Meeting Notes 1997-Aug-13

Prof. Lockemann from the University of Karlsruhe and the Forschungszentrum Informatik in Karlsruhe visited us because he was interested in learning what our current and future plans are, to avoid duplication, to coordinate his integration project with TSIMMIS, and to investigate possible cooperations.

Some of the related projects at Karlsruhe use mediators/wrappers in the context of agent architectures. One of the focuses are the investigation of scaling problems in large networks and the automatic generation of mediators and wrappers. At the moment, they would like to use TSIMMIS and concentrate on the automatic specification of mediators. Another aspect is the event management in CORBA environments where they investigate the generation of wrappers for event sources (eg triggers in Oracle DB).

He asked us some prepared questions of which I list some of the more interesting ones.

Wrappers

1. What is the standardised architecture for wrapper generation?
   Ramana explained architecture. See the SIGMOD '97 picture. It is a bit misleading wrt some of the inputs and their relevant internal components.

2. What happens if 2 or more templates at a wrapper match different parts of query. E.g., x :- goal1, goal2 and each goal is matched by a different template?
   Due to our philosophy this case cannot occur at a basic wrapper but is dealt with at the mediator level (which itself can act as a more powerful wrapper if necessary).

3. What is supported regarding templates?
   
   • RHS cannot be non-terminal (unlike '95 paper).
   • General SQL cannot be described.
   • But: We support parameterised templates

4. Does TSIMMIS support active sources?
   We refered to QSS. Lockemann mentioned work by Chakawathe (Florida) and Arne Koschel (Karlsruhe).

5. How are result schema generated?
   Current implementation: by hand based on template. Could be automated. Lockemann expressed some interest in this topic (cooperation?).

6. What are future plans?
   We mentioned among other things
   • Persistency (caching)
   • Templates vs. Schema changes (possible cooperation with Lockemann)

Other topics which were shortly touched: cost models.

Mediator

1. Mediator architecture?
   Ramana explained architecture and stressed similarities and differences to wrappers. Noted that schema is implicitly contained in OEM objects.

2. What is handled by mediator?
   Selection, Projection, Join, Nest, Unnest, Union for conjunctive queries. No negations, no disjoints.

3. What is failure handling?
   At the moment: none. Reference to DISCO project (Paris). Also mentioned use of persistent mediators/wrappers if source fails.

Lockemann shortly mentioned some ideas for improving/changing query plans based on trader information. E.g., traders know about cost or availability of sources.