From: Steve Arnold (steve.arnold4_at_verizon.net)
Date: Sun Feb 01 2004 - 23:17:18 PST
This is not the first paper on Grapevine, but in this paper, the authors are
giving us on update on the system, describing how they have implemented it,
how it has been successful, and how they have encountered a few problems.
They start out by providing a brief overview of how Grapevine is structured.
It is a distributed system for managing registrations (users) and mailboxes.
In the Xerox implementation, they had 17 servers and a network with gateways
connecting them. This is a significant scaling up from the initial
implementation and it sounds like Xerox had made this an important business
application.
Probably one of the most important things about Grapevine is that it was
truly a distributed application. It had a system of distibuting the
resources throughout the network: there is no single controlling server.
When the user community grows, you add more registries (usually hardware),
not necessarily better machines. It also employed replication in several
ways.
The configuration was very open-ended. It seems like you needed some experts
to help you decide how to configure it (where storage should go, how things
are connected, how groups are laid out, etc). In their experience,
configurations could have a major impact on performance. Perhaps they needed
a separate utility for heuristically finding an optimal configuration.
Grapevine was also used for other applications, such as authentication for
file systems. It was a central point of authentication and permissions,
rather than maintaining several different systems. It was a bit like we
might see in Windows networks today.
They go into several problems that they encountered. I thought it was
interesting that they became more and more resistant to make changes because
of the potential for regressions. I'm guessing it was no where near the
magnitude of today's systems!
This archive was generated by hypermail 2.1.6 : Sun Feb 01 2004 - 23:16:28 PST