bcn@cs.washington.edu (Clifford Neuman) (12/06/90)
Archive-name: ftp/research/prospero/1990-12-03
Archive: june.cs.washington.edu:/pub/pros-srv-usr.tar.Z [128.95.1.4]
Original-posting-by: bcn@cs.washington.edu (Clifford Neuman)
Original-subject: More information on Prospero
Reposted-by: emv@ox.com (Edward Vielmetti)
Here is more information on Prospero. This message describes the
features and goals of the system. Some of the features described in
this message are not yet supported in the prototype.
~ Cliff
---
The Prospero Distributed Operating System
Main Goal
As systems grow larger, it becomes harder for the user to organize,
identify, and locate the information and services that are of
interest. The Virtual System Model is an approach to organizing large
systems that makes it easier for users to keep track of and organize
such objects. Prospero is a distributed operating system based on the
Virtual System Model.
Advantages
Prospero allows users to build their own ``virtual systems'' from the
files and services that are available to them. It is easier for users
to keep track of objects in the resulting name space. Tools are
provided that allow a name space to be specified as a function of
other name spaces. This relieves part of the burden of keeping name
spaces up-to-date. Additionally, the tools that are provided allow
information to be organized in ways that make it easier for other
users to identify new information of interest to them, but which they
have not yet seen.
A disadvantage of supporting multiple name spaces is the lack of name
transparency: the same name may refer different objects when used by
different people. Prospero addresses this problem by supporting
closure.
Description
A virtual system defines a view of the world centered around the
user. Those files and services of interest to the user have short
names, while the names of objects that the user is less likely to
access are much longer. Users can specify parts of their name space
as a function of one or more other name spaces. This is accomplished
using two tools: a filter is a user defined program, associated with a
link, which changes the way one views objects seen through that link;
a union link takes the contents of a linked subdirectory and makes
them appear as if they are part of the directory containing the link.
Naming in Prospero is ``object-centered''. In object-centered naming,
multiple name spaces are supported, and each object has an associated
name space (forming a closure). By supporting closure, it is always
clear which name space is to be used when resolving a name.
The global name space forms a generalized directed-graph, but users
don't see it as such. Instead, they see the name space associated
with the active virtual system. Each virtual system specifies the
node in the directed graph that is to be the root of its associated
name space. When projected from that root, the name space appears to
be hierarchical.
A prototype is presently implemented on top the Unix operating system.
The prototype focuses on the file system and allows users to organize
references to information scattered across Internet archive sites.
Implementations are planned for other systems.
Objects that are accessible through Prospero may reside on systems of
different types. The Prospero protocol makes extensive use of types,
allowing full support for heterogeneity. Because Prospero is intended
to scale across administrative boundaries, support for heterogeneity
was an important requirement in its design.
Prospero runs as a network service. Each system that contains
objects that are to be accessed through Prospero must run the server.
The servers collectively implement a distributed directory service,
independent of the mechanism used to actually store the objects. In
addition to directory information, the directory service also stores
information about individual objects. This information includes
replication, closure, and access control information, a list of cross
references, and a (possibly incomplete) list of back-links to the
directories which name the object.
Client applications can run on any system, not just those running the
server. No kernel changes are required. Several applications
supporting the Prospero name space are provided and a library
redefining the open system call allows existing applications to use
the Prospero name space. A kernel implementation is planned. Such an
implementation would avoid the need to relink applications, and would
thus reduce the size of the resulting application binaries. The
kernel and non-kernel implementations will interoperate, so users can
choose whichever implementation is most suitable for their
environment.
The Prospero protocol is presently implemented on top of UDP (in the
TCP/IP suite of protocols). The protocol could be easily implemented
on top of other protocols, and simultaneous support for multiple
protocols has been made possible by the extensive use of types,
especially in passing network host names and addresses.
Access to individual objects in Prospero is mediated by existing
access mechanisms, and the security for such access is defined by
those mechanisms. Access to read or write directory information may
be specified by access control lists associated with directories, or
with individual links within a directory. Prospero uses
interchangeable authentication mechanisms. Authentication information
is passed as an authentication type field, and a field for data
specific to that type. The prototype uses an authentication type of
``assertion'', a future release will incorporate Kerberos
authentication, and other authentication mechanisms may be added as
needed.
Objects in Prospero may move around, and the links to the object will
continue to work. This is accomplished through a combination of
forwarding pointers, callbacks, and timeouts.
Scalability was a key consideration in the design of Prospero.
Prospero is intended to tie together pieces from numerous systems that
are geographically distributed and which cross organizational
boundaries. In addressing scale, Prospero is not just concerned with
the demands placed on the system. The demands placed on the user are
just as important. The user-centered view supported by Prospero is
one way of limiting those demands.
Virtual systems may be set up by users or organizations. They may be
created for individuals, projects, or any other purpose for which a
separate view of the world might be useful. It is expected that
organizations will maintain directories organizing objects in different
ways, and that these directories can be incorporated into the virtual
systems of those that need the information. Among these directories
might indices by author, project, subject, or any other key fields.
Users can find object by looking for them in the appropriate index, or
by browsing through related virtual systems.
STATUS
A prototype is running at the University of Washington and at a few
sites elsewhere on the Internet. A beta release is available. Those
interested in being a beta test site should contact INFO-PROSPERO.
MAILING LIST
INFO-PROSPERO@CS.WASHINGTON.EDU is a mailing list on which
announcements will be made as new pieces of the system become
available. To be added to the list, send a request to:
INFO-PROSPERO-REQUEST.
Contact: B. Clifford Neuman bcn@cs.washington.edu
Department of Computer Science and Engineering +1 (206) 543-7798
University of Washington, FR-35
Seattle, Washington 98195
References:
@TECHREPORT{vsmtp,
AUTHOR = "Neuman, B. Clifford",
TITLE = "The {V}irtual {S}ystem {M}odel: A Scalable Approach to
Organizing Large Systems (A Thesis Proposal)",
INSTITUTION = "Department of Computer Science, University of Washington",
YEAR = 1990,
MONTH = "May",
NUMBER = "90-05-01"}
@ARTICLE{nfclosure,
AUTHOR = "Neuman, B. Clifford",
TITLE = "The Need for Closure in Large Distributed Systems",
JOURNAL = "Operating Systems Review",
MONTH = "October",
YEAR = 1989,
VOLUME = 23,
NUMBER = 4,
PAGES = "28--30"}
@INPROCEEDINGS{wsvsm,
AUTHOR = "Neuman, B. Clifford",
TITLE = "Workstations and the {V}irtual {S}ystem {M}odel",
BOOKTITLE = "Proceeding of the 2nd IEEE Workshop on
Workstation Operating Systems",
YEAR = 1989,
MONTH = "September",
PAGES = "91--95",
NOTE = "Also available as University of Washington Technical
Report 89-10-10. Also appears in the
{\it Newsletter of the IEEE Technical Committee on
Operating Systems}, Volume 3, Number 3, Fall 1989"}
@TECHREPORT{vsmldos,
AUTHOR = "Neuman, B. Clifford",
TITLE = "The {V}irtual {S}ystem {M}odel for Large Distributed
Operating Systems",
INSTITUTION = "Department of Computer Science, University of Washington",
YEAR = 1989,
MONTH = "April",
NUMBER = "89-01-07"}
@UNPUBLISHED{mrdvsm,
AUTHOR = "Neuman, B. Clifford",
TITLE = "Managing Replicated Data within the
Virtual System Model",
NOTE = "Department of Computer Science and Engineering,
University of Washington",
YEAR = 1990,
MONTH = "May"}