Minutes

Routing SIG

Thursday 2 September 2004, Sheraton Fiji Resort, Nadi, Fiji

Meeting commenced: 4:00 pm

Chair: Philip Smith

The Chair introduced the SIG and explained the agenda.

Open action items
BGP wedgies - Bad routing policy interactions that cannot be debugged
Happy packets: some initial results
IPv6 routing status report
Status report on BGP and BGP: The Movie

Open action items

Philip Smith, Cisco

The Chair reported that there were no open action items.

Top

BGP wedgies - Bad routing policy interactions that cannot be debugged

Timothy Griffin, Intel Research

Presentation [pdf | ppt]

This presentation examined one class of anomalies that has developed in the BGP system. The presenter noted that BGP policies make sense locally, but globally they may result in multiple stable routings and some routings are not consistent with the intended policies. A 'full wedgie' occurs when an unintended routing is installed, and no single group of network operators has the knowledge necessary to debug it. The speaker used a simple example to show how this might occur. The speaker noted that as routing got more and more complex this scenario would become more common, and that the solution would require manual intervention to fix, which is not a good thing in a dynamic routing situation.

The speaker recommended that network administrators be aware of this problem. He suggested that interdomain communities should be defined with care and consistently implemented. He also noted that tools which could enumerate all stable routings would be useful, and that he has begun to look at developing these types of tools.

Questions and discussion

It was asked whether route caching was responsible; the speaker clarified that it was not, and that it was simply the nature of BGP.
It was asked whether the Internet Vendor Task Force was developing a draft to address this. Geoff Huston reported that the IVTF was not currently looking at this.
It was asked whether these situations had been used as a competitive or anti-competitive measure. The speaker reported that he had not had experience of this, but knew of similar scenarios where these sorts of anomalies had been used as a competitive measure.
There was some discussion of the technical and political aspects of BGP's development that had led to this situation.

Action items

None.

Top

Happy packets: some initial results

Randy Bush, PSG

Presentation [pdf]

This presentation addressed the relationship between control plane instability and data plane instability. The speaker noted that there was a lot of pessimism about Internet routing, often blamed on BGP and problems associated with it. The speaker compared BGP to white bloods cells, which are not part of the problem, but part of the cure.

The speaker contended that the measure of routing quality is whether the users' packets reach their destination. The presenter referred to these as 'happy' packets and explained that metrics exist for measuring them. The speaker discussed one method for measuring this using a BGP beacon.

The speaker reported investigating the relationship between packet loss and the number of BGP announcements, and reported that no correlation appeared to exist. The speaker reported on a number of conclusions that the research seemed to indicate.

The speaker also revisited a presentation which he presented a year ago, entitled 'BGP is chattier than we think', looking at ways to improve routing speed.

Questions and discussion

A question was asked about whether these issues applied to IPv6; the speaker did not know.
It was commented that routing, like evolution, did not have to be the perfect, but only 'good enough'.

Action items

None.

Top

IPv6 routing status report

Philip Smith, on behalf of Gert Doering, SpaceNet AG

Presentation [pdf]

The Chair noted that the author of this presentation has reported on the IPv6 routing table for a number of years. The presentation examined a range of statistics regarding the IPv6 routing table. These indicate that the number of ASes in the routing table is growing steadily. The report shows a number of people announcing two prefixes, and provides some reasons for this. It also examined the 6-to-4 prefix, and noted the number of prefixes in the routing table, and the growth trends in this through RIRs and the 6bone.

The report also included a number of interesting observations within the IPv6 routing table, called 'ghosts', which are BGP withdrawal bugs. There have also been a number of accidental hijacks, prefix leaks, weird AS path leaks, and invalid AS numbers.

The presentation reported that the 6bone will be phased out by June 2006, and noted that overall the IPv6 infrastructure is improving, moving towards production quality. He also noted that while the US is catching up to Asia and Europe on IPv6 allocations, it is still lagging in announcements.

The report concluded with a number of IPv6 routing recommendations. These included no peerings over bad tunnels, applying incoming prefix filters to peers, filtering private ASes and overly long paths, and not giving unrestricted IPv6 transit to peers unless asked to.

Questions and discussion

It was noted that some organisations define all tunnels as bad tunnels in terms of peering over them, and recommend against using them.
It was noted that for those involved in deploying 6-to-4 reverse, the current low level of activity in 6-to-4 was a good opportunity.
It was noted that it would also be worth doing further research into Teredo, which is currently not in heavy use, but is widely available.
It was noted that in a real-world situation, tunnelled IP is significantly worse than VPN tunnelling, and that this needs to be taken into account.
It was noted that there will be a number of IPv6 allocations larger than /20, and that it would be worth measuring the level of disaggregation necessary to avoid filtering.
It was noted that not all /35s are historical, meaning that there are three bits of disaggregation available to those with /32s, and it was argued that this can be very useful and should be allowed to continue as long as the /35s are put in a routing registry.
It was noted that some recent research by Kenjiro Cho on tunnels had found them to be remarkably cleanly engineered.

Action items

None.

Top

Status report on BGP and BGP: The Movie

Geoff Huston, APNIC

Presentation [pdf | ppt]

The speaker presented a historical view of the BGP routing table, and recent growth and developments in address announcement, fragmentation, unique ASNs, and average AS path length. The speaker noted that it appears that the amount of absolute fragmentation is growing. The presentation included an animated film of the progression of address space allocation, routing, and visibility.

Questions and discussion

It was asked whether there was a comparable movie for IPv6, and then speaker noted that there was, but that nothing much happened in it.
It was noted that AS numbers represented the fastest growth rate. The speaker also noted that if the A-class address space areas which are currently being neglected were actively managed, it could buy another 10 years of use for IPv4.

Action items