...Shore
This overview document is based heavily on the paper ``Shoring Up Persistent Applications'' presented at the 1994 ACM SIGMOD Conference. This document was created by editing that paper to reflect the state of and plans for the Shore system as of the alpha code release.
...Shore
This research is sponsored by the Advanced Research Project Agency, ARPA order number 018 (formerly 8230), monitored by the U.S. Army Research Laboratory under contract DAAB07-91-C-Q518.
...Shore
Copyright (c) 1994 Computer Sciences Department, University of Wisconsin - Madison. All Rights Reserved.
...data
SDL is very closely related to ODL, a data definition language recently proposed as a standard by ODMG, an OODB vendor consortium [Cat93].
...peer-to-peer
The alpha release supports only single server operation.
...database
In the alpha release, heaps are not available to application programmers; for now, they are only used internally, for storing the variable-sized components of an object. Also, demand-paging of large heaps is not implemented in the alpha release, so an object's heap is brought in its entirety when the object is faulted into the object cache.
...calls.
The alpha release does not include a Unix system call library; that is planned for a later release. It does include an NFS server, so Unix system calls can be used with Shore objects if the SHORE file system is NFS-mounted on a Unix host.
...OODBs.
Many of the commercial systems use a tree-structured name space for naming databases, but not for naming or organizing individual persistent objects or collections.
...objects
in the alpha release, modules and type objects are not special objects
...stamps.
The semantics of time stamps are slightly different from those of Unix in order to make them efficiently maintainable while retaining their usefulness to applications that rely upon them.
...section.
In the alpha release, cross references can be followed through NFS, but they cannot be removed through NFS. When a dangling cross reference is followed through NFS, somewhat odd behavior results: the NFS client (host) issues a message, ``name not found'', in which ``name'' is the name of the cross reference.
...server.
In the alpha release, only the NFS file server is provided; the Unix compatibility library is not yet available.
...defined.
In the interest of brevity, some of the details have been omitted.
...binding.
Other bindings are planned, of course, but work on them will not begin until Shore is fully operational and delivering good performance through its C++ binding. Moreover, note that the features described here differ somewhat from their description in the SIGMOD-94 Shore paper. The description here represents the alpha release C++ binding; we expect that the binding will become cleaner over time.
...definitions.
Note, however, that any Shore application can create type objects. For instance, one could write a graphical schema design tool to create type objects and install them in the database.
...will
in a future release
...server.
At present, disk volumes are not replicated.
...applications.
In the alpha release, remote server support is not provided; the server-to-server data shipping and caching code is still under construction at present.
...server.
An OID contains a volume identifier. The server uses a global volume-location service to find the appropriate server and establishes a network connection if necessary.
...large
with the alpha release, the object cache does not remove or replace objects
...system
the alpha release uses no authentication
...database.
An SQL server value-added server is an example of a rather different use of the Shore Server; the upper layers and type system of Shore would essentially be thrown away, and the facilities provided by the Shore Storage Manager would be used in the construction of a completely different, customized server.
...protocol
planned for a later release
....
The alpha release does not include multi-granularity locking, as that code is still being written along with the server-to-server code mentioned earlier.
...server.
Generally, the computational cost of the redo is small enough to be ignored, especially when compared to the cost of receiving a page of data via the network.

zwilling@caseus.cs.wisc.edu
Thu Nov 3 14:14:30 CST 1994