[OAI-implementers] deep nesting problem in DP9

Xiaoming Liu liu_x@cs.odu.edu
Tue, 12 Nov 2002 13:05:20 -0500 (EST)


I have a question regarding DP9 service and hope to solicit some consensus
in OAI community (After discussions with Jeff Young). DP9 is a gateway
service which allows general search engines, (e.g. Google, Inktomi) to
index OAI-PMH-compliant archives. For more information please see 


DP9 uses resumptionToken to handle large archive, a resumptionToken link
is put at the bottom of each listidentifiers screen. Each time a web
crawler follows a resumptionToken link, it takes them down another level
in the page hierarchy. In the case of XTCat (4M+) records, there will be
8000+ pages deep. But actually most crawlers will only do 4~5 levels, so
there will never be a chance of whole XTCat being harvested. 

Jeff and me discussed this question and came up several possible
solutions. All of them require some levels of actions in data provider
side, We hope the final solution is "general" and that's why I post it in
the list.

1) Create many small bins based  on timestamp and sets, the DP9, of
course, must be intelligent to do the correct split. This probably can be
done by a partial pre-harvest. This solution should be applied for most
data providers. But it fails if a data provider has a large number records
with same datestamp. 
  pro) no change in OAI spec.
  con) data provider should not have a large number of records with same

(snippet from Jeff's email)
2) Create a new verb named ListResumptionTokens that
returns a complete set of stateless resumptionTokens.
  pro) easy to implement in DP9 side.
  con) requires modification of the OAI spec.

3) Define a new <description> element to be returned in the Identify
response that provides with the information DP9 need to automatically
generate stateless resumptionTokens.
  pro) doesn't require any modification of OAI spec.
  con) requires individual repositories to voluntarily provide this

Probably somebody can come up with a better idea, and we may  reach a
consensus of which way to go ;-) Please send to the list if you
have any input. 

best regards,