Hacker News Reader

Comment by the8472 | Hacker News Reader

Source-specific IP Multicast support on the open internet, which probably needs some serious magic to make it feasible on core routers, if it is solvable at all.

Some shabby server in someone's basement could stream data to millions. CDNs would no longer be needed.

View on HN

There is a great fear of Multicast from many network engineers I talk to. This comes from at its core, it being too easy to create cycles. At the internet scale the potential for cycles would be even higher, b/c the TTL needed would have to be on the order of ~20-30 hops, and many engineers will just set this number too high.

Now I don't have experience with Source-specific Multicast, which is supposed to resolve the cycle problem, but I think there remains a lot of fear of Multicast in the industry.

Can anyone else comment on cycle issues?

Also, in terms of scaling, Multicast requires of copy of the packet on the router to every port which is registered as a listener to a Multicast address, this will increase load on the larger routers.

About scalability, a possible compromise might be making some distance-based limitations. I.e. limit subscriptions per prefix or adjust packet drop rates.

That way regional SSM can still be useful and local islands (which whould hopefully grow over time) could be bridged by unicast links or - if they have overlap - by leapfrogging.

Even at the ISP level there would be still be use-cases for SSM.

That sounds a lot like taking the D out of DDoS.

Isn't this how multicast IP television works anyways? With IGMP and PIM?

Multicast on top of MPLS isn't really a thing with respect to the backbone routers.

Reply on news.ycombinator.com