IAB-sponsored working on 18-19 October 2006 on Routing and Addressing topics. As a background comment, there appears to be some doubt from some parts of the operations area that the existing IETF activities can provide effective solutions for the scaling issues relating to routing. Objective of the workshop was to develop a shared understaanding of the problems that operators are facing with the current routing and addressing framework.
The scalability problem is critical and near term, and in particular the capability to further scale the FIB in an economically feasible way is a relevant observation. RIB dynamics are also an issue here, where update rates are inflating faster than RIB size. And the requirements on routing also increase in terms of scale factors.
The overloading of IP address with both locator and identifier semantics is a problem. There was some uncertainty about what these two terms meant. Is this an inherant problem or a solution vector? This may have implications whe considering mobilty protential growth.
There was some discussion about the timelines of scaling criticality. It does not appear to be an "event" time, but rather a potential for cost escalation for network operators.
The workshop report is being prepared, and some consensus as to the nature of the issue here is being gathered. The IAB may work on the architectural aspects of the locator/id split.
The problem is that scaling remains an issue with Ipv6 - it shares the same routing technology base as IPv4 without any noteable difference in the IPv6 routing and addressing approach. The ROAD effort in the early 1990s saw issues with exhaustion of Class B nets, the explosive growth of the routing table and the eventual exhaustion of the 32 bit IPv4 space. The resolution of these issues saw CIDR and strong address aggregation in routing as a short term mediation, and IPv6. The CIDR short term mediation has been effective for over a decade and no further substantive work on routing was done since. The overloaded semantics of addressing does not combine ease of deployment and use with strong compression of the routing space. The scaling problem may not become obvious for a further time interval here.
However the trends of update rates and processing, and peak processing rates for reasonable convergence performance imply concern over the 3 - 5 year window for routing on deployed and deployable routers. Also projections of parallel IPv4 / Ipv6 routing deployments indicate potential to see routing tables of the order of 0.4M to 0.6M entries if this were the case today, and projections over a 5 year timeframe heading into 1M - 2M routing entries.
The routing hardware does not necessarily tightly track the peak price performance of the chip fabrication industry, and the use of specialized ASICs in routers does not give high volume low unit price yeilds in fabrication. The observation is that routing growth greater than aorund 1.3x/2 years may infer higher unit costs in packet routing.
The growth drivers include the lack of any other mechanism for TE-styles and address agility outside of inflating routing pressure. Also there is considerable pressure for PI address assignments from industry, again in a desire to maximise the utility and flexibility of addresses at the expense of the routing domain.
There is the feeling that the RAWS workshop did achieve reasonable consensus on recognising this as an important issue and is generating some productive thoughts in this area.