Minutes from the Operational Statistics Working Group meetings at the San Diego IETF, March 16-20, 1992. Participants: Chris Myers chris@wugate.wustl.edu Nevil Brownbe nevil@aukuni.ac.az Brian Shiflett bshiflet@icm1.icp.net Jim Alfieri jdal@troy.cc.bellcore.com Ken Goodwin goodwin@psc.edu Stefan Fassbender stf@easi.net David Waitman djw@bbn.com Robert J Reschly Jr reschly@brl.com Ron Roberts roberts@jessica.stanford.edu Henry Clark henryc@oar.net Frances Yeh fyeh@rafael.jpl.nasa.gov Al Wiersma wiersma@nsipo.nasa.gov Ursula Sinkewicz sinkewic@decvax.dec.com Miriam Nihart miriam@ltning.zso.dec.com Dale S Johnsson dsj@merit.edu Vikas Aggarwal vikas@jvnc.net Pushpendre Mohte pushp@cerf.net Bill Bliss billbl@microsoft.com Tom Easterday tom@cic.net Frank Solensky solensky@clearpoint.com Linda Liebengood ldl@ans.net Kary Robertson kr@concord.com Hock-Koon Lim lim@po.cwru.edu Bill Mannings bmanning@rice.edu First session, Tuesday March 17. 1. Review of the OPSTAT document. The document on "A Internet Model for Operational Statistics" was reviewed. Decisions: - In listings of MIB variables suggested for gathering the name shall be the last part of the fully qualified MIB name as long as this uniquely defines the variable. - A suggestion of adding InErrors and OutErrors was not approved as these variables are counted differently in different MIB implementations. - Valid types for bandwidths and protocol types should be explicitly enumerated. - The timezone part of the timestamp in the datasection is redundant and is moved to the devicesection. - It shall be more clearly stated that document gives recommendations and polling and saving periods are not mandatory. After these changes have been included the meeting decided that the document shall be submitted as Internet Draft. Changes suggested that would make major changes necessary can be discussed and decided on during the six month Internet Draft reviewing period. 2. SQL Database The meeting discussed possible use of SQL database technique As SQL is optimized for other purposes than retrieval of serialized data the meeting concluded that SQL was not appropriate to use in retrieval of statistical data. 3. Implementations of the OPSTAT model The meeting made a review of NOC's prepared to implement the operational statistical model when the Internet Draft is submitted. Below NOC's who were represented at this meeting expressed interest: Merit (Dale Johnsson) New Zeeland (Nevil Brownbe) RIPE NCC (Daniel Karrenberg) EUnet (Daniel Karrenberg on behalf of Joy Marino) 4. Need of theoretical framework The meeting discussed the need for a theoretical statistical framework as current thinking to some extent is based in practical experiences. Kim Cluffy, SDSC, is writing a PhD thesis on analysis of wide area network. The output may have a lot in common with the OPSTAT work. Bellcore has developed models for statistical analysis of data from packet switch networks which also could show beneficial for the OPSTAT work. Contact shall be taken with Bellcore to investigate if their models also could fit in the Internet architecture. Session 2, Wednesday March 18. 5. Review of future activities. At an early stage the OPSTAT group discussed possibilities of using a client/server based system for retrieval of statistical data. The meeting agreed that such model would be useful to offload network equipment from SNMP processing and to enforce access control of statistical data. Some similar system may already exist. As examples were mentioned a system developed by DEC. 6. Review of existing tools. NASA have tools that currently is configured for 1 minute polling. The 1 minute polls are stored internally in the program and 15 minutes average and peaks are being stored onto secondary storage. The NASA statistical tools may be made publically available. Milo Medin, NASA, expressed the need of differentiating between statistical tools and monitoring tools. SNMP access is not the same as access to statistical data. A client/server based system may show useful in accessing logged data when there is no "public" snmp access. Another method would be to give access to the tables and diagrams produced from statistical data. OARnet have tools that currently does weekly loggings of the data. The tools are more oriented towards logging of error conditions. RIPE NCC has tools developed from the ISODE snmp code using gawk with snmp capabilities. These tools are already adopted to the OPSTAT thinking and changes to reflect the lastest storage formats may easily be included. Daniel Karrenberg gave a short presentation of these tools. RICE University are using enhanced Merit tools. The tools are written for AIX and SUN Sparc and could be made publically available.