1. Distribution Alternatives for Superimposed Information Services in Digital Libraries
Sudarshan Murthy, David Maier, Lois Delcambre
Department of CSE, OGI School of Science & Engineering at OHSU

2. Digital Libraries: Where to Put the PartsOr
Digital Libraries: Where to Put the Parts

3. Superimposed Information*
*Picture courtesy of Prof. James Pankow, Dept. of EBS, OGI
Superimposed Information*
People often superimpose new information onto existing information
Annotations, summaries, …
They use many means
Mark up paper
Place sticky notes on the paper
They combine existing information and their interpretations to get “their” view
25-May-19
Distribution Alternatives for Superimposed Information Services

4. Distribution Alternatives for Superimposed Information Services
Group
Item
25-May-19
Distribution Alternatives for Superimposed Information Services

5. Superimposed Information Services in DLs
Related DL patron activities
Create, view, and share annotations and summaries
Superimposed information services can support and potentially extend these activities
25-May-19
Distribution Alternatives for Superimposed Information Services

6. Component-based Architectures
We use a component-based middleware architecture, SPARCE, to support superimposed information management
Component-based architectures have been used to support annotation and metadata over DL information
OAI-PMH, ODL
Flexibility of deployment (distribution) is a benefit of using component-based architectures
25-May-19
Distribution Alternatives for Superimposed Information Services

7. Distribution Alternatives for Superimposed Information Services
Goals
Propose distribution alternatives for architectural components to add superimposed information services to a digital library
Propose metrics to compare performance of the alternatives
We use SPARCE to illustrate the alternatives
25-May-19
Distribution Alternatives for Superimposed Information Services

8. Distribution Alternatives for Superimposed Information Services
Outline
Motivation
Background
Superimposed information management, SPARCE
Distribution Alternatives
Metrics
Example alternatives
Discussion
Conclusion
25-May-19
Distribution Alternatives for Superimposed Information Services

9. Superimposing Information
Overlaying new information on top of selected, existing information
Add new data
Impose new schema or model
Mark is a reference to base element
Many implementations, ~ one per base type
Addressing scheme depends on base type
Heterogeneous sources: Word, Excel, PDF, HTML,…
25-May-19
Distribution Alternatives for Superimposed Information Services

10. Distribution Alternatives for Superimposed Information Services
Excerpts and Contexts
Name
Value
Excerpt
Finding of No Significant Impact … not be necessary
Font name
Times New Roman
HTML
<p><b><span style='font-size:14'>Finding of No Significant Impact</span></b></p> <p><span style='font-size:12'>Based on ...</span></p>
Excerpt is the content of a marked region
Context is information related to a mark
What constitutes a context varies
A mediator called context agent retrieves context of a mark
25-May-19
Distribution Alternatives for Superimposed Information Services

11. Distribution Alternatives for Superimposed Information Services
SPARCE
The Superimposed Pluggable Architecture for Contexts and Excerpts
Middleware for superimposed information management
Provides mark and context management services
Use the same programmatic interface to work with any base type
25-May-19
Distribution Alternatives for Superimposed Information Services

12. SPARCE Reference Model
Context Manager
Base Info
Superimposed Application
Base Application
Mark Manager
Superimposed Info
Marks
25-May-19
Distribution Alternatives for Superimposed Information Services

13. Distribution Alternatives
25-May-19
Distribution Alternatives for Superimposed Information Services

14. Distribution of Components
Distribution affects latency, feature set, maintenance cost, load, security, …
Latency example: SOAP-based web service with 400-byte array input and output
Local: 7.87 ms
LAN (1 hop): ms
WAN (> 18 hops): ms
25-May-19
Distribution Alternatives for Superimposed Information Services

15. Distribution Alternatives for Superimposed Information Services
Metrics
Latency
e.g., Time to serve a patron’s request
Reduces as distance reduces
Load on server
e.g., Number of active processes
Increases with the number of components
Maintenance cost
e.g., Effort to update a component
Thin clients are less expensive to maintain
25-May-19
Distribution Alternatives for Superimposed Information Services

16. Distributing SPARCE’s Components
Context Manager
Patron Application
Base Application
Base Info
Mark Manager
Superimposed Info
Marks
Patron’s computer ?? ??
DL Server
25-May-19
Distribution Alternatives for Superimposed Information Services

17. Distribution Alternatives for Superimposed Information Services
Latency
tpm
tmb
tbb
Patron App
Managers
Base Apps
Base Info
Tpm
Tpm = tpm + tmb + tbb
The overall latency Tpm may be reduced by reducing tpm, tmb, and tbb
25-May-19
Distribution Alternatives for Superimposed Information Services

