[Date Prev] | [Thread Prev] | [Thread Next] | [Date Next] -- [Date Index] | [Thread Index] | [Elist Home]
Subject: tables.zip
BTPers, The enclosed zip contains the current version of the state table html and the "proofs" of the transitions. Slightly modified from the version I put on the screen at the end of yesterday - it now really is the state table, so the entries as you see them show the new state the endpoint role goes to when the event occurs. If it is a hypertext link, it will take you to an example trace which includes that transition. If the entry is just in red, then the simulation runs haven't produced an example (at least since I last cleared the history files which was after making the last appreciable change. (Adding the ability to send RESIGN/no-rsp-req *before* receiving PREPARE, I think (states b1, c1)). I've run about 3 million trials since then. Eventually the unseen events will occur, or we can conclude they can't happen and take them out - I've seen events require 5 million trials or so to turn up. The exerciser program keeps the "best" example of each transition, with best defined as shortest. Sometimes this means it's a rather degenerate case (e.g. disruption occurs immediately afterwards), but it avoids getting the cases where the same message is sent repeatedly. (the next-event choosing bit of the exerciser isn't terribly bright) The state tables shown are as they are currently, having applied the decisions in San Jose, including: hazard (problem) can be detected and possibly reported *before* a persistent record of the problem is made - this inevitably makes it possible that the problem report is lost (when neither side makes a persistent record before they crash) resign/no-rsp-req can be sent before PREPARE is received (implementations with a persistent active state will typically ask for a response, those that have no such, and don't mind if a message lost causes cancellation will not ask for a response THESE TABLES ARE STILL DRAFT and are part of the ongoing work of the BTP TC - I'm circulating them to help people review the defining tables for correctness. (The exerciser can find if the tables are inconsistent, or can lead to an undesirable final state (e.g. contradiction where neither side ever noticed, but it can't check if all the things that are supposed to be allowed are allowed) Unpack the two files to the same directory and open the tblproofindex.html in your browser. Enjoy! Peter ------------------------------------------ Peter Furniss Technical Director, Choreology Ltd email: peter.furniss@choreology.com phone: +44 20 7670 1679 direct: +44 20 7670 1783 mobile: 07951 536168 13 Austin Friars, London EC2N 2JX
[Date Prev] | [Thread Prev] | [Thread Next] | [Date Next] -- [Date Index] | [Thread Index] | [Elist Home]
Powered by eList eXpress LLC