Formerly known as Wikibon
Close this search box.

Systems of Intelligence Enable Digital Transformation & Differentiation

1. Systems of Intelligence Premise

Systems of Intelligence enable Digital Transformation and Digital Differentiation, which are survival imperatives for all business and governmental enterprises. In addition, Systems of Intelligence integrate existing Systems of Record (i.e., standard ERP systems) with real-time advanced analytics. Advanced hardware and software architectures are required to accelerate existing Systems of Record, and also accelerate/parallelize real-time advanced analytics. This strategy will deliver the fastest route to enabling Digital Transformation and Digital Differentiation.


2. Executive Summary

Digital Transformation and Digital Differentiation

Systems of Intelligence are key to implementing Digital Transformation and Digital Differentiation. Digital Transformation requires basic transformations of institutional and business processes that are “standard”. Digital Differentiation requires advanced transformation of key processes that enable significant and sustainable competitive advantage.

Systems of Intelligence need advanced IT technologies to integrate existing systems of record with real-time analytics. To implement Systems of Intelligence, enterprises need advanced IT technologies that have only become available recently. In addition, these technologies need to operate in concert wherever the data is created and stored, and minimize the movement of data between locations.

The first step in implementing Systems of Intelligence is to enable the current Systems of Record to operate more rapidly. The use of advanced technologies should improve the elapsed time of the core systems of record systems by 50% or more. This will create time (a “latency buffer”) for real-time advanced inference analytics to run.  These real-time inference analytics need  highly parallel systems with specialized processor technology like GPUs. The inference analytic programs are built from large scale analytic systems which are not real-time. One technique for building the inference engine is to develop analytic and artificial intelligence (AI) models, and then use machine learning to train and improve the model. This process can take weeks or months. The output of this development process is a real-time inference program which can be called via an API by the system of record.

2.1 Avoiding Conversions of Systems of Record

Some vendors and professionals argue that the current Systems of Record should be converted to a more modern platform. This may be possible for small companies or very small systems. However, converting large scale Systems of Record will bring years of delay and a very high risk of failure. Wikibon strongly believes that building on the application value of existing systems of record is quicker, less expensive, and less risky. CXOs should reject any Digital Transformation & Differentiation strategy that requires conversion of existing Systems of Record.

The new “UniGrid” type system architectures will improve the performance of almost all Systems of Record. UniGrid architectures are the name given to architectures that offload storage and the storage network components from the central processing unit. It allows direct, very low latency, any-to-any connection between any storage volume and any processor by the use of NVMe over Fabric (NVMe-oF) or equivalent technology to connect the system components. The processor, accelerators (e.g., GPUs), NVMe flash storage, and network components are attached to PCIe, with technologies such as Liqid allowing these PCIe devices to be composable. The number of components in a UniGrid can scale to be in the 1,000s, with true any-to-any interconnectivity.

UniGrid architectures will enable faster execution of real-time analytics and AI – reducing elapsed times sufficiently to allow time for real-time analytics to complete during the execution of the business transaction or process.  UniGrid architectures will also be the basis for any distributed cloud topology (See Wikibon UniGrid Research entitled “UniGrid 1: Digital Differentiation Requires Performance Infrastructure” for more details). This same technology will be the platform that also enables future increased operational automation, reduces security and extended downtime risks, and improves compliance with internal and governmental regulations.

2.2 Hybrid Cloud

The key operational processes of most large enterprises (and the data supporting them) are typically widely distributed across many locations. A distributed multi-cloud strategy is therefore required to enable Digital Transformation and Differentiation across an enterprise. The key imperative is to minimize moving large amounts of “heavy” data, and instead dynamically move the processing to the data. This interconnected multi-cloud strategy is only feasible and effective when there is a common hardware architecture and software stack in all of the clouds. These architected clouds need to be available both on-premises, in cloud vendor locations, and at the Edge. This allows maximum flexibility for placing processing and data dynamically, without having to manage and test different architectures in each cloud.