18. Distribution Alternatives for Superimposed Information Services
Alternative A
tpm (Low)
tmb (Low)
tbb (Low)
Patron App
Managers
Base Apps
Base Info
Patron’s computer
DL Server
Base documents are usually downloaded once (per session), so tbb is likely to be low
Patron must have all base applications available locally
25-May-19
Distribution Alternatives for Superimposed Information Services

19. Distribution Alternatives for Superimposed Information Services
Alternative B
tpm (High)
tmb (Low)
tbb (Low)
Patron App
Managers
Base Apps
Base Info
Patron’s computer
DL Server
Minimize number of round trips to managers
Mark activation would be unable to exploit base apps on patron’s computer (“call back” may be possible)
25-May-19
Distribution Alternatives for Superimposed Information Services

20. Distribution Alternatives for Superimposed Information Services
Alternative C
tpm (Low)
tmb (High)
tbb (Low)
Patron App
Managers
Base Apps
Base Info
Patron’s computer
DL Server
Minimize number of round trips to base apps
Mark activation would be able to exploit patron’s base apps
25-May-19
Distribution Alternatives for Superimposed Information Services

21. Distribution Alternatives for Superimposed Information Services
Alternative D
tpm (High)
tmb (High)
tbb (Low)
Patron App
Managers
Base Apps
Base Info
Patron’s computer
Middle tier
DL Server
Managers could be shared by more than one DL (server)
Mark activation would be unable to exploit patron’s base apps
25-May-19
Distribution Alternatives for Superimposed Information Services

22. Distribution Alternatives for Superimposed Information Services
Alternative E
tpm (High)
tmb (Low)
tbb (Low)
Patron App
Managers
Base Apps
Base Info
Patron’s computer
Middle tier
DL Server
Managers and base apps could be shared
Mark activation would be unable to exploit patron’s base apps
25-May-19
Distribution Alternatives for Superimposed Information Services

23. Summary of Alternatives
Maintenance Cost (Location)
Round-trip time
Alternative
Patron apps
Managers
Base apps
DL Server Load
tpm
tmb A
High (P)
Low B
Low (P)
Low (D)
High C
Medium D
Low (M) E
Locations: P=Patron’s computer, D=DL Server, M=Middle tier
Prioritize among latency, server load, maintenance cost
Prioritize the latency terms tpm and tmb to minimize
25-May-19
Distribution Alternatives for Superimposed Information Services

24. Distribution Alternatives for Superimposed Information Services
Discussion
Heterogeneous distribution alternatives
Patron apps of different capabilities; a patron may use DLs that employ different alternatives
Sharing annotations and marks
Marks may be shared or replicated
Mark manager could be useful for other services such as indexing
Security
Minimize #interface points; narrow functionality
25-May-19
Distribution Alternatives for Superimposed Information Services

25. Improving Performance
Minimize number of round trips
Batch requests where possible
#round trips may not be an issue for high bandwidth. e.g., bandwidth between middle tier and DL server likely higher than between patron and DL
Replication
Replicate DL server to handle large loads
Instantiate manager modules on patron’s computer for local base apps; use DL server at other times
Cache context
25-May-19
Distribution Alternatives for Superimposed Information Services

26. Distribution Alternatives for Superimposed Information Services
Conclusion
Enhancing digital library services with superimposed information services can benefit patrons
Several distribution alternatives exist, each have trade-offs that DL architects must consider
25-May-19
Distribution Alternatives for Superimposed Information Services

27. Questions? Visit http://www.cse.ogi.edu/sparce 25-May-19
Distribution Alternatives for Superimposed Information Services

28. Distribution Alternatives for Superimposed Information Services
Related Work
InfoBus: Roscheisen and others, 1998
UC Berkeley DL Project: Wilensky, 2000
ODL: Suleman and Fox, 2001
PMH: OAI, 2002
XOAI-PMH: Suleman and Fox, 2002
FEDORA: Staples and others, 2003
COM, CORBA, SOAP
25-May-19
Distribution Alternatives for Superimposed Information Services

29. Placement affects Latency
Mean round-trip time (milliseconds)
Input
Output
Local
LAN
WAN
None
2.94
3.13
146.74
10 integers
3.36
137.53
One 400-byte array
7.87
10.53
689.39
Mean round-trip times for SOAP-based web service method calls
25-May-19
Distribution Alternatives for Superimposed Information Services

30. Distribution Alternatives for Superimposed Information Services
Mark Manager
25-May-19
Distribution Alternatives for Superimposed Information Services