Want to read Slashdot from your mobile device? Point it at m.slashdot.org and keep reading!

 


Forgot your password?
Close
typodupeerror
×
Networking Linux

Linux and Multiple Internet Uplinks: a New Tool 80

Posted by Soulskill from the fault-tolerance-for-all dept.
New submitter Alessandro Zarrilli writes: Linux has been able do multipath routing for a long time: it means being able to have routes with multiple gateways and to use them in a (weighted) round-robin fashion. But Linux is missing a tool to actively monitor the state of internet uplinks and change the routing accordingly. Without it, from a LAN perspective, it's like having a RAID-0: just one uplink goes down and all of your LAN-to-WAN traffic goes down too. Documentation and examples on the subject are lacking; existing solutions are few and deeply integrated in firewall/routing specific distributions. To address these issues, a new standalone tool was just released: Fault Tolerant Router. It also includes a complete (iptables + ip policy routing) configuration generator.
This discussion has been archived. No new comments can be posted.

Linux and Multiple Internet Uplinks: a New Tool More

Linux and Multiple Internet Uplinks: a New Tool

Comments Filter:

  • Strange (Score:5, Interesting)

    by ledow ( 319597 ) on Tuesday March 03, 2015 @04:15PM (#49174687) Homepage

    Strange.

    I was using routing patches to Linux nearly 7 years ago to do this (admittedly it wasn't in the stock kernel, but the patches weren't huge)... you were able to specify multipath and multiple gateways and if one route went down, the others were prioritised and would take over, and also your upstream etc. were balanced properly and took account of failing routes automatically without any kind of daemon etc. running.

    I ran a school off multiple ADSL and even 3G connections with it - the only manual maintenance I ever had to do was to put the ADSL modems onto a SMS-controlled relay (SMS came in on the same 3G stick!) because our ISP would often give us "dead" sessions if they'd had problems (where you'd get PPP and an IP and a remote gateway but couldn't do anything across them) and we were then able to manually reset if necessary. My bursar and I used the system for five years like that, only ever resetting it to enable VPN when all the upstream routes had got dead sessions, and that less than once or twice a year.

    And, no, we didn't have to do much. It was a stock Slackware install with one set of patches to a (2.6?) kernel to enable the multipath routing etc. Pretty well advertised at the time, one plain page of simple patches (I remember porting them myself to a newer kernel version, just before the new diffs came out), I'll try and dig it up.

    And "RAID-0 for upstream"? Bollocks. It "just worked" whatever interfaces were up (proven by it would even include the 3G PPP interface whenever it came up, and that only came up when we manually instructed it to connect as it cost money).

    Not saying this isn't good software, but it's by far not the problem the summary purports it to be, not a first by any means, and certainly not "new".

    • Re:Strange (Score:5, Informative)

      by ledow ( 319597 ) on Tuesday March 03, 2015 @04:20PM (#49174739) Homepage

      Think it was:

      http://www.ssi.bg/~ja/ [www.ssi.bg]

      Seems to still be updated.

    • Re: (Score:2)

      by TopherC ( 412335 )

      I wonder how this is different from channel bonding / link aggregation? I looked into this a few months ago and don't remember all the details but there's a "bonding" kernel module, which can run in some modes entirely in kernel space, or in a user-space-assisted mode. There is a round-robin mode but there are several others that include fault tolerance and load balancing. LACP can be used in cooperation with other network elements including switches if you want something that spans a local network.

      I had li

      • Re: (Score:2)

        by ledow ( 319597 )

        LACP would, indeed, fulfill the purpose but relies on you being able to obtain LACP support on upstream connections from your ISP. LACP must be enabled and known about on both ends for it to do anything.

        It's not always true that you could get support on upstream connection, but they are many, and multiple, types of bonding that provide similar facilities.

        However, in terms of being able to get disparate connections that can be conjoined without specific support on the other end or high-end hardware, there a

      • Re:Strange (Score:5, Informative)

        by msauve ( 701917 ) on Tuesday March 03, 2015 @05:28PM (#49175411)
        Link aggregation works at layer 2 (e.g. Ethernet). Basically, make multiple cables between 2 devices look like a single one.

