OBJS Technical Note
Information Access Services
Craig Thompson
Object Services and Consulting, Inc.
July 1998
One of the often cited kinds of agents are information agents that are
intended to access heterogeneous data sources. Below is an outline
of the kinds of properties and services that are found in I*3 systems.
The DARPA I*3 program mostly dealt with querying heterogeneous data sources,
something that coABS needs to do. While most I*3 like systems do
not invoke the agents paradigm, some do, for instance the MCC
InfoSleuth project. Some of these notes come from an overview
paper by Nancy Lehrer (ISX) on the ACME mediator architecture (sent to
me for review, not yet released). ACME is intended to be a prototypical
I*3 architecture and implemented system.
-
heterogeneous data sources and domain object models
-
properties:
-
LAN or WAN at known location or dynamically available
-
relational
-
OODB
-
information retrieval
-
simulators
-
geographical informaiton systems
-
semi structured sources
-
html
-
xml
-
other formatted sources
-
image
-
information integration services
-
data source wrappers
-
ifilter
-
subscribe, notify
-
monitor
-
push
-
pull
-
persistence
-
replication
-
caching
-
query decomposition
-
multi tier queries - get info from one source to complete query
-
source discovery
-
trader, brokering, yellow pages, matchmaking
-
source selection
-
transformation, translation services, semantic integration and transformation,
includes unit conversion
-
domain and ontology services
-
term translation
-
correlation services, name/place
-
query translation
-
fusing
-
stream
-
reflection
-
control
-
properties
-
indexing
-
working in parallel
-
iterative query reformulaiton
-
change propagation
-
if data sources change (alternate source)
-
if query changes
-
context of query
-
statically
-
dynamically - specified in plan, case-based reasoning, workflow
-
metrics
-
operation effectiveness - info quality, timliness and cost of retrieval
-
breadth of coverage - completeness, data source complexity
-
maintenance or evolution over time
-
metadata properties
-
operations covered
-
data access systemic properties
-
scaleable
-
plug and play, open, component-based
-
transparency of data location, access langauge, and protocol
-
user queries
-
static set vs dyanamic and frequently changing based on user task and need
-
distributed
-
all local to user machine, distributed on LAN, on WAN
-
how homogeneous is content
-
uniform across all sources
-
unique per source
-
overlapping sources
-
partially redundant w inconsistencies
-
incomplete
-
homogeneity of information sources
-
all in same query language
-
syntatic differences
-
multiple kinds
-
source size - # of entity types
-
semantic impedence w user
-
no vs all query term translation
-
source responsiveness
-
quick (<10 sec; medium - up to 60 sec; slow - <3 min; very slow -
<3 min; batch overnite; unreliable
-
cost
-
availability, permission, quality, quantity, capabilities, type,
-
synchrony
-
synchronous query and response
-
asynchronous
-
interuptable
-
desired response control
-
all answers at once
-
client controls
-
server controls
-
top N answers
-
requence of data source changes
-
never, seldom (<1/yr); often >1/mo; continuous - real-time feed
-
frequency of user query changes
-
never, seldom (<1/yr); often >1/mo
The above does not yet take into account the multi-query optimization and
channel architectures of the DARPA AICE program.