Multisearch Architecture

Map: <query, service> - the inital key, value pair are inputed to the system. The query can be hashed. The service is represented by a ServicdeWritable object.

Different kinds of Clients can be generated by the Map instruction which will be needed to call the correct services. The OutputCollector will collect the ResultSet output, and we can move to the reduce part of Hadoop.

Reduce: <query, ResultSet> - Since the Reduce function will also contain the merge, we should only get one output. We might need to use the combine functionality to speed up the connection.

This is a much simpler than the older architecture from 2007:

The older version also could not load servers on the fly, it required an XML file to define everything. It also required a heavy middleware OGSA-DAI, which is good for groups launching many services, but very hefty for those with only one or two.