.m From dahlin@postgres.Berkeley.EDU Tue Aug 31 11:51:17 1993
17.op/TEA.DAP.dahlin.etal
.ls 2
.na
.LP
Design of a Large Capacity Object Server Supporting Earth System 
Science Researchers

Mike Dahlin, Cliff Mather, and Randy Wang

(Professors T. E. Anderson and D. A. Patterson)

MICRO 1992-93 and (NSF) IRI-91-16860*

We are investigating file systems to integrate multi-terabyte
storage and wide area network access for a large number of users.
There are two primary challenges to providing such
a system: performance and reliability.

We are currently working to use aggressive caching to 
improve performance as much as possible. First 
we will use protocols such as write-back and 
write-ownership to minimize server traffic. We will also
take advantage of cache-to-cache transfers of data. Finally,
we will use clustering and directories to coordinate 
different caches of a local area network to reduce server
overhead, reduce the number of redundant copies of
cached data, and allow parallel access to large files.

Reliability is also a big concern. A system spread over hundreds
or thousands of caches, tens of LANs, many tertiary storage devices,
and several WANs has a large number of individual components that
may fail. Further, as file systems store more and more data, 
the potential cost of data loss or unavailability increases. 
Data replication can address these problems, but managing these
extra copies entails many of the costs caching was meant to
avoid.


*Support for Mike Dahlin