R I P E 3 8 A M S T E R D A M Routing Working Group Session 24-Jan-2001 Draft Minutes Chair: Joachim Schmitz (JS395-RIPE) Scribe: Engin Gunduz Participants: 97 A. Preliminaries (Joachim Schmitz) The chair man welcomed the participants, and distributed the participants' list to sign in. Engin Gunduz volunteered to take the minutes. The WG agenda was accepted without changes. The minutes of the previous meeting at RIPE37 were accepted by the participants, and are thus declared final. B. The Hermes Project (Giuseppe di Battista, University of Rome) [ The presentation will become available shortly at the RIPE web server ] URL: http://www.dia.uniroma3.it/~hermes The idea of the Hermes project is the integration and visualisation of Routing Registries Information and BGP Routing Data. Consequently, major targets are to o explore the Internet interactions at AS level o visualise interactive maps o show inconsistencies Hermes is based upon a o 3-tier architecture o repository of routing information o mediator for the information integrity. o graph drawing module After presenting the general outline and functionality, Guiseppe showed a demo prerecorded with a screencam from a life example around AS137. This demo included - peering activities (neighbor ASs, and routes) supplemented with more info from the IRR database - in/out routing policies - paths to another AS with routes from that AS and description from the IRR supplemented by AS macros involvede, plus which ASs are part of them Another demo showed more details around routing policies, where the AS was initially selected by its name in the IRR - showing inconsistencies of registered policies of connected ASs The results can be viewed as text and in a graphical map. Guiseppe continued with more details on the internal structure of how Hermes works. In particular he described - three tier architecture - data integration - mediator for data processing - graph drawing around the difficulties introduced by locally sparse and extremely dense parts requiring complex algorithms Eventually, current and future work was summarized: - moving from RIPE-181 to RPSL format (March 2001) - adding new sources of information (RIS project?) - distinguishing between different sources of information (June 2001) - visualisation and animation of route flaps (June 2001) - looking inside an AS: traceoute (June 2001) - looking inside an AS: OSPF (June 2001) Finally, a brief distinction compared to other similar projects was given: o skitter (Caida) [global view of Internet] o ASExplorer (IPMA) [specific view of one single AS] o Hermes [in between] In the brief following discussion, the impression was that Hermes is a useful tool, comprising information from different sources. Q (chair): It is not entirely obvious which data comes from which source. Is this shown in the tool? A: It should be relatively easy, and is actually planned. Q (Ping Lu): What are the means of output? A: So far, the data can only be printed. Since this is a Java applet, it is difficult to produce output in many formats and save them to local disk As final summary, Guiseppe stressed that he welcomes suggestions like this, and asked for any additional input to be sent to him and his group (contact details on the web page http://www.dia.uniroma3.it/~hermes). C. RIS Report (Henk Uijterwaal, Thomas Franchetti, Antony Antony, RIPE NCC) [ Presentation is online at: http://www.ripe.net/ripencc/pub-services/np/Talks/0101_RIPE38_HU/ http://www.ripe.net/ripencc/pub-services/np/Talks/0101_RIPE38_AA/ http://www.ripe.net/ripencc/pub-services/np/Talks/0101_RIPE38_TF/ ] The RIS group gave three presentations 1. Henk Uijterwaal In the introduction, Henk listed a few administrative items - RIS has progressed that for that it will be turned into a regular operational service very soon - To support this move, they are now hiring a 2nd network engineer which will bring the headcount to 2.5 FTE's plus 1 student (from 1-Apr on) - Henk again announced the PAM-2001 workshop Then, a sumamry of the highlights in the past months was given - a new database was introduced to keep up with performance requirements - more powerful hardware was installed, and a scaling analysis will be done with it - currently there are 4 Remote Route Collectors (RRC) at RIPE NCC, LINX, AMS-IX, SFINX; anybody interested in peering to put data into RIS is welcome to contact Antony Antony - two more RRCs are planned, investigating where to place them - work has been done on the user interface, with output on: web, email - vanilla, strawberry, chocolate: which view of the routes do we have in the RIS? The flavors distinguish different amounts of routing info, ranging from filtered and restricted output (vanilla) to the full set of routes (chocolate); RIS prefers to see more - following several requests from different sources (ISMA project, Hermes, ASExplorer,...), the raw data is now available (announced yesterday), check out URL http://abcoude.ripe.net/ris/rawdata During the next months the RIS team plans to - move to production during 2001 - proceed with route collection and expand it - improve and extend the user interface - proceed with network research Henk stressed again that he very much welcomes welcomes input from the community on this project. Joachim reminded Henk of the action, the Routing WG has put on the RIS team, to also implement route flap statistics. Henk confirmed that this is on the list of topics to analyse. 