The Chair introduced the SIG and explained the agenda. He explained that the presentation due to be given by Phillip Harris was cancelled as Phillip had to return home.
Preamble
Geoff Huston, Telstra
The speaker noted that in 1998, the routing table began a trend of exponential growth at a rate greater than Moore's Law. Following the end of the Internet boom in 2000/2001, this pattern changed. The speaker also began tracking 40 separate BGP feeds at that time.
The speaker explained that while approximately 1.8 billion IP addresses have been allocated to date, 0.5 billion of those allocated addresses are not currently being advertised.
In 2002, there was a significant drop in the growth rate of announcements. The speaker was not sure of the reasons for this decline.
The speaker noted that there is a lot of noise in the BGP table; within the blocks covered by large announcements, there are smaller fragmented announcements made from within this space. The fragmented routes appear to be announced to for reasons of local resiliency and local traffic engineering. The speaker commented that half of the global table seemed to be expressions of local policy.
The speaker reported that the growth of the number of unique Autonomous Systems over the past few years is slowing.
The speaker explained that his research was showing that the growth of the Internet is no longer exponential, but is growing in a gentle linear fashion.
The speaker highlighted the information available in the CIDR Report and suggested that participants access the website for more details.
Top
Allocation vs announcement
Geoff Huston, Telstra
Presentation [pdf]
The speaker explained that the motivation behind his research into the relationship between allocations and announcements was that although many ISPs perform prefix length filtering, due to accidents, there have been significant leaks.
The presenter explained that his research using RIR and BGP data showed that a quarter of the allocations made by RIRs in the past year are not being announced. Of the 3641 past year's allocations advertised, 2938 advertisements matched the allocation sizes exactly. The remaining 1206 allocations have created over 7000 announcements. The presenter explained that it appeared that the networks using these remaining 1206 allocations are cutting up their allocations into /24s using old classful methods of addressing.
The presented noted that it appears that a fifth of the networks on the Internet are continuing to use old classful methods to manage their address blocks.
The speaker commented that historical data showed that when CIDR began to be deployed, networks began to use better announcing techniques.
Questions and discussion
- It was questioned whether a network filtering aggressively for fragmented announcements would still allow packets to reach their destination. The speaker explained that packets would still arrive because most of the fragmented routes had larger prefix announcements above them.
- The speaker suggested that LIRs advise their customers not to advertise smaller announcements as the LIRs are already advertising larger aggregate announcements containing the customer blocks.
- The speaker was asked if he had contacted any of the networks announcing /24s and asked them why they were doing those announcements. The speaker replied that he had not contacted networks, but was aware of others who had done so, with varying degrees of success.
- The Co-chair stated that he had asked a number of network operators in the past why they were announcing /24s, with little success.
- The Chair asked the audience for comments on their announcement and filtering policies. One network operator present commented that his network does filter fragmented announcements.
- It was questioned if work was being conducted on the size of announcements and withdrawals in the routing table. The speaker answered that this work was being undertaken and demonstrated that longer prefixes churn more.
- It was noted that the interaction of various instances of BGP often turns a single announcement or withdrawal into a flood of announcements or withdrawals from further away. The RIS could have a lot of useful data on this, but there is so much data that it is an intimidating task to approach. The Chair noted that the Co-chair and his associates are conducting experiments in this area using a single prefix to announce and withdraw, with interesting data results.
- The speaker was asked if the presenter would think it appropriate to create some strongly worded suggestions on reducing fragmenting announcements in the routing table. It was suggested that perhaps the APNIC Secretariat could incorporate such suggestions in APNIC training in the region. The speaker explained that there were no new policy implications in his research findings.
- It was noted that in the late 1990s, when the BGP table included approximately 100,000 entries, it was thought that there would be a meltdown at 500,000 entries. The speaker was asked if there was any new figure for the maximum number of routes that can be dealt with by a router currently. A Juniper staff member in the SIG was asked to answer this. He confirmed that routers could now deal with millions rather than hundreds of thousands of routes and that, currently, BGP thrashing is not seen as an issue that is hurting router performance.
- It was observed that the presentation had showed that, prior to 1989, there was a 90% correlation between allocations and announcements; it was questioned why this was not 100%. The speaker explained that this was a result of his methodology as he did not have a snapshot from historical times. He acknowledged that the 90% figure could be misleading and stressed that fragmentation of announced allocations started to occur in the late 1980s and not earlier.
- It was asked if the visibility of announcements would be affected by putting local exchange points in countries. The speaker had no data on that issue.
- The speaker was thanked for the detailed work he was doing on the issue of announcements versus allocation sizes.
- A SIG participant commented that their network may be one of the networks fragmenting their announcements because they are a subsidiary of a larger telco that has separate peering announcements. There was discussion about how this could be changed.
Action items
