Opher Etzion has blogged a good summary of one of the more “interesting” sessions at Dagstuhl, a.k.a. the “industry feedback” evening session. My list of the “most interesting points raised” includes:
Marco Sierio from Sweden talked about his ASP/SaaS model for delivering event processing – proof if needed that event processing applications can be deployed as services or in the cloud. His RuleCore application does track n trace for trans-Europe shipments, with customer rules sent as events directly from his customer to the system.
- The “rule updates are just events” philosophy makes perfect sense in my mind, albeit one requiring considerably more event validation / security / etc etc than the simple sensor events that drive the application. A similar approach is used to update the template-defined rules in TIBCO Active Service Gateway, for example.
- Marco mentioned one side effect of the application being that the shipping company discovered truckers were apt to make stops outside – and therefore effectively associating their ad-emblazoned trailers with- er, “establishments they would not normally want to be associated with”. Real life meets monitored life?
- Some of the “issues” in learning to “do CEP” he finds in customers are
- understanding the declarative (rather than procedural) model, and
- understanding asynchronous events (rather than “request-reply”, a.k.a. simple SOA).
- Richard Tibbetts talked about some of his “observations” about developer practices in building CEP systems from his customers at Streambase – and applicable to all CEP development and customers, for sure.
- Lack of discipline in MOM message design: developers create too many message types and replicate information inside messages. I have heard this before, and continue to wonder whether there is now a need for “Master Event Management” …
- A tendency towards premature optimisation: some EP operations are simply not used enough to warrant much effort in optimising them. Use a profiler to see what needs optimising *after* the system is running …
- Use incremental development – don’t try and replicate the “big bang” and wait until functional completion before unit-testing!
- Event patterns tend NOT to be favored by developers, compared to lower level developer constructs like queries.
- Badrish Chandramouli from Microsoft explained the genesis of the Microsoft stream processing engine (MSSI) from its routes in the CEDR research project in Microsoft Research.
- MSSI is used internally for web analytics on the Microsoft Hotmail system, doing user classification for the on-line ad system at a rate of 1M events per day (i.e. probably 20-50 per sec), as a part of a system that uses map reduce as a part of the machine learning.
- The MSSI product is validated internally by translating sample continuous queries into SQLServer queries via some temporal algebra translator, and event streams into database tables; if the outputs from SQLServer and MSSI don’t match they know they have a problem!
- Although MSSI is brand new to the market, they are already looking at High Availability features for a future release, and topics like real-time data mining…
- IBM covered 2 of their (at least 3, maybe 4) CEP technologies:
- Martin Hirzel from the IBM System S (a.k.a. Infosphere Streams) research group talked about their progression from research tool to product, and the development of their Streams Processing Language.
- Udo Pletat talked about using the IBM RFID tool that exploits the IBM Labs’ Amit ECA rule engine. I hadn’t realised Amit had gone into real-world use, so will need to do more updates on the CEP Market time-graph!
