Slashdot Mirror
Linux Based Router
Troy Larsen
sent us a link to a yahoo press release proclaiming a company
selling a Linux Based Router.
Up to 40 gigs per second and up to 128 100mb ports. Yum.
← Back to Stories (view on slashdot.org)
← Back to Stories (view on slashdot.org)
> That and some recent NFR results showed that
9 99/Feb/0110.html
> linux dropped 90% of the packets on a 47mp line
> whereas the BSD's dropped 5% and Solaris on an
> Ultra1 dropped > 10%.
NFR is written specifically for the Berkeley Packet Filter using read(), which Linux doesn't implement (Linux supports filtering but doesn't buffer the incoming packets to allow read()). End of story. Why don't you compare the speed of Linux 2.2 DECnet versus the DECnet in the latest OpenBSD while you're at it?
However, if you look at the latest patches from Alexey Kuznetsov on:
ftp://ftp.inr.ac.ru/ip-routing/lbl-tools
you will discover a kernel patch and a patch to libpcap labeled "turbo". This patch implements packet capture via a ring buffer shared between kernel and user space. This is something even BSD doesn't have- they must go through an additional buffer copy in the read() call.
This patch also eliminates 2 out of the 3 main problems with the Linux packet capture code outlined in this message from the NFR mailing list:
http://www.nfr.net/nfr/mail-archive/nfr-users/1
The remaining problem, #3, may or may not still be there, but only involves copying a few bytes here and there and so isn't a very big deal if it does exist.
It will be interesting how the different OSes stack up doing packet capture in the future.
I seem to recall that someone was building a Linux "router distro" based on Debian. Is that this company? I also hear that Debian thinks it is good for people to be taking their main distro and tweaking it for various special purpose apps. Are there any others I'm not aware of in the pipeline?
That wasn't FUD. He said "both are good" (after quoting a sarcastic Linus comment, granted).
As for the comparison, I share his experience. One of my boxes is an ancient 90 Mhz Pentium running Linux 2.0.x, and it has no problems at all saturating my 100mbps ethernet. E.g. I use it to burn CD's from a master over the net at 4x, which I believe is pushing the limits.
I've kept the network and that box otherwise fairly quiescent while doing so, figuring there's no point in asking for coasters, but still, it's handling the net, the scsi device, and the CD burner software all at once.
I personally have no idea if FreeBSD/NetBSD/OpenBSD are better or worse in these regards, and it doesn't matter to me (except for being interested in knowing when to recommend one thing or another for various purposes). If Linux didn't exist, I'd be using BSD; I have a personal interest in it.
Professional Wild-Eyed Visionary
I've been using, installing, and recommending Xyplex equipment for years. It is without exception well-designed, intuitive, and based only on open standards. They're a good company with good products - it's a shame that they have to live in Cisco's shadow. I'm excited about this new product and hope to get a chance to tinker with one soon. There are just so many possibilities this opens up...
Tired of FB/Google censorship? Visit UNCENSORED!
I've always been perplexed by people saying "xxx distribution is ok to start with, but no one else should bother with it." (Usually directed at RedHat.)
Let me freely admit that my main Linux machine was originally a RedHat box. Why? I was new to installing linux, for starters. But I also wanted something I could loan to interested friends and relatives that they might be able to install on their own.
I have never bought another distribution. I may pick up Debian or Slack some time in the future, just to compare. But here's my point: My box is Redhat originally, but I feel I have become an intelligent user (compiling the kernel and other software, learning how to manually setup services, boot configurations, etc.). Why would I now go out and buy any other distribution as a "graduation" from RedHat? My machine is my customized version, I have already gone beyond any distribution in terms of what works for me.
(I can see for specialized uses like LRP it is quite a good idea to use another distribution.)
+LO
A reasonably priced switching fabric under the control of open source code which and have modules added to it could be very cool.
Existing Linux based routers are great for the low end, but they can only scale to a certain point because the CPU is involved in all packet forwarding.
