From: Jim Whitehead (ejw@ics.uci.edu)
Date: Mon Jul 17 2000 - 10:34:54 PDT
Adam Rifkin writes:
> Speaking of which, I can't believe I'm missing TWIST...
>
> http://www.isr.uci.edu/events/twist/twist2000/
>
> Greg or Lucas or Beberg or Jim or anyone else there, how about posting a
> summary of what's been going on for those of us too busy to attend?
For me, these were the major messages I learned at TWIST (Mark, Adam, Lucas,
Roy, Greg, feel free to add on to this).
I came into TWIST thinking that centralized and decentralized are attributes
of an entire system, and left realizing that they can only be used to
describe aspects of a system. That is, it is usually incorrect to describe
a system as being totally centralized, or totally decentralized, since most
systems have aspects of both.
Centralized and decentralized are distinct from localized and distributed.
The localized/distributed distinction is a technical one: a system is
typically considered to be localized when its parts are contained within a
single computer, or comprises a single processing node. A distributed
system is one that has many processing nodes. The
centralization/decentralization distinction is concerned with organizational
boundaries, and organizational *control*. Typically, when we say
"centralized system", we mean a "centrally *controlled* system". So, you
could have a distributed system still be controlled by a single
organization, and hence it is a centralized system.
But, there are many dimensions on which this analysis can take place. One
good example was the Internet: it is a decentralized system to the extent
that multiple organizations control the operation of various subnetworks,
but yet the IETF provides a fairly centralized authority for architectural
change control over the Internet. Of course, even here you need to be
careful, since it is possible for organizations other than the IETF to field
new Application layer protocols, and even some Transport layer ones, but
only the IETF really has change control over TCP and IP.
I thought Alvina Nishimoto's talk about her experiences at HP was
interesting because she reported on the trend of customers abandoning
distributed solutions in favor of localized ones (although in general her
talk was too much of a marketing spiel for an academic workshop). That is,
Moore's Law is increasing the power of computers to the point where many
applications can be hosted on a single computer, and their growth rate is
typically outpaced by advances in Moore's Law. Adam Beberg also touched on
this point in his talk:
http://www.isr.uci.edu/events/twist/twist2000/presentations/beberg/ Adam's
talk was also good in that it mentioned some of the driving technical
factors for localized and distributed applications.
- Jim
This archive was generated by hypermail 2b29 : Mon Jul 17 2000 - 10:43:08 PDT