        Multipath routing works at layer 3 (e.g. IP), you can send outbound packets to multiple routers for further forwarding. It works when there are "real" routes between the hosts (i.e. not behind NAT).

        This is one step beyond that, since it also does connection tracking and will work with outbound port NAT, so you can have a private network connected to multiple ISPs.
      • Channel bonding is a different thing, but you need upstream support to let it work. It's for example what these guys are doing (letting you connect to their server on the cloud): http://speedify.com/ [speedify.com]

    • Linux has been able do multipath routing for a long time: it means being able to have routes with multiple gateways and to use them in a (weighted) round-robin fashion. But Linux is missing a tool to actively monitor the state of internet uplinks and change the routing accordingly

      No the author is lacking knowledge of such tools. There are at least three major implementations of common routing protocols for Linux.

      It is actually surprisingly simple. You activate OSPF on your Linux box and it will add routes on working links and remove them again if the path is no longer viable. Works both ways too, so your routers will not try to deliver traffic to your box on a dead link. Instead the routers will route around the problem.

      • Re: (Score:2)

        by Macfox ( 50100 )
        From reading the details on the site, it's more sophisticated. It interacts with iptables, which traditional routing protocols do not.
        • Exactly! I wanted to create a well documented and fully functional system targeted to the average Linux guy, not to a network specialist. It cares about everything: creating iptables configuration, setting ip policy routing rules, monitoring the uplinks, warning the administrator.
      • Does OSPF work also with cheap DSL routers Telecom Italia or Vodafone provide us? I doubt it. Unfortunately we're not in Silicon Valley here... ;-)
    • If I'm not wrong, ja kernel patches detect a dead route by pinging the interface gateway. In my experience, in my environment (Italy), this is not enough to detect a dead gateway: we often get dead DSL links or very disturbed radio links. In both cases the interface gateway is up and running, so the uplink would not be detected as down. And this unfortunately doesn't happen here "less than once or twice a year": we don't have a very nice internet infrastructure here in Italy, in some rural places it can hap

  • Ruby?? (Score:4, Insightful)

    by Schlopper ( 413780 ) on Tuesday March 03, 2015 @04:28PM (#49174831)

    I do not want to install Ruby on my firewall/gateway along with all of its douchebaggy dependencies and gems/crystals/unicorns/whatever-the-fuck-they're-called. This is networking, not some hipster web 2.0 app.

    • Re:Ruby?? (Score:5, Informative)

      by John Bokma ( 834313 ) on Tuesday March 03, 2015 @05:05PM (#49175189) Homepage
      Ruby is a scripting language like Perl or Python. Rails is the "hipster web 2.0 app", which is a library (framework) for Ruby.

    • Then, mr hardtoplease, wait for the systemd based implementation, it is right around the corner (holding a knife).

    • It's not aimed at embedded systems, so dependencies are not a problem for me. Then, installing dependencies with bundler it's a joke. It's clear you're not a Ruby fan! ;-) I agree it would have been better to write it in C, but I'm a little rusty on C, today I almost develop everything on Ruby, so it was a natural choice for me.

  • I am unclear if this would work for a "single stream" like a video webcast source in a remote location using multiple cellular links like jetpacks. I can easily see how any given network request would go to one link or another depending on availability, but I'm assuming that this would not be able to take a network video stream and parse out portions of it to the various uplinks and then at the far end put the stream back together (in order to get either extra bandwidth or extra reliability) to forward on

    • I would use OpenVPN. One VPN going from each uplink to your data center. Then use channel bonding on the OpenVPN interfaces on both sides: remote location and data center. Of course you would need a modified version of Fault Tolerant Router to detect if an uplink goes down, otherwise your "channel bonded" connection would go down too...

  • I would be happy enough if I figured out how to mark the packets so that my routing actually works. I already have a shell script to switch connection from fiber to 3g, but incoming connections only work on one at a time. However, I'm not willing to run Ruby on my router to solve that issue...

    • No, you needn't Ruby on your router if a working packet marking solution is all you need. Just run Fault Tolerant Router in demo mode on... Windows, if you wish! Grab the command output and use the same commands on your router, using Bash or whatever you like.