In something like this, I/O porcessors, or some specialized ASICs do the scut work of moving data around making low level routing decisions. The OS and CPU only deal with managing the switching fabric. They generally only get involved when things change, such as when a packet comes in for a new destination, or when a route to a destination changes, etc.
Having this management layer available for manipuation could enable linux developers to bring out a new class or network applications.
For example, if this hardware is cheap enough and presents clear interfaces, it should be relatively easy to implement a high performance URL aware HTTP load balancer which could eclipse the performance of any sort of software/general-purpose hardware solution by multiple orders of magnitude.
I was going to say that this would be pretty expensive, but given that this router lists at $300,000, the I2O route could be quite practical.
The I/O processor would forward packets based on a local flow cache. Packets not matching the flow cache would generate an event which would be handled by a module running on the linux OS.
Incoming HTTP requests could be forwarded to a linux module which would spoof the handshake to get URL information. It would then open a connection to the apppropriate web server, populate the flow tables with the appropriate entries and then hand the flow back to the IOPs. The IOPs would forward packets and rewrite headers as needed.
Maybe we can use it to convince Cisco to hand out the source code for their 12.X IOS. That is some code I would like to get my hands on.
... exciting, but not ready for primetime.
Unfortunately, the product doesn't seem to be up to snuff yet. No talk of OC-48 Packet over Sonet, or even OC-12 POS, no talk of GigE, no mention of BGP??? Looks like one to keep your eyes on and watch how it develops
You were led to believe correctly. The deeper question that yours leads to is whether or not any operating system's IP stack is up to speed with a custom-designed solution. Juniper, I know, is making some gigabit-class routers based on the BSD kernel, but nobody has been able to take away Cisco's IOS market share yet ... is it worth our time to try to route with OS-based systems? Is a hacked version of routed or gated robust enough to pass 20 or 30 gig across a switching and routing backplane? More power to them if they can, but i have doubts.
pardon me replying so much, but I work on Ciscos all day and am interested in any other players in the market. A couple of the issues I see are:
-- express/optimum switching. Here, we are using the first packet headed toward a destination over a certain port to evaluate through the router's access lists, and then allowing the remaining packets toward that destination over that port to flow through the switching hardware without being evaluated by the processor? Is a BSD or Linux-based gated able to handle the logical concept of flows to optimize access lists and route processing?
-- nifty features: yes, they're standards based, but does the gated gsr support soft inbound and outbound soft reconfigs on bgp? What about nifty things like HDLC, which beats the hell out of PPP. ISL trunking between your switches and routers? Fast Etherchannel? Fast drops on access lists, which saves immense amounts of processor time when you are writing a smurf filter. Easy disabling of directed-broadcast, while we're talking smurfs. Rate limits on circuits?
What I'm saying here is that a BSD or Linux based router is going to have to sell itself to a lot of people who maintain internet backbones for a living. It is going to have to have a feature set that meets or exceeds our current vendor's. While I can see the applicibility of a BSD or Linux based router on the low end, I have yet to see an entry that I would trust a nationwide backbone to.
A little more than processing OSPF maps is taken care of in software:
-Processing of access lists
-Managing switching path logical connections
-Maintaining all processing for your Interior Gateway protocol(s) of choice
-Maintaining all BGP route maps, route tables from upstream neighbors, route reflection client or server processes, metric processing, etc.
-Maintaining the master route table, where BGP and IGP routes are held and routing decisions are made
-Any and all network management
-Any and all console diagnostics, line card monitoring, environment monitoring, power monitoring, etc.
-Handles any high-level encapsulation, especially tunnels and encapsulation of Appletalk, DECnet, IPX, X.25, etc.
That strikes me as a little more than a configuration front end.
Here's a question for anyone who's worked with both Linux and IOS:
Would it be possible to compile Linux to run on the Cisco equipment? IOS supports downloading a new "Flash ROM" for IOS updates, so I'm guessing that Linux could run on a Cisco router/switch/hub for a lot less than the several-thousand-dollar IOS software?????
Please enlighten me...
LOAD "SIG",8,1
LOADING...
READY.
RUN