2.3 The Business Case for Systems of Intelligence

The business case for Systems of Intelligence is far superior to traditional data warehouses and big data applications operating separately from the Systems of Record.  This is because Systems of Intelligence focus on business process automation and reductions in business cycles times in real-time. Initially, the benefit of the Systems of Intelligence approach will be to “make room” for adding real-time analytics. The existing Systems of Record will run faster and at lower cost, with better utilization of compute, storage and network, and fewer software licenses. A more significant business benefit will be the systematic automation of “standard” business processes by providing advanced models and analytics, either from implementation by enterprise IT or from extensions to ISV applications.

The highest value benefit – and most  difficult to realize – will come from developing new ideas for digital differentiation providing sustainable long-term business value. If these ideas can be implemented, the rewards will be overwhelmingly valuable.  Wikibon expects that large enterprises that have established the infrastructure and operational foundation for the first two components (2.1 and 2.2 above) will be in a position to experiment and fail fast on new ideas for fundamental long-term differentiation.

2.4 Bottom Line

There is a strong initial business case for the first step towards Systems of Intelligence, based on faster Systems of Record performance. Over time, Systems of Intelligence will become the foundation for implementing sustainable strategic digital differentiation, with the ability to dramatically improve business innovation cycle-time. The potential benefits for enterprises will range from ensuring survival to overwhelming long-term business value.

2.5 Research Organization

The research and analysis of our premise is organized into three major parts:

  1. Designing Systems of Intelligence
  2. System of Intelligence in an tier-1 Oracle Environment
  3. The Business Case for the first step towards Systems of Intelligence

2.6 Scope of Research

This research focuses on Tier-1 applications requiring very high levels of performance, availability, security and recoverability. It assumes that current systems of record are using a Tier-1 Oracle database implementation. In addition, it assumes each cloud is implemented using hyper-converged principles. Wikibon refers to these hyper-converged clouds – whether on-premises or dedicated – as “True Private Clouds”. Each cloud in a multi-cloud topology is assumed to be identical in hardware architecture and software stack. The reference architecture assumed in this research is Oracle database cloud systems either on-premises (Oracle Cloud at Customer) or in an Oracle Cloud facility.  There are other viable database and system solutions that are available, and Wikibon will publish research on these later in 2018.

Dave Donatelli was interviewed recently on theCUBE addressing some of these issues. The clips are Donatelli discussing his keynote presentation on cloud strategies with John Furrier and Peter Burris.

3.0 Designing Systems of Intelligence

This research report focuses on implementing Systems of Intelligence for large enterprises. Large enterprises have already created business differentiation and have developed processes, procedures, and IT systems to support this differentiation. The challenge for large enterprises is to develop and maintain new digital differentiation. Wikibon believes that Systems of Intelligence built on UniGrid principles are the foundation for the next generation of large enterprise applications.  Part 1 of this research looks at the general principles involved in developing and deploying Systems of Intelligence.

3.1 Systems of Intelligence

Systems of Intelligence are systems of record integrated with real-time advanced business analytics.  In general, the value of non-real-time (batch) analytic systems of record is to make managers and professionals smarter, and enable them to improve the business processes from the top-down. Wikibon research on this top-down approach indicates that CIOs and business leaders find the return on current batch analytic systems ranging from “disappointing” to “moderately successful”. The root cause is less the system itself, but the much lengthier time it takes the organization to take advantage of batch analyses.

3.1.2 Examples of Systems of Intelligence

A System of Intelligence employs real-time analytics embedded in systems of record to help automate manual processes and enable new business capabilities. There are many examples of these new capabilities – real-time price adjustments based on supply and demand, improved service levels, reduced stock levels, real-time fraud detection, and using multiple sensors to recognize employees, customers and “persons of interest” at facilities. These systems respond with the appropriate actions in real-time.  Examples of such workloads might include:

  • Executing a complex fraud detection process in real-time in the submission workflow for (say) insurance claims, and automatically accepting valid claims, while putting aside others for further review.
  • Dynamically changing a call-center script during a customer call, as a result of an analysis of supply chain issues, customer return history, customer mood change, or other reasons.
  • Real-time adjustment of prices and delivery dates to react to demand and supply, customer return profile, weather disruptions, and many other possible factors.

