|
|
IP-Multicasting in ISP Usage: Multicasting NetnewsNetnews (or Usenet news) is one of the more prominent systems for electronic communication that make up what is now loosely called "the Internet" in the media. Usenet operates by flood-distributing articles between participating news servers. The Usenet is experiencing growth problems like any element of the thriving Internet environment. It is widely recognized that NNTP, the article distribution system in use in the Usenet, is running into scaling problems. In today's NNTP system, there is little relationship between network layer topology and the NNTP interconnection topology, as nicely illustrated by the EU-to-US connectivity: Also, the path from the author to a reader of an article involves a number of NNTP hops, each of which can experience significant queueing. The result is that a mid-level news site in Europe such as the one at the University of Bremen is seeing an average article delay of 7.5 hours. This is in marked contrast to the sub-second hop-by-hop article relay times sometimes boasted by operators of core news servers. Fortunately, there is an existing solution for efficient information dissemination in the Internet architecture: IP multicasting, which allows sending a packet to a group address so that it is received by all hosts that have joined that group. It is then the job of the network to efficiently distribute the multicast packets to all receivers based on its knowledge of the network topology. Newscaster is based on two protocols: The News Distribution Protocol (NNDP) is used to distribute compressed batches of articles, by partitioning the batch into packets that can be multicast without being fragmented on the way and sending out these packets in a properly paced, rate-limited fashion. Using forward error correction (FEC), NNDP achieves good performance in the presence of losses. Newscaster servers continue to use NNTP for background transmission of articles that were missed during connectivity outages. The News Distribution Coordination Protocol (NNDCP) solves the multiple entry problem: NNTP is often able to distribute articles to multiple servers at about the same time. As, in a multicast scenario, there is no way to ask all the receivers whether they already have received an article, this means that servers regularly multicast redundant copies of the article in parallel. Newscasters announce their intent to send an article using NNDCP, which is based on the consistently ordering reliable multicast transport MTP/SO. The first newscaster to announce "wins" the right to multicast the article. After a number of design iterations, the Newscaster system now is a workable system for introducing large-scale Netnews multicasting into the global Internet. The Newscaster implementation runs with the most popular news servers and is in operation at sites in Europe and the USA. Source and documentation are available from the Newscaster web site; the global experiment is coordinated globally through the newscaster mailing list newscaster@mathworks.com. |
|
|
|
|
|
|