  • That's one of the points of zebra and the suite of tools that it brings to the table.

    The kernel shouldn't do more than it already does unless you want to move the kernel into systemd as well.

    http://en.wikipedia.org/wiki/G... [wikipedia.org]

    Zebra is not, by any means, new.

    • Oh hell, this story is nothing more than a slashvertisement for some guys new pet project, the first commit is barely 2 weeks old.

      Could you people use google and a cloupon before you reinvent the wheel using a shitty scripting language, please.

      • It would be nice if we could get rid of stories like this and all those stupid kickstarter "gimme money to build this cool new thing" ads.
      • First commit is 2 weeks old because until then the program was internally developed and used, as stated in the README. 2 weeks ago it was put on GitHub and released as GPL.

    • unless you want to move the kernel into systemd as well.

      Shhhhh...you're giving them ideas.

  • But hey, it hasn't been on Github until now. Why the hell is this news?

  • OpenWRT with mwan3 (Score:5, Informative)

    by AlreadyStarted ( 523251 ) on Tuesday March 03, 2015 @04:36PM (#49174913)

    OpenWRT package mwan3 has similar functionality without the complication of multipath.

    http://wiki.openwrt.org/doc/ho... [openwrt.org]

  • Not to lose my karma (Score:3, Insightful)

    by Yew2 ( 1560829 ) on Tuesday March 03, 2015 @04:38PM (#49174933) Homepage
    but as a network engineer, this snippet is painful to read. Inaccurate to say the very least.

  • Many other tools for multipath (Score:5, Informative)

    by klapaucjusz ( 1167407 ) on Tuesday March 03, 2015 @04:39PM (#49174937) Homepage

    There's a lot of multipath-related work being done right now, at the IETF, within OpenWRT, and independently.

    We've been working on providing multiple routes automatically [arxiv.org] (disclaimer -- I'm a co-author). As to actually making use of the multiple routes, the solution that currently works best is MP-TCP [multipath-tcp.org], a set of kernel patches that allows TCP to use multiple routes simultaneously, with no modification to applications. Other solutions are SHIM6 [wikipedia.org], which works below the transport layer, and Multipath Mosh [arxiv.org], which works at the application layer.

    I'm pretty confident we'll be able to have most of this stuff enabled by default in mainstream Linux distributions by the end of the year.

    • Re: (Score:2)

      by Zarhan ( 415465 )

      Well, this example is just a simple load balancing with outbound NAT and nothing as fancy as the stuff you cited which aims for e.g. multipathing single sessions like MPTCP does.

      Been working on that same area.

  • gwping (Score:3, Interesting)

    by ManiaX Killerian ( 134390 ) on Tuesday March 03, 2015 @05:29PM (#49175425) Homepage

    There is a small shell script called gwping, which can be used to do the exactly same thing, easier and simpler. It's ~150 lines (with comments and everything) and takes 10-20 minutes to setup with the policy routing and everything, we don't need an overbloated runtime to do something so simple.

  • dualgate_multinet.sh supports dual-gateways and multiple subnet vlans.

    https://github.com/sodonnell/b... [github.com]

    This script can easily be extended to support more than (2) gateways, and can support various VLAN/subnet configurations and isolation.

  • We've had a functionally equivalent capability with Shorewall + LSM (Link Status Monitor) for years now. Setup (2) ISP connections, route a proportion of your traffic to each one and when the link goes down, all traffic goes to the other link.

    The hard part of the equation is your public DNS records that need to change to whichever IP address is "active" (or round-robin between the two). But most DNS service providers have a solution for that as well.

    Naturally, it's not as seamless as multi-path, but i
  • High Availability (HA) [linux-ha.org] is not an option for many Linux users (or to me for my home systems) so I use iproute2 [linuxfoundation.org] (which is built into all common Linux distributions). With a few simple rules, one can make outbound traffic go out on the interface that it is associated with. For example: I could have multiple DNS A records for a host (using either single or multiple network interfaces) and have that host respond to client requests via the same interface on which they arrive.

    Iproute2 has worked out very well

  • Pity it's written in ruby, that's a bit too heavy for most small routers...

Slashdot Top Deals

The one day you'd sell your soul for something, souls are a glut.

Close