2. Thomas Franchetti In the second part, Thomas presented some results from data collected by RIS - statistics of numbers of prefixes by length show a noticable increase of /24, and /32 - origin ASs announcing only one prefix - distribution of prefix lengths again shows the increase of /24, and /32 - statistics of how many ASs announce certain number of prefixes most ASs announce only very few prefixes, number of announcements per AS is on average decreasing The graphs shown will be made available on the web before RIPE39. Conclusion: - it is necessary to view larger periods to see trends more clearly. - there is a lot of data in RIS, and lots of potential ways to analyze it, but not easy to find ways of presentation; this definitely needs to be explored Joachim finds the parallels interesting between the RIS project results, and other analysis like the one by Philip Smith (distributed on the Routing WG mailing list), or the "standard" CIDR report by Tony Bates. Joachim is eager to see more results at RIPE39, and also welcomes any input from the community; after all these stats are just a glimpse at the wealth of data collected at RIS. 3. Antony Antony In the third part of the presentations on RIS, Antony presented some peculiar RIS observations: - regarding negligence in aggregation, it is amazing how big a number of more specifics is announced from small prefixes (up to several 1000s in a /8) - looking at AS path length, a path with 123 ASs was found, which is very peculiar; this causes several problems, e.g. Cisco actually resets because of missing buffer space; there is also a known Cisco bug where an empty AS path may appear, and thus care must be taken if filtering on AS path length - if AS numbers are considered as such, roughly 82% of RIPE NCC assigned AS numbers are visible by RIS, and the question arises: where are the others? Actually, the RIPE region is still in pretty good shape because the global average is 50%. Following up on this, Daniel Karrenberg added some observations and comments. He noted that the full feed they get from a neighbor consists of 81,272 prefixes, but the maximum number of prefixes from a peer is 100,461. In the beginning, the difference between the extremes was just a couple of thousands, but obviously has expanded tremendously. As Randy Bush has pointed out during the EOF session, it is possible to go down to around 80k prefixes without loosing connectivity. Possibly, people announce redundant data. RIS may be used to catch the source of those redundant data. Daniel was musing whether the NCC should suggest filter lists but would want more discussion on that. D. Routing Registry Consistency Checking (Shane Kerr, RIPE NCC) [ The presentation is available at http://www.ripe.net/ripe/meetings/archive/ripe-38/presentations/rrcc-ripe38/... dex.html ] Shane reports that the RIPE IRR currently holds around 24k route objects and 4k AS objects. The quality of the contents varies widely, because there are people who really update their objects, and others who don't. To improve data quality, the data in the IRR is compared to what is found on the Internet. However, the goal is not just reporting, but help to improve the IRR and make it more useful. This is the idea behind the Routing Registry Consistency Checking (R2C2). The following slides gave an overview of the approach, the status, progress, and the plans of the project. Besides route objects and AS objects (routing policies), there are also supporting objects like AS macros which form a set of data to be compared to RIS data. Several reports and tools are in the make, like - summary report - specific report - router configuration checker - IRR correction wizard The project has reached a status where comparisons are possible, and mismatches may be identified. Prototype interfaces and a prototype correction wizard are available. A web page for the project exists with a practical example of the prototype interface. The next steps include polishing and finishing current software, publishing a summary of the project as RIPE-201 document, and seeking community input. Potential possible directions include - a view from specific route collector on request - checking aut-num policy with AS paths - a router configuration verification Additional suggestions include working together with other IRRs and on request from the community develop further applications. This was the last topic in the Routing WG session itself. The second slot was dedicated to a joint session of the Routing and Database WGs dealing with RPSL transition issues. Y. Input from other WGs There was no input from other WGs. Z. AOB There was no other business. Summary of Open Action Points from the Routing WG 36.R1 on RIPE NCC During the transition phase to the RPSL software - verify with RADB on their test suites for RPSL implementations - coordinate with RADB on consistent mirroring of databases (NRTM) - coordinate with RADB on consistent whois interface of databases including irrd status: ongoing 37.R1 on C.Panigl, P.Smith, D.Karrenberg, J.Schmitz Provide updates to RIPE-210 - general editing, review - explanation of damping parameters - examples by implementation - script to determine TLD/root nameservers for "golden network" list status: ongoing 37.R2 on RIS team, Henk Uijterwaal Implement route flap analysis from RIS data status: ongoing Engin Gunduz, Joachim Schmitz, February 2001