[OAI-implementers] on the OAI-PMH data model
Fabio.Simeoni at cis.strath.ac.uk
Fri Mar 18 11:54:50 EST 2005
> we are.
good, yours is example number 2 I've gathered, after Library of Congress.
I suppose you also generate the different formats on the fly rather than
actually store persistent versions, right? That is, I assume that, for each
item, you only persist record datestamps for each format you disseminate,
along with a master record datestamp (like Library of Congress), is this
correct? And also that every time you actually change the item/master-record
-- in the day-to-day operation of your repository -- you also update all the
datestamps you are maintain.
In any case, you're doing this to potentially avoid SPs having to repeat a
historical harvest of records which have not actually changed but appear to
have because a mapping of the master record onto another format -- say a
newly adopted format of no relevance to old SPs -- has. Is this also
>From a repository-centric perspective, this seems to me quite a service you
do to SPs and the OAI-PMH. I find it interesting for it shows that, even
within the OAI-PMH, some service costs may be absorbed by DPs rather than
I wonder now if DP could be as good-willed as you are and yet avoid the
bookkeeping you are doing by:
(i) avoid updating the item/master-record datestamp when a format mapping
(ii) use out-of-band techniques to tell harvester which are interested in
the mapping change that, as a result of that change, they should do a
All the other harvester would still be spared from doing costly and yet
irrelevant historical harvests. Of course, this requires DPs to know who's
harvesting them and for what.
So, at least 4 possible scenario for DPs:
1) only one dissemination format: don't bother and attach datestamps to
2) many dissemination formats: bother and keep/mantain datestamps for each
3) many dissemination formats: bother, attach datestamps to
items/master-records, and tell only some harvesters to re-harvest in case of
4) many dissemination formats: don't' bother (and be evil)
Any comments appreciated.
More information about the OAI-implementers