Delivering Real-Time Insight With the IBM z Analytics Portfolio
Everyone’s perception of “current” is different. Consider this example: If you came to an intersection, would you look at a picture taken a day ago, or even an hour ago, to determine whether it was safe to cross? Of course not. To do so could put you at great risk. Using old data to make business decisions is no different. According to a Forrester study, the more data moves away from where it originates, the more outdated insight becomes. Outdated insight can lead to poor decisions and potentially harmful actions.
Real-time insight matters, and IBM Z* mainframe and z Analytics software deliver the low latency, high-performance environment you need to take advantage of your enterprise data where it originates. This is where IBM z Analytics technologies excel, helping provide answers to questions in real time for better fraud detection and prevention, risk reduction, enhanced customer experiences, and more.
Real time depends on two important aspects. The first is that data must be current. Data that’s moved from platform to platform implicitly has built-in latency. The second aspect is performance. Analytics must provide insight in a timely manner. Having one without the other is not real time. A platform must provide both low latency and high performance to deliver real-time insight.
The IBM z Analytics portfolio is built on the concept of real time. Core to the portfolio are the IBM Db2* Analytics Accelerator and Machine Learning for z/OS*.
IBM Db2 Analytics Accelerator for z/OS
Coupling the IBM Db2 Analytics Accelerator with Db2 for z/OS delivers an industry-leading, data-serving platform. It supports high-performance queries for operational reporting, data warehousing, business intelligence, virtual data lakes and machine learning.
The IBM Db2 Analytics Accelerator transforms the mainframe into a hybrid transaction and analytical processing (HTAP)1, 2 environment that supports transactional and analytics workloads concurrently, efficiently, and cost-effectively. These benefits include:
- High-speed analysis. Gain rapid insight from enterprise data to support time-sensitive decisions.
- Real-time analytics. Leverage business-critical data where it originates to integrate real-time insight with real-time operational decisions.
- Simplification. Simplify infrastructure, reduce off-platform data movement and free up compute and disk resources.
- A secure IBM Z perimeter. Safeguard valuable data under the control and security of Db2 for z/OS.
The IBM Db2 Analytics Accelerator delivers true HTAP, which IBM defines as real-time insight from live transactional mainframe data. True HTAP is a unique, enterprise grade scale-out approach for your real-time analytic needs.
Here’s how it works: As data is written to Db2 for z/OS, the changes are asynchronously replicated to the IBM Db2 Analytics Accelerator. When a new query arrives, the Db2 optimizer decides whether the query should be executed within Db2 or the IBM Db2 Analytics Accelerator. Online transaction processing queries will be executed within Db2. Applicable resource-intensive, complex queries will be sent to the IBM Db2 Analytics Accelerator. If the most recently committed data isn’t required, the IBM Db2 Analytics Accelerator immediately executes the query. If the HTAP option is specified, the IBM Db2 Analytics Accelerator will hold the query, apply the most recent changes up to the point in time that the query was executed and then run the query. A maximum wait time to apply all changes may be specified. With the IBM Db2 Analytics Accelerator, this query-routing process is transparent to the user.
The advantage the Db2 for z/OS HTAP approach has over other HTAP architectures is its heterogeneous scale-out capability. This assures industry-leading performance for mixed workloads. Analytical processing doesn’t degrade transactional workload performance because transactional and analytical queries are each processed within separate compute resource pools that can be upgraded independently.
The IBM Db2 Analytics Accelerator is designed for high-speed analysis of your enterprise data. It drives out cost and complexity and enables real-time analytics on transactional data as it’s generated. This enables you to leverage your business-critical data where it originates and integrate real-time insight with real-time operational decisions.
Machine Learning for z/OS
Machine learning scoring (which is used to make predictions) also requires high performance and the most current data. Scoring services can be called several thousand times per minute. With significant transaction volumes, just a few additional milliseconds can impact revenue by millions to tens of millions of dollars.
With Machine Learning for z/OS, the scoring process can be colocated with transactional applications and data. This minimizes execution time, supporting high throughput and fast response times while delivering consistent service-level agreements. In the most current version of Machine Learning for z/OS, the scoring engine saw a boost in performance, especially when running within a CICS* region. With the scoring services running within the same CICS region, COBOL developers are able to use a link to Liberty calls, which deliver fast performance. IBM internal measurements showed that a round trip call by a CICS transaction to score an average- to large-sized model3 ranged from 1 to 5 milliseconds.
Machine Learning for z/OS offers incredible performance and low latency for scoring, and provides an end-to-end machine learning solution with capabilities that ensure fast model development, deployment, and monitoring. It leverages a hybrid cloud approach to model life cycle management and collaboration. Models can be built and trained on your platform of choice (including IBM Z) and then easily deployed on Machine Learning for z/OS using modern RESTful APIs.
IBM continues to invest in Machine Learning for z/OS across four key areas:
1. Integration
• Leverage IBM Db2 Analytics Accelerator as a data source
• Incorporate Operational Decision Manager to enable rule-based decisions with machine learning insights
• Integrate SPSS* Modeler to prepare, organize and visualize data in addition to selecting machine learning algorithms without requiring significant programming or math experience
2. Compliance
• Support auditable machine learning through traces across the entire model life cycle
• Deliver scoring request policy-based tracing to support regulatory requirements
3. Freedom
• Leverage SPSS modeling and visualization tool
• Support customized transformers and estimators
• Support custom libraries for third party integrated model development environments
• Integrate natural language processing packages
• Expand machine learning libraries to include modeling capabilities, such as XGBoost
4. Simplification
• Use z/OSMF workflow and RESTful APIs to manage scoring services (e.g., start/stop) directly from the administration dashboard UI
• Simplify installation and upgrades
Turn Insight Into Opportunity
IBM z Analytics minimizes the time between when data is generated and a decision is made, helping to close the decision latency gap. The IBM z Analytics portfolio provides high performance and low latency, delivering real-time insight from real-time data.
Learn more about IBM z Analytics at ibm.co/2K6GHky.
1. “Market Guide for HTAP-Enabling In-Memory Computing Technologies,” gtnr.it/2lcGdi4. Retrieved 2017-04-15
2. “Hybrid Transaction/Analytical Processing Will Foster Opportunities for Dramatic Business Innovation,” gtnr.it/1onVMLR. Retrieved 2017-04-15
3. This IBM simulation included a 1GB XGBoost + R model with 2000-plus trees (with a tree depth of 20) and 200-plus input fields