Re: Summary, was Re: Every time ..., was Re: General formula

[Lynn.Wheeler@xxxxxxxxxxxxx]

>From the AADS/business standpoint ... there is alwas an account record someplace
containing a bunch of fields and the certificate contains R/O replication from
some set of those fields (taken at some point in time). The question then is
two-fold ... 1) frequency/probability that specific information is changed in
the master account record and 2) procedures/processes of propogating
invalidation/update signals to all the locations containing read-only copies
(and 3) discontinuities
introduced with latency in the invalidate/update signals).

This characterization has been extensively modeled for cpu caches, database
caches, replicated databases, filesystem caches, and replicated filesystems.
Modeling scenerios are typically master/slave (master/original location persists
over long duration with one or more read/only copies), single writer (only one
master at a moment, but it may float), and multiple writers. Certificates tend
to most closely resemble master/slave ... with the account record at the
certificate issuer representing the "master copy" (with the number of places a
certificate information replication occuring potentially arbritarily large).
Sometimes the database models will also look at things like duration and size of
information inconsistency and the effect on correct operation (although this
tends to be more data/application sensitive).