Jul
3
Written by:
Event Zero Administrator
7/3/2008 4:04 PM
We moved to a new venue today with functioning air conditioning! This building is a 14th century building that used to serve as the cloister for the church next door (oddly enough the other building was only a few years old). The format of the meeting has changed from concurrent tracks to a single program that everyone attends. VERY long day with 11 presentations grouped into 4 sessions. Given the volume of the presentations, I’m going to provide some key points from the sessions. In general this day covered a much wider range of topics both from academia (mostly) and industry.
Session 1: Content Based Pub/Sub
There were two academic presentations in this track related to content-based networking presented by researchers from Trinity College Dublin and Helsinki University of technology. These presentations covered the pros and cons of content distribution networks, the general landscape, and alternative approaches. In general this subject is not extremely relevant for the types of applications that we are looking to create at the moment, but is something to consider moving forward. In general, most enterprise EP applications only require pub/sub frameworks that will be used by a very few subscribers and the EPN is in charge of propagating relevant derived events out to clients. However, a few topics of interest did come up. The first presentation covered extensions of the Sienna CBN to create a Knowledge based network. The extensions extended expressions of interest in content from a single value to a “bag” of values (and associated set operators). There were a few slides discussing ontology and ontological relationships – sounds like a good method/tool to describe and infer relationships. This is a requirement we have seen in the past related to the ability to model devices and their relationships to interpret independent alarm events as related to a larger group (ex. An intersection with multiple traffic lights or interrelated network components). Interesting that ontology was mentioned for this purpose in an IBM case study slide later in the day(see below).
Session 2: Filtering and synchronization
3 presentations in this track. The first presentation considered a distributed system in a low bandwidth environment where high volumes of event streams are sent to numerous monitoring applications. Real world case would be the new generation of wireless sensor devices (this topic was mentioned in numerous case studies and presentations). The general approach combined collection of events from the low bandwidth network, regional multicasting (rebroadcast to a segment on a higher bandwith network) , and ultimately content based filtering on the multicast clients. The next presentation discussed a technique (bloom filter based routing) to optimize content based publish/subscribe. Generally this technique improves the efficiency of the publish/subscribe environment by eliminating the need to re-evaluate content at multiple nodes in the overall system. The last presentation discussed optimized techniques for peer to peer event notification in the context of multiplayer online games. The technique optimized traffic on the overlay network (used to exchange player information) by optimizing the ability to detect and delete obsolete events via the introduction of a “gossip” protocol.
Session 3: Event processing in sensor networks and runtime environments
The last track wasn’t extremely relevant, but this track was really good. IBM did a presentation that described the “Intelligent oil field - a really good example of using event processing to provide operational intelligence and improve the bottom line. The scenario: many geographic locations, many types of devices, and many different interests in the associated event data. The main use case involved multiple assigned production engineers each monitoring their own set of assets (and being able to model the event sequences for their domain expertise). They described a 4 tier system (unlike the usual presentation, app, db 3 tier) as follows: 1 – user, 2 – early event warning (EEW), 3-sensor publisher tier (OSI,etc.), and 4 - ontology layer to manage relationships/hierarchy of devices, labels, assets, users, and sensors. They presented some stress test analysis of 50 assets w/ 6 data channels each (300 streams) . One of scenarios described was sand production and acoustic detectors to monitor the amount of sand being sucked into the processing equipment. The amount fluctuates normally and patterns are detected to stop the machinery before serious damage occurs. This is a perfect example of where simple threshold monitoring (typical in most SCADA environments) is not adequate. Rather, detection of patterns is the key to keeping production moving and sparing VERY expensive equipment. Challenges mentioned were working with other vendors and learning the industry.
Session 4: Reasoning and complex event detection
Two interesting presentations in this group. The first was a presentation by Franz Inc. They showed some pretty amazing questions you could ask AlleroGraph DB. Really cool for data mining, but not real time event detection – they spoke mostly about how fast they could load the database. The interesting thing is that their database schema consists of one table - 3 columns (really 4 for a unique rowid) where everything is indexed. Using the concept of triples (partition/key, name, and value) they can answer all kinds of cool questions with an infinitely extensible schema (by just adding more predicates – new rows) on the fly (ex. Find all meetings that happened in December within 5 miles of Berkeley that was attended by the most important person in Jans’ friends and friends of friends) . The other interesting presentation in this lot was by SAP. They are working on event processing using Rete Algorithm for pattern matching. They added extensions to handle temporal relationships to handle Allen’s temporal relationships (generally describes every combination of intervals that exist).
Tags: