On Tue, Dec 9, 2008 at 10:46 AM, David Meyer <[EMAIL PROTECTED]> wrote: > In today's world, hosts have pretty much complete > information about which <source,destination> pairs are > available.
That really isn't true in *today's world.* Even in the best case scenario, the application (NOT the host) only knows the destination addresses which provide a given service. The app has no idea whether those destinations are a single host where the addresses could reasonably be interchangeable, multiple hosts where the app is expected to connect to one of them and stick to it or some combination of the two. In IPv4 the host also has no clue which of the source addresses besides the primary one on the exit interface are valid candidates for remote operation. > But yes, if you convert GSE into something like SHIM6 (by > letting hosts know their source RG), then the locator > path liveness problem is something like O(n^2). Not necessarily. You may be able to compute the path-source-liveness separately from the path-destination-liveness and let the host scholastically converge on the currently-optimal pair from the two separate functions. If so then the overall path liveness problem is only O(n). Think about it: a dead destination adds relatively evenly to the source deadness function. Only source deadness adds unevenly to the source deadness counter, identifying a source that can be excluded. Even in the worst case where each source works with one, only one and exactly one destination, the randomness in the function will push convergence towards one of the sources which will feed back into the destination function so that it likes the matching destination. Same for the destination function. On Tue, Dec 9, 2008 at 10:59 AM, RJ Atkinson <[EMAIL PROTECTED]> wrote: > Would you agree that: IF the host selects the Source Locator, > which (as you say) is NOT what the GSE documents say, then > the source host could use the same strategy as in today's Internet > (e.g. search the space of available <s,d> pairs to find a workable pair). How do you determine "workable pair" in a stateless protocol like UDP? Regards, Bill Herrin -- William D. Herrin ................ [EMAIL PROTECTED] [EMAIL PROTECTED] 3005 Crane Dr. ...................... Web: <http://bill.herrin.us/> Falls Church, VA 22042-3004 _______________________________________________ rrg mailing list [email protected] https://www.irtf.org/mailman/listinfo/rrg
