Research Agenda And Road Map

What are the interesting problems and questions in event processing research?

Guiding exemplary questions:

Explore the synergy with delay tolerant networks (DTNs)
- Unification - pub/sub is actually an overlay structure on delay-tolerant networks; need to explore the synergy of pub/sub and DTNs when modeled and implemented in this way
Other synergies?

One-size-fits-all usually fits nothing particularly well; in addition to sharing events across systems, we need to further explore how a PS system designed for constrained environments (e.g. sensor networks) can federate with PS systems designed that do not suffer from similar constraints.

Attendees: Joe S., Peter P., Jian L., Alex W., Vinod M.

Streams have a temporal or type relationship among them.
Pub/sub is a datagram model. Each event is self-contained with no imposed relationships.
How to introduce notions of application level flows to event-based systems?
- Based on event type/content, {source, sink} pairs, etc.
  - Complicated by anonymity and many-many communication.
- An application "flow" may consist of many network-level flows.
- How to express, configure, manage these flows?
- Who specifies the flow (senders, receivers, 3rd party, ...)?
- What can we do with the flows? QoS, resource reservation, monitoring, etc.

Joe: Let's build it from scratch instead of adding layers to existing architectures/models.

Real time (QoS in general)
- Was never built into IP. A mistake.
- Is QoS in event-based systems {equivalent, subset, superset, etc.} of QoS in networks.
- Anonymity in pub/sub adds new QoS issues.
- Many-many communication adds new QoS issues.
- What are the real time requirements in event-based systems?
- Who specifies the requirements (sender, receiver)?
- Is pub/sub appropriate for real-time applications? What are the applications?
Mobility semantics (don't hide mobility from the application)
- Don't bother sending a messing if only <50% of devices will receive it.
- Ensure message is delivered to N of M subscribers.
Reliabiliions that need anonymity.
- applications where non-overlapping groups communicate.

Wireless networks break many existing protocols.
- Sperical propagation model breaks routing protocols based on fixed links.
- Conceivable that most devices will be wireless in the future.
Many, cheap devices
- Achieve scalability, relability, redundancy without expensive replication
- Clustered architectures in unstructured, dynamic environments.
- Cost is a factor.
- Earlier group communication research ignored cost.

Layer on top of existing infrastructures
Virtual infrastructures (Genie, etc.)
Clouds of network with their own protocols with gateways (with custody transfers) to bridge them.
Migration path from old to new infrastructure.

ty (delivery semantics)

Crucial for real systems.
Best effort not good enough.
100% is too hard.
Applications must be able to specify different requirements
- How to express this?

Pub/sub is most appropriate for:
- applications that need anonymity.
- applications where non-overlapping groups communicate.

Wireless networks break many existing protocols.
- Sperical propagation model breaks routing protocols based on fixed links.
- Conceivable that most devices will be wireless in the future.
Many, cheap devices
- Achieve scalability, relability, redundancy without expensive replication
- Clustered architectures in unstructured, dynamic environments.
- Cost is a factor.
- Earlier group communication research ignored cost.

Layer on top of existing infrastructures
Virtual infrastructures (Genie, etc.)
Clouds of network with their own protocols with gateways (with custody transfers) to bridge them.
Migration path from old to new infrastructure.