Although these systems are technically challenging, there is the potential for far greater returns on investment from real-time analytics than traditional batch analytic systems.  Outstanding advances in real-time artificial intelligence (AI) are being realized in many workloads – ranging from automated driving to Apple’s iPhone face recognition capability.  These systems will often involve Edge devices operating on edge data from multiple sources in real-time. A case study of a wind farm looks at the costs of different ways of performing Edge computing, and concludes that Edge computing that reduces moving “heavy” data is usually the most cost effective solution.

The specific technical implementation of Systems of Intelligence can vary. The simplest system would be to use the same database for both transactions (usually row-based) and analytics (usually column-based). Low latency is a key design principle for all hardware and software, so the use of low-latency IO and very large memories can reduce the elapsed time for these analytics. When the size of the analytic database is much larger, a different type of analytic database with greater parallelism may be required to improve elapsed times.

3.1.2 Technical Implementation

Figure 1 illustrates a process for enterprises moving from separate systems of record and separate analytic systems into integrated Systems of Intelligence. Today, data warehouses and data lakes provide the basis of analytic systems. These systems typically take data from systems of record, data streaming from the Internet, data aggregators, data from suppliers and partners, and many other sources. This data is transformed (usually ETL – Extract, Transform, Load) using mainly batch processes. These systems are not real-time, and therefore cannot inform the system of record in real-time. Figure 1 shows that the initial stage (Phase 1)  is to start to create batch models of business process optimization, acceleration and automation with a strong focus on real-time automation, e.g., automating pricing updates so they are more real-time.

Phase 2 is to identify a specific process that can be automated, and take a subset of the model in Phase 1 to implement it in real time (red section of Figure 1 below). Typically, minor changes would have to be made to the current system of record to action the input from the real-time analytics. However, the payoff of this incremental approach of continuous improvement is the best way to realize Systems of Intelligence in a controlled, sustained manner.

The section below entitled “The Technical Challenges of Systems of Intelligence” deals with the technical aspects in more detail.

Steps to Systems of Intelligence
Figure 1 – Steps for Creating Systems of Intelligence, which use “UniGrid” architectures to increase the amount of data that can be processed in real-time by orders of magnitude. 
Source: © Wikibon 2018

3.1.3 Digital Transformation & Differentiation with Systems of Intelligence

Digital transformation is the use of data to transform business processes by enabling the use of huge amounts of additional data, from IoT at the Edge to consumer systems in the cloud. The challenge for enterprises is to integrate the digital transformation into Systems of Intelligence.  Besides better infrastructure efficiencies, the key potential value of accomplishing digital transformation is that it can enable digital differentiation for the enterprise.  While 75-85% of typical business and IT processes are “standard” and relatively undifferentiated, the remaining 15-25% processes are critical in that they are used to created specific sustained business differentiation.

3.1.4 Examples of Digital Transformation & Differentiation

The most important and most difficult task is creating advanced digital differentiation for competitive advantage. This requires sufficient volume of product or services to learn from, and long-term access to the data about performance and usage of the product or service over time. The following are just two examples illustrating the value of real-time analytics:

  • Uber’s real-time modification of pricing based on complete information on supply and demand, and actual experience of what happens when prices change.
  • Caixa Bank in Spain offers real-time feedback to all staff who interact with customers on their satisfaction with Caixa. This is a key advantage in maintaining overall customer satisfaction, and ensuring that the staff focus on the right issues and opportunities at the right time.

3.2 The Technical Challenges of Systems of Intelligence

In order for enterprises to digitally transform and differentiate themselves, and create Systems of Intelligence, they must meet the technical challenges of implementing such systems. These systems will process hundreds of times more data than existing systems of record, and require access to data 1,000 times faster. This data will be spread across the enterprise and its suppliers, partners and customers, and many cloud providers.

3.3 Principles for Designing Systems of intelligence

Wikibon believes there are six key principles in successfully tackling this challenge.

