On Wed, Dec 30, 2009 at 2:03 PM, Jared Mauch <jared@puck.nether.net> wrote:
Back at the Toronto NANOG I bumped into someone who had an interesting solution to the multihoming problem.
What they had was a machine that would key/sequence the packets and send them out each connection (so if they had 2, it would send a copy out each).
That's exactly what I was alluding to and you may have spoken to the person that wrote the tool I was thinking of, as that's pretty much what I described. (He and I both operate out of Toronto.)
Whichever got there first, was decapsulated and forwarded on. Any duplicates/late packets were dropped. This meant that they would always have the speed of the fastest link for either up or down.
With similar links (my allusion to low latency between the far ends of the upstreams across a local *IX), you really reduce jitter as well. Happy voip. I've used it, it works, just need to get it out there. Esp out here, for my voip because my latencies go up and down, so I'd rather have my packets go out twice and first one wins. (I've assisted with customers that have this service running today and have for a couple years, but I havent set it up locally here yet as I havent had a real need for reliability til I went all VOIP. I used to use plain mpppd across multi providers mainly for agg bw, but that's not nearly as good as this solution for reliability.)
They also had a method to load-share to bond the two (or more) links together.
As I mentioned, I think based on ToS or packet size. And can even pound through packetloss with duplicate packets down the same link (though I dont think that's implimented yet).
It was some custom solution they built, but something I would like to see a link to or open-sourced.
Still is and still hasnt been moved into a proper wide-deploy testing and marketing phase. I think it would be useful, but wanted to gauge your reaction. In fact, Im not sure what the next proper step in the whole endeavour is. If anyone is intersted in testing/using/assisting with marketing/selling it, contact me off list and Ill describe the particulars. Note it aint my tech, I just work closely with the developer. On Wed, Dec 30, 2009 at 02:08:18PM -0500, Dorn Hetzel said:
I guess that method presume some cooperating box out there on the net somewhere to coordinate the far end?
Also what I alluded to, you need a provider running the COE side of things (and if they go down you lose everything except your basic links, assuming the same one isnt responsible for both links). But we're looking at colo reliability for the COE - done right should be up into the mutli-9s. /kc -- Ken Chase - ken@heavycomputing.ca - +1 416 897 6284 - Toronto CANADA Heavy Computing - Clued bandwidth, colocation and managed linux VPS @151 Front St. W.