From: Elias Sinderson (elias@cse.ucsc.edu)
Date: Wed Jun 02 2004 - 11:56:20 MDT
Michael Wilson wrote:
>Needless to say, the SIAI will /not/ be using your Puny Earth Science. The bare /minimum/ capability likely to be used here is something [...] which incorporates a fast ACID transactional database [...]
>
>
ACID transactions are generally not well suited for use within large
distributed / decentralized systems due to the inherent difficulty of
achieving simultaneous agreement or concensus among computational peers.
According to Lynch [1], "the impossibility of consensus is considered to
be one of the most fundamental results of the theory of distributed
computing." Indeed, assuming that one can create AGI upon the foundation
of a consensus-based architectural style may well be folly, especially
given that the only extant model of intelligence (consciousness,
sentience, or whathaveyou) appears to not follow this model. Instead of
the DB-centric ACID properties that sfotware engineers have come to
cherish, consider the BASE principles of (B)est effort data transfer,
(A)pproximate estimates of current values from data already received,
and (S)elf-centered trust management, and (E)fficient summarization of
data to be sent [2].
Regards,
Elias
[1] Lynch, N. A. Distributed Algorithms. Morgan Kaufmann, San Mateo, CA,
1996. 904pp.
[2] Khare, R., Extending the Representational State Transfer (REST)
Architectural Style for Decentralized Systems. Doctoral Dissertation,
University of California, Irvine, CA, September 2003.
http://www.isr.uci.edu/~rohit/DecentralizingREST.pdf
This archive was generated by hypermail 2.1.5 : Tue Feb 21 2006 - 04:22:38 MST