3.3.1 Remove Undifferentiated Work from IT Avoid RYO Snowflakes

Figure 2 – Snowflake photographed by Don Komarechka, and included with his permission. Every snowflake is unique.
Source: Don Komarechka 2017. See his website for an inspired collection of unique snowflakes, and the work of Wilson Bentley on the relationship of temperature to the types of snowflake crystals. 

The traditional way to design IT infrastructure is to take the best infrastructure technologies available and stitch them together. A typical installation might include EMC arrays with local synchronous replication and remote asynchronous replication, EMC Data Domain backup system, IBM p-series servers and AIX operating system, VMware virtualization and system management software, BMC management software, CISCO network and switches, and other systems. These are held together with processes and procedures for testing, managing patches, managing updates and implementing enhancements. Automation of the processes comes from internal development implementing and supporting management software, scripts and APIs.

The advantage of the approach has been lower costs of equipment procurement. However, the significant disadvantage is that “snowflake” (see Figure 2) one-off solutions are implemented that create complexity because of the number of moving parts, and the elapsed time to patch, update, enhance, and test the hardware, firmware, and software across the entire installation. IT management, to its detriment (see the Equifax discussion in the next section), often avoids making changes to the system or applying patches quickly because of the risk of downtime. Too Complex to Manage

One recent powerful example of these risks of the failure to apply timely patches was the hacking of personal records of 140 million US citizens at Equifax. Beyond the fact that 140 million people have years of financial risk and uncertainty in front of them, Equifax’s brand has been severely damaged – resulting in a significant reduction of its market capitalization and requiring widespread changes to the Equifax board and senior management.

Wikibon strongly believes that large enterprise systems have become too complex to be managed by even the best manual operational teams. Many examples in the past year have shown long-term outages in well-run operational systems. Apple, AWS, British Airways, Delta, Google, JetBlue, Microsoft Office 365, SalesForce, Southwest, Symantec Cloud, Twitter, United and many others have all suffered significant and costly downtime and attracted negative media attention. The weak links are very complex infrastructure from many suppliers, often resulting in human operational staff making human mistakes. Outsource Undifferentiated Work

As a result, Wikibon strongly believes that infrastructure hardware and software selection, maintenance, and automated operations should be outsourced to vendors with the skills and scale to run infrastructure effectively. An example would be the Systems of Intelligence reference system from Oracle which enables Oracle to provide a fully integrated and tested update every quarter for both on-premises Oracle Cloud at Customer and Oracle Cloud. Wikibon expects greater integration with Oracle SaaS systems, and expects Oracle to develop and deploy real-time operational AI systems, using both real-time and historic data from customers.

3.3.2 Avoid Rewriting Current Systems of Record & Accelerate Development of New Systems

There is a school of thought that current systems of record should be rewritten using modern methods and then moved to a public cloud as a starting point for digital transformation. Any IT professional that has been involved in conversions and rewrites knows the practical problems of changing the engines of an organization while the “plane is still in flight”. In large organizations the process almost always takes years longer than originally estimated. The efficiency of the enterprise is almost always significantly negatively impacted during this time since current systems need to be “frozen” to facilitate conversion, leading to the systems being less productive.

Wikibon strongly recommends that enterprises minimize changes to existing systems of record, as shown in Figure 1 above.

3.3.3 Make Space in Current Systems of Record by Offloading

To create Systems of Intelligence, real-time analytics will be added in parallel to existing systems of record. This requires systems of record to complete faster, in order to allow sufficient time for the analytics to run and guide the operation of the process.

New systems architectures are available (and many more are in the pipeline) that will improve the elapsed time of work units in current systems of record. Flash storage drives, for instance, are replacing much slower magnetic drives, and are already widely deployed. New system “UniGrid” architectures can offload the IO and networking processing and services from the main servers, and release up to 50% of the cycles. These systems are reducing the IO elapsed times from 2-5 milliseconds in traditional storage arrays down to 200 microseconds, a factor of 10 to 25 times faster.

