Where classical business intelligence only supplies information about what happened in the past, Fast Data is different. A Fast Data architecture detects, affects, and acts on situations as they are happening. Event processing is at the heart of the Fast Data architecture. It processes a flow of incoming events to identify situations that need attention—and as soon as identified—responds to bring about the best outcome, which could be risk prevention, cost savings, revenue maximization, or customer satisfaction improvement.
You Need Big Context
As in human decision-making, the more context available for automated decisions, the better the decision will be. Context is built from data—historic, reference, or profile data—which is accessed by decision logic. Because the event processing application needs to respond to incoming information in real time, in-memory computing becomes critical for enabling fast scans through contextual data.
You Need Big Memory
In-memory computing is enabled and scaled in two ways, horizontal and vertical.
Horizontal scaling—using the combined storage of dozens of small machines—works well if you can co-locate data with the logic that uses it. When this co-location isn’t possible, you need cross-node communication to find and access the data, which adds traffic that impacts performance and may limit scalability. If this is the scenario, vertical scalability using large, modern CPU and memory may provide a better option.
Vertical scaling—adding processors and storage to individual nodes—can allow increasing capacity on the fly. As an example, the Bull’s bullion server can be dynamically reconfigured to provide elastic scalability across 2 to 16 processors and 48 GB to 24 TB of memory. This type of solution can be very affordable because you don’t need to invest in huge capacity up front. In many situations, vertical scalability is an easy choice for handling more complex rules and greater volumes of events. It can also simplify application development and management because data and work do not need to be partitioned.
Here’s a technical brief showing real-world scalability benchmarks for the TIBCO event processing platform on bullion machines.
