Editor's note: These minutes have not been edited. Date: Thu, 4 Jan 1996 10:23:59 -0800 From: Cengiz Alaettinoglu Subject: RPS WG Minutes RPS WG Minutes By Ramesh Govindan, and Rusty Eddy Meeting 1: Wed 12/6: -------------------- Cengiz introduced the agenda items. Curtis Villamizar presented the status of the transition of ANS away from advisories. He reported that the aut-num object for ANS has been gradually cleaned up and that ANS no longer uses advisories. Curtis also talked about the status of ANS aggregation work; some proxy aggregation is currently being configured manually. Finally, he listed some ongoing configuration tool development work. Andrew Partan asked how different from shortest-path route selection was ANS's routing policy. Tony Bates of MCI said that about 10-15% of their routing deviated from shortest path. Brian Renaud presented the RADB cleanup efforts. Since the last IETF, some effort but not a whole lot has been done to clean up. Merit is also collecting routing information and will integrate some of this into the database. Tony Bates talked about MCI's registry. Using RIPE code. Quite heavily used. About 25K routes registered in the database. A question was asked about coordination between MCI's database and other databases. Tony Bates replied that currently MCI needs to be notified of the need for such coordination. Cengiz Alaettinoglu discussed an extension to RIPE-181 for the BGP Destination Preference Attribute. He also presented a usage of this attribute. Cengiz then talked about the use of AS expressions; logical combinations of AS macro and AS numbers. We need an additional ThisAS keyword to disambiguate a macro reference. Several ambiguities in the unrestricted use of AS expressions: mostly these have to do with the lack of information about the universal set of peers. Solution is to restrict the kinds of AS expressions that can be specified. Curtis talked about ways of handling aggregation in the IRR. Need to identify: aggregates formed on exit, entry, components, components passed unaggregated, aggregation boundary. This is done using communities with specified names. Also basic need to associate "external" information with an object. Several mechanisms proposed for this. Also described the levels of possible aggregation and the kinds of associations necessary. Some discussion of the amount of coordination needed to achieve provided level coordination The second part of Curtis's talk dealt with changes to the language to specify aggregation. Several options: change route object to include inbound vs outbound aggregation (problem: authorization model i.e. how to check if someone is allowed to aggregate a route), modify the aut-num object (too much junk in the object already), have a separate aggregate object. Prefers the association/attachment mechanism. Apart from that, need a way to specify what a collection means (i.e. to specify the aggregate). One way would be to use a "name" to specify the function. Finally, Curtis talked about a syntax for representing IP address prefixes in the language. This included exact and more specific prefixes or prefix length ranges. The chair agreed with the need for such syntax and the proposed syntax as well. Cengiz summarized the mailing list discussion about AS path regular expression syntax. He described how the initially proposed syntax compared with that from cisco and gated. He also showed examples of different syntaxes. The general consensus seemed to keep the integer-based regular expression syntax. Cengiz's talk titled "communities and macros". He gave a background description of RIPE communities and how they differed from macros. As currently defined, communities have an authorization problem. Macro objects however contain the objects they reference and do not have a similar authorization problem. Proposed a mechanism for changing the authorization model of communities using a "referent-maintainer-list". Some discussion about the need for this in the presence of route macros. Also described a pp-macro attribute which is syntactic sugar for simplifying policy specification. Jessica Yu talked about a proposal for an ISP Tag in the IRR. Basically we need some way to associate tags/flags with route for several reasons (e.g. router configuration using outdated data). Some discussion of the efficiency of such a proposal (to tag a route, need to modify as many route objects as there are routes; can use route macros) as well as whether this breaks the authorization model. Meeting 2: Thu 12/7: -------------------- - Multicast policies: Deborah Estrin multicast is used to achieve resource sharing. policies that descriminate against senders and recievers can defeat the purpose of resource sharing. policies that are based on mcast addrs only make sense when mcast addrs have a fixed meaning. mcast addrs are not