New memory technologies such as ultra-fast flash from Samsung, DRAM/Flash DIMMs, and (perhaps) Intel 3D XPoint are increasing the amount of main (DRAM) memory and enabling in-memory databases. New systems following the Wikibon “UniGrid” architecture will have access to any data from any processor in as little as 50 microseconds over very high-speed low-latency networks, for cluster sizes of 1,000’s of processors. These architectures will allow current systems of record to run much faster, and also allow the parallelism and bandwidth for advanced real-time analytics.

One example of this type of architecture, Oracle’s Exadata and Exalytics on-premises and cloud systems, will be discussed in the second half of this paper to illustrate the features required. There are also other new UniGrid architectures with significant offload and parallelism capabilities, such as the IBM Z-series mainframes, iguazio’s Unified Data Platform, Excelaro, Mellanox, and reference architectures from Micron.

3.3.4 Create a Multi-cloud Infrastructure with Identical Node Architecture, and Move Processing to the Data

Data is created in every part of an enterprise, from every person and machine including IoT system data at the Edge, mobile devices on people and vehicles, multiple cloud systems, and multiple systems at every location.  One school of thought is that all data should be brought to a central cloud for processing. While this is a useful model for some types of systems, Wikibon and other research has shown that the costs of data transport and the elapsed time to transfer data make this approach a complete non-starter for most environments and most large enterprises.

The alternative approach is to provide identical nodes of processing power and move code to the place where data is originally created. After the processing and data reduction, it is then more economic and faster to move a small amount of the original high-value data to other locations (probably <0.1%). This approach is strongly facilitated by a common multi-location, multi-cloud strategy deployed to support Systems of Intelligence. The reference example in this research is the Oracle database cloud services that can run identical systems either on premises (Oracle Cloud at Customer ) or in an Oracle Cloud facility. This is a unique capability for systems based on Oracle Enterprise databases.

Multi-national legislation, such as the European Union’s GDPR coming into effect in May 2018, has also mandated that data must be held and processed within country, with strict rules on retention, and the ability of individuals to remove data.  This is another driver for multi-cloud strategies both within countries and between countries.

3.3.5 Use Parallelism to Reduce the Elapsed Time for Real-time Advanced Analytics & Artificial Intelligence

Wikibon has developed a four-level topology for advanced analytic and artificial intelligence systems includes:

3.6.1 Level 1: Real-time Edge AI Devices

Level 1 devices are usually production devices consisting of highly integrated sensors and processing to solve a business problem with inferences on a large amount of data volume. Examples are the Apple mobile face recognition system on the iPhone X at a manufacturing cost of $122, and Nvidia  DRIVE™ systems to support car and truck manufacturers implementing AI systems at a starting price of $600. Both these devices are based on volume consumer ARM chips and GPUs (See Wikibon research entitled “IoT Devices will Dominate Edge Hardware & Data Creation in 2018 & Beyond”). Most of the data analyzed at the Edge for these use cases will not be stored long term. These devices will usually be managed by the vendors supplying the Edge devices.

3.6.2 Level 2: Real-time Edge Systems

Level 2 systems integrate the smaller amounts of data from Level 1 systems and manage the flow of data to other Edge production systems (for example sending employee arrival data to payroll, and to an application optimizing production as part of a packaged ISV manufacturing system). These systems will usually be managed by the ISV and/or the user’s IT/OT department.

3.6.3 Level 3: Real-time Analytics Bound to Systems of Intelligence

These are production real-time systems that provide the inference results of real-time analytics to the systems of record (e.g., advanced fraud detection analytics for online insurance claims, or real-time pricing analytics adjusted by customer and/or supply and demand). These inputs can automate processes that previously required specific business processes and staff, as well as  operator functions to enter data into the current systems of record.

