>>>>> "Sanjai" == Sanjai Narain <[EMAIL PROTECTED]> writes:
Sanjai> The key simplifying observation is that there is a set of Sanjai> fundamental protocols/distributed algorithms (P/DA) whose Sanjai> compositions make up an end-to-end system. DNS, DHCP, Sanjai> SENDMAIL are all examples of these. We capture the answer to Sanjai> the question "what does it mean for a group of agents Sanjai> executing a particular P/DA to be correctly configured?" Sanjai> This becomes a model of the configuration logic for that Sanjai> P/DA. We now compose these models to create a validation for Sanjai> the whole system. This was the idea behind the Service Sanjai> Grammar work in which the compositions are programmed in Sanjai> Prolog, not full first-order logic. See the papers Using Sanjai> Service Grammar to Diagnose Configuration Errors in Sanjai> BGP-4. <http://www.argreenhouse.com/papers/narain/scp-BGP.pdf> Sanjai> /Proceedings of USENIX LISA Conference, /San Diego, CA, Sanjai> 2003, and Building Autonomic Systems via Sanjai> Configuration. <http://www.argreenhouse.com/papers/narain/Autonomic.pdf> Sanjai> /Proceedings of AMS Autonomic Computing Workshop, /Seattle, Sanjai> WA, 2003. Yeah, there are certainly a group of core protocols that are really important, but in order to rely _solely_ on analytical tests, you could need a model for each protocol/program. I am mainly pointing out that the space for unix systems is a lot larger than for network management; with network management, there are a much smaller set of protocols that you need to model network behavior. Compared with the wide range of software that may be vital at a given site on unix systems, this is a much more tractable problem. >>> Behavioral tests aren't as powerful as the analytical ones that >>> Sanjai does, but they sure are a lot more incremental. Sanjai> An example of where behavioral tests don't seem appropriate Sanjai> is in testing single points of failure. We cannot fail Sanjai> individual components and links to see whether end-to-end Sanjai> service hold up, can we? An analytic approach is Sanjai> needed. However, this is also not an either/or Sanjai> situation. Behavioral and analytic tests are complementary. Yeah, I agree completely. There are several cases where behavioral tests can't quite give you the big picture. They should be used with analytical methods where possible. My main point was that behavioral tests fit with the methods that administrators are already familiar with; modelling isn't so common yet. -nld _______________________________________________ lssconf-discuss mailing list lssconf-discuss@inf.ed.ac.uk http://lists.inf.ed.ac.uk/mailman/listinfo/lssconf-discuss
