router software, some hardware and their sales force to really hurt Cisco. Xorp has a BSD-style license, so Sun and MS will have no problem taking it proprietary when it's good enough.

Routers from Joe's garage are never going to be a threat, but routers from Microsoft or Sun would be an instant threat, when Xorp becomes as useful and reliable as IOS.

see xorp's website

It's BSD Licensed, so Yes, MS could take and use it, much like their TCP/IP stack.

Check out the xorp project then, xorp includes a version of bgp and compiles on both freebsd, linux and macos.

I've actually found cases where the SMP on FreeBSD 4.x was superior to Linux. In order to test Linux's networking performance vs. FreeBSD, I changed a program for an algorithm to just run in the background calculating the Ackermann Function.

Anyways, the base rate was to run two Ackerman's at once, thus causing 100% USER CPU usage on both CPUs. The base rate for FreeBSD 4.62 was 15.5 Ackerman's per time period, vs. Linux's v2.4.18 14.0 during the same time period. Now this isn't a smoking gun, but the hardware was identical, and they were both running on custom compiled thin as possible kernels under the same duress.

Why would anyone do this? Well, my goal was to eat up all USER CPU and see how much I could rob from user with system under severe network abuse. Needless to say, that both OS's did very poorly, with FreeBSD having a clear edge, when the interface was brought to promiscuous mode to listen to a packet flood. FreeBSD degraded less, but in both cases an almost useless amount of CPU was left over for USERland. FreeBSD with RX polling turned on - a feature that practically seems unique to FreeBSD, from the XORP router project. I am aware of polling endeavors in Linux but was never able to get them working. As usual with FreeBSD, 'features' aren't creeping in, so they tend to work. I even changed the polling to work under SMP (it wasn't designed to) and it worked in a situation where it shouldn't have. The usefulness of RX polling cannot be stressed enough, its imperative to consider the live-locking of interrupt driven kernels when dealing with massive amounts of bandwidth. If interested, see: 'Eliminating Receive Livelock in an Interrupt-driven Kernel', USENIX 1996, its amazing to me livelock still happens over 5 years after stuff like this gets presented to the public.

So, how bad is FreeBSD SMP? As far as I was concerned in my test, 2.4 Linux SMP seemed inferior (in my case) to FreeBSD on identical hardware. Are people touting Linux's big bad SMP zealots. Most probably, most good kernel hackers think highly of FreeBSD, particularly the VM. I find it amusing that RedHat is not porting to SPARC or Alpha anymore, and yes FreeBSD 5 is planned stable on IA64, IA32, SPARC64, PowerPC [stable planned a bit later, probably when a real PPC gets offered by IBM - die Motorola PPC, die] and Alpha. Clean code and standards compliance begets portability.

As far as saying "SMP" is better. Linux may have a better approach, but like my example, and I am sure there are others, empirical tests say a whole lot more. It's important to keep the machinery well oiled and coherent, which is something I think FreeBSD does rather nicely. Empirical tests such as mine prove that approach and theory and real life are different.

FreeBSD - it's coherent, well documented, "thin," bloody fast, BSD licensed so call it your own. You can see that well written code goes across architectures; the FreeBSD discipline is allowing them to easily stay stable on several platforms. I have run several tests that suggest that even FreeBSD 4.X is 'better' than Linux at various things, let alone 5.0. The VM subsystem is superior [2.5 is catching up]. Most big companies provide virtual servers with FreeBSD, such as Verio. The biggest irony of all is how small the FreeBSD community is compared to legions of hackers and companies trying to improve Linux. Yet why is Linux fragmented so horribly? You will eventually come to understand why this is the only free and open commercial grade OS there is. You will know what you are missing when you finally get a coherent UNIX. GCC, the C library and the kernel are all a matched set, not of this he said she said GNU-of-the-day distribution crap or fake compilers from RedHat and frozen broken CVS snapshots of the C library [RedHat again, with a fake C-lib on RH8]. FreeBSD is used by Juniper as the core OS, with network processors instead of 'real' network cards. It's beautiful. A full version of FreeBSD, relabeled JuneOS, with an IOS-like CLI for those who need it and superior design and interfaces. The UFS2 filesystem is also incredible. I really, really like XFS for Linux, but the Linux kernel maintainers won't merge it in [to 2.4] but have a myriad of vastly inferior filesystems merged into Linux [ext3 fake journaling, Reiser fsck for fun FS, JFS which is robust but slow]. RedHat's refusal not to embrace XFS with open arms boggles my mind. UFS2 addresses this problem. A fast, robust logging filesystem that is stable and in the kernel. I think UFS2 is a far superior improvement to UFS than was ETX3 to EXT2.

Anyways, I don't think I'll wait for Linux kernel 2.6 or any of the flavors of Linux distributors to come out with something stable, well documented, coherent with UNIX as a standard and each other. Don't be fooled, LSB is a standards base, but you don't get decades of discipline, you get maybe a years worth of un-actualized planning. FreeBSD 5.0 is here. This project needs a better installer, and some 'for workstation use' cleanups, and probably a better package system, although, there are lots of people who like PKG and PORTS much, much better than RPM or DEB. Another annoying omission [and yet another Sun self-screwing maneuver] it that it is difficult to get a JRE/JDK to run natively [1.1, 1.2, 1.3 and 1.4 are available as ports] and Sun does not provide one [they are apparently planning one]. People have lots of luck though using the Linux binary emulator, FreeBSD can run everything Linux does in binary form and it's easier to port to. Another good reason to develop for FreeBSD is this: Linux has /usr/include/linux. That in and of itself is a reason not to start there for development work. World, see a more beautiful future, one which was paved with the golden road made of FreeBSD - Certainly FreeBSD has a place, and in my opinion it clearly deprecates Linux in some situations. Particularly if you need to have a nice server box stay up forever or stay GPL-virus-free. [this said affectionately, I like the GPL, but you may not be able to afford giving your intellectual property to the world but would like to contribute in some way nevertheless. If it's a non-novel concept, the "community" will just implement it out of need/demand, if it's too difficult for the hackers to trivially add, then it might just be worth calling intellectual property.]