3.6.4 Level 4: Development Systems for Real-time Analytics

These systems support the development of systems used in Level 1-3 systems above. The models used to develop and train (machine learning,  Bayesian analytics, etc.) real-time analytic development systems are a superset of the real-time models with many more sources of data, and some historic data from the production systems. These systems are batch, have less time-constraints and more cost constraints, and are sporadic in the usage of compute resources. As a result, these systems are often run in cloud systems. The parallelism is often supported by technologies such as Nvidia,  a volume supplier of GPUs, that support volume consumers (e.g., gamers) and enterprise developers  Oracle has announced the availability of Nvidia’s Tesla P100 GPUs on X7 hardware, initially available as bare metal instances on Oracle Cloud infrastructure. The maximum throughput is about 21 TFLOPS of single-precision performance per instance. Wikibon expects this will also become available on Oracle Cloud at customer and the Exadata x7.

3.3.6 Avoid “Snowflakes” and Utilize Volume Economics

Deployment of Systems of Intelligence will exacerbate the problem of IT’s current focus on undifferentiated workloads. For Systems of Intelligence to be viable, Wikibon believes that undifferentiated work on infrastructure, maintenance, optimization, and automated real-time operation has to be moved to vendors, who can take advantage of volume across many customers and systems. This volume aspect enables more rapid system upgrades and security patches, and higher levels of support from vendor staff.

It is clear that the hardware and software in Systems of Intelligence is best integrated by a vendor who can also maintain and update the system. In some cases, the vendor will need to create OEM agreements with suppliers of system components, but the key principle should be to minimize the number of different organizations that touch the system.

Another important advantage of volume approaches to infrastructure is that advanced real-time analytics can be applied to Systems of Intelligence to optimize performance, detect security breaches, balance workloads against business objectives and recover from failure. The most significant source of failure in flying planes, driving vehicles, or operating data centers is the real-time human operator. Removing fine-grained, real-time decisions from operators and enabling them to focus on developing and testing new automated processes will be an important benefit of Systems of Intelligence.

Snowflakes may be unique and beautiful, but for IT they are extremely costly, and will block migration to Systems of Intelligence.

4.0 Systems of Intelligence in an Oracle Database Environment

Part 2 of this research takes the general principles of developing and deploying Systems of Intelligence, and analyzes how these principles can be used in a challenging large-scale Oracle Environment. This type of environment is potentially challenging because of the size, complexity, and advanced functionality already present in the current systems of record and in the analytic support systems.

4.1 Is a Tier-1 Database Required?

Oracle database systems can be expensive. This analysis assumes that the cost of the Oracle databases is currently justified for the business benefits they provide. In our experience with customers, Wikibon has found that Oracle’s business practices make reduction of Oracle license costs difficult to achieve, so this analysis assumes that there are no direct savings from reduced licensing, but rather that the licenses can be used for other purposes.

4.2 Oracle Integrated Systems for Systems of Intelligence

This section takes six major principles for designing Systems of Intelligence discussed in Part 1 of this research, and applies them to an environment where there are currently Oracle-based systems of record, and a significant use of Oracle technology for data warehouse systems. In addition, the capability of implementing multi-cloud systems with the same systems architecture and software stack is evaluated.

4.3 Six Major Principles

4.3.1 Remove Undifferentiated Work from IT

The reference Oracle Cloud strategy is a significant contributor to removing undifferentiated work from large enterprise IT, typically inherited from the traditional “snowflake” method of IT’s stitching together the best individual infrastructure technologies available. Instead, the time-consuming work of setting up processes and procedures for managing patches, updates and enhancements to the infrastructure can be given to the vendor.  This allows IT resources to be redirected to implementing real-time analytic systems to enhance the current systems of record to achieve greater business value.

The advantage of this integrated approach is the reduction in the number of moving parts, and the elapsed time taken to patch, update, enhance, and test the hardware, firmware, and software. This reduces the risk of security breaches, such as the one that struck Equifax.

It also enables new applications and application updates to be deployed more rapidly. When Wikibon interviewed Oracle customers who are implementing Oracle’s cloud strategy, the overwhelming reason for adopting the strategy was that it enabled  faster time-to-value for applications and application updates to the business.

