RFC005

TITLE: DAS Caching
Author Matthew Pocock (mrp@sanger.ac.uk)
Dependancies: Directory services (e.g. RFC 2), Authentication
Version: 1
Date: 15 August 2001


Requirements
------------

There are likely to be multiple DAS servers either side of the atlantic. 
Requests can potentialy become prohibitavely slow over these distances. 
Some method to cache DAS data would potentialy remove the need to 
explicitly set up mirror servers with all the ID cannonical copy issues 
this may introduce.


Mechanisms
----------

A seperate class of servers termed DAS-Cache could be introduced which 
localy store the results of some requests. Caches can potentialy 
retrieve data from other caches, which in turn retrieve data from caches 
untill some cache fetches it directly from a DAS non-cache server.

This would, for example, allow centralised DAS-Caches to be put in place 
geographicaly arround the web, as well as within each organization and 
potentialy even in parallel to an individual browser. Requests from the 
browser can be cascaded through the cache hierachy when apropreate, 
reducing the need to move data redundantly between DAS nodes.


Issues
------

It may be necisary to define DAS-Cache services for each DAS service 
individualy, or it may be possible to define a single DAS-Cache service 
that caches all DAS-compatible data.

Multiple DAS requests potentialy overlap in what they requre. Any cache 
technology would idealy understand this and be able to optimaly fetch 
data for parts of the query from local stoorage and other parts from 
remote stoorage.


Relationship to other DAS/2 services
------------------------------------

DAS-Caches should be registered with a directory service allong with 
other DAS services. They must be associated with enough meta-data to 
allow clients to discover which caches are appropreate for fetching what 
data.

Potentialy, caches may end up providing a gnutella-like infrastructure 
for replicating data near to data consumers. This will potentialy lead 
to a requirement for data validation (how many times have you downloaded 
files from gnutella to find that the content doesn't match the file-name).