fixed and mainly transient. can have admistrative scope. neighboring restrictions. downstream credential restrictions, sender restrictions. - SDRP route construction agent: Cengiz Description of route construction using the IRR. Gave an example topology and walked through an example. Discussed a modification to the aut-num obj within the IRR, need to register SDR supported routers. provided an example of proposed modifications to sdrp policies. A discussion insued about how to set up policies to counter the sdrp example. - Real time database mirroring: Cengiz for David Kessens Overview of the need to efficently mirror information in seperate databases. described how it works, current example and what comes next. (giving a serial number?) Curtis: Suggestion, the next step with mcasting, take a look at wb and reliable mcast. find what is out of sequence and request the missing pieces. Marten: use unicast first: Elise: why use the time stamps vs. serial numbers (???) Inter-Registry Data Exchange: Jerry Scharf CIX collaborative effort with the InterNIC. they have funding and programmers, waiting for the work in progress to complete. the design will be complete before any development is done. they hope to have working functionality before next ietf. do not want to desing a distributed database. they are working on the data distribution model, including a powerful authentication model. will look at work from the ipsec. showed a first cut of an object-description breakdown. Jerry provided his email addr to solicit comments CIDR assistant - Cengiz CIDR assistant is useful because incorrect use of CIDR can cause routing problems, holes. Discussed types of aggregation level 0-3. gave an example and a walk through, including examples of all levels 0-3. Cengiz also spoke about proxy aggregation to be supported by proxy aggregation. Cengiz also spoke of future issues: allow holes, output for cisco and gated. identify policy changes to provide better aggregation. IRR Visualization - Rusty The tool plots the intetrnet toology, is an interactive look at irr data. It can save and open different views of IRR. It uses tkined and scttoy. Tkined is a gui based internet editor, it is a drawing tool: draw nets, connect them by links, hosts. Scotty is extended tcl with netwirking support such as udp, snmp, tcp things. It can construct views, explicityly thru as expansion, or implicity thru tool usage, such as prpath, prtraceroute, etc. For example it plots the domains on the paths output by prpath. It can group domains into icons. The display gets messy when the views are opened since the views are not opague. Current icon placement options: draw peers clockwise, geographic icon placement, longitute and lattitide info on a map, not implented. No geographic information is available in IRR. Tkined can do whois queries to find geographic information using DNS etc. Planar algoritmnic icon placement can find centers of the graph, and draw around it. Modified fish eye view: one can move your mouse around, and the portions of the graph around the mouse becomes less dense. Another way is to push out things around the mouse, open and collapse groups as you move around. Tony Bates said this tool may get people to use irr. He emphasized the geographical placement is important, and suggested minimalistic hooks in the IRR for geographical information. Michael Patten suggested to use the LOC atttribute of DNS. He said NIMROD visualization implemented layers, but no information useful for layers was available in the IRR. Curtis suggested to look at integrated cicuit placement algorithms, this would be under planer placement algorithms. IRRTOOL - Rusty Click on an AS on the visualisation and launch the IRR tool or use it stand alone. It does irr queryies, creat and edit objects, gui front end to the tools, pops up editor for edit/creats. - Peval and RtConfig: Cengiz Peval inputs a policy expression, make expansions and evaluates the expressions. It is used as a library or as a backend tool. RtConfig and CIDR Assistant uses it. RtConfig is a tool to configure routers in cisco, gated and rsd formats from the IRR data. Cisco configuration is under construction. - pmatch and prpath: Cengiz Pmatch is a backend tool and library which lists which policy terms in an autnum matches a given destination. prpath, prtraceroute and prconn can use it. It uses the services of the RADBserver. prpath is originally written at ripe. what's new: c++/c/bison/flex, understands many new extenstions including as path regular expressions. whats' next: fuzzy matches, listing each action taken as the route propagates. Tony is concerned with documentation, including drafts. Cengiz agrees, and posts a Request for Volunteers. Tony Bates, Marten Terpstra, Curtis Villamizar and Elise Gerich volunteered to help Cengiz write a draft rpsl specification.