Why Splice Machine?
Adaptive Applications Require a New Data Platform
Modern applications need to be programmed to learn and adapt, rather than be programmed for only a finite set of prescribed actions. Adaptive applications are able to ingest and process large volumes of data in real-time, use AI and ML to learn from the data, and predict the best course of action. This ‘Rearview Mirror’ is critical to support sound decision making, but the “Windshield” is what allows the Adaptive Application to process incoming data and execute the actions required by the in-the-moment situation.
Sample Adaptive Application Scenarios
The need to bring together real-time analytical processing and in-the-moment transactional decision making is common to all industries. Tolerance is rapidly decreasing for the latency involved in ETL processes that move data between transactional and analytical systems. Here are some application scenarios that require concurrent access to up-to-the-moment learning and real-time action:
A prospect enters your eCommerce site. You have 20 milliseconds to make her an offer that matters. Do you know what she was researching this morning?
A patient in the IC has developed acute sepsis. Does anyone know? Is she deteriorating? What caused it?
A truck has broken down. Which stores will experience stock-outs? Should another truck be re-routed to restock them?
There is sharply increased traffic on the network. Is it legitimate or malignant? If malignant, what configuration changes will stop it with the least disruption?
A cat jumps in front of a speeding autonomous vehicle. Should it stop, swerve or keep on driving? What if it is a child or a shopping bag blowing in the wind?
Our predictive maintenance model tells us a key production line needs to be stopped. What is the impact on order fulfillment? What can be rescheduled?
The Need for a Hybrid DBMS
The need to bring together transactional and analytical processing goes back as far as the introduction of MPP Data Warehouse solutions in the 1990’s. The efforts to reduce the latency and complexity of ETL (extract-transform-load) processes to move data between operational and analytical systems has lead to two different types of approaches:
Scale-Up: Engineered hardware/software systems for in-memory computing. These are powerful systems, but often prohibitively expensive.
Scale-Out: Hadoop-based distributed systems. These are cost-effective and very effective for analy, but they are complex to maintain, and delegate many data-management tasks to the application layer.
The best approach to support Adaptive Applications is delivered by a new generation of hybrid database management systems, that integrate the transactional and analytical workloads and make them easy to operate. Analyst firms Gartner and Forrester refer to these systems as HTAP and Translytical, respectively.
The Adaptive Application Platform
To support Adaptive Applications a Hybrid DBMS is foundational, but not enough. Additional requirements are:
1. It needs to support fast data ingestion that can operate concurrently with the transactional and analytical processing driven by other applications on the platform.
2.It needs to support the major frameworks for AI and ML such as Spark MLLib, TensorFlow and H2O.
3. It needs full ANSI-SQL and ACID-transaction support to facilitate application development and concurrent application workloads
4. It needs to be able to scale dynamically, so users can provision just enough infrastructure for their needs
Why Splice Machine
The Splice Machine Application Platform has been designed from the ground up to support the requirements of Adaptive Applications. It offers:
Support OLTP and OLAP. No need to run multiple systems and high-latency, error-prone processes
ANSI SQL / ACID transaction processing to easily migrate existing skills & applications, including PL/SQL code
Ingest fast streaming data, develop, test and deploy machine learning models, collaborate with notebooks
Reduce management, deploy faster, reduce risk. Scale as you grow, provision only what you need.