4.3.2 Avoid Rewriting Current Systems of Record & Accelerate Development of New Systems

Oracle has many options for improving the performance, availability, security and recoverability of existing systems based on Oracle Enterprise Edition Database. In addition, Oracle has a broad range of integrated systems (e.g., Exadata, Exalytics, Big Data Appliance) that will support the implementation of advanced analytics, either within the same systems running the system of record, or in highly coupled systems using an integrated InfiniBand network. It is unlikely that current systems of record will need rewriting, and it is likely that their offload techniques (see section below) will allow sufficient room to add real-time analytics.

In addition, Oracle will allow exactly the same Oracle database software to run across multiple locations either on-premises or in the Oracle Cloud, which can be the foundation of an effective multi-cloud strategy.

4.3.3 Move Processing to the Data

Oracle has a number of integrated systems such as Exadata that can be placed in different clouds and on-premises locations.  With Oracle Platinum Service these can all be updated automatically every three months. Oracle also has a range of advanced database solutions such as Oracle Data Guard and Oracle GoldenGate that allow consistent copies of data to be kept in multiple locations. In addition, Oracle has an appliance backup and recovery solution (ZDLRA) that will integrate backup and recovery of Oracle Databases across different locations.

In addition to these products, Oracle also offers Oracle Cloud service. The Cloud at Customer is an on-premises solution that runs exactly the same systems software as the Oracle Cloud service. This allows full delegation of system maintenance to Oracle, which in turn enables Oracle to use volume and experience to improve and better automate Oracle system maintenance over time. Oracle has indicated that there will be a full range of Oracle Cloud services and Cloud at Customer alternatives for all the Oracle integrated products.

Oracle also has a range of Oracle Data Appliance (ODA) systems that are suitable for applications in remote locations and at the Edge. The ODAs have good mechanisms for lowering the license costs for Edge systems where appropriate.

4.3.4 Make Space in Current Systems of Record by Offloading

Oracle’s Exadata X7 offers significant offload capabilities starting with IO processing from the application servers to the storage subsystem. New Exadata software also delivers in-memory performance from shared storage for OLTP workloads.  Exadata X7 software can automatically uses DRAM as a storage server memory cache to speed-up access to data with algorithms that avoid caching the same data in the database server cache and storage server memory caches.  In addition, the IO latency on the Exadata X7 storage server has been reduced to 250 microseconds, further enabling offloading.

Oracle also offers Smart Scan, which offloads some data intensive SQL operations from the database servers to the storage servers. SQL data filtering is processed in parallel across all the storage servers as data is read from flash, and directly relevant data for a query is sent to the database servers.

4.3.5 Use Parallelism and New Architectures to Reduce the Elapsed Time for Advanced Analytics

Oracle Exalytics, Big Data Appliance, and Oracle Database in-memory extensions on Exadata both support in-memory analytics, an important technique for real-time analytics. The Exadata storage server can offload compression using Oracle’s Hybrid Columnar Compression (HBC) feature. This can speed up processing in intensive read environments, and helps to enable some real-time analytics. This technology utilizes a combination of both row and columnar methods for storing data, and can enable the use of databases for both system of record (mainly row-oriented) together with analytics (mainly column-oriented).  This hybrid approach achieves the compression benefits of columnar storage for real-time analytics, while reducing the performance shortfalls of a pure columnar format for systems of record. Wikibon advises users to focus on the performance speedup of HBC, rather than the storage savings.

At the moment, Oracle does not support integrated special processors such as GPUs (Graphical Processing Units, e.g., Nvidia cards discussed above). Executives should push strongly for commitments from Oracle to provide GPU and other integration capabilities. 

4.3.6 Avoid “Snowflakes” and Utilize Volume Economics

The most important feature of Oracle’s cloud strategy is that it allows enterprise IT to avoid or minimize IT snowflakes for existing systems of record, and more importantly ensures that snowflake approaches do not get in the way of deploying Systems of Intelligence. Wikibon believes that the reference Oracle integrated private and public cloud solutions will enable it to deliver database infrastructure maintenance and optimization more quickly to its customers than “snowflake” solutions. This should enable much quicker upgrading of systems, more rapid application of security patches, and better levels of crisis management from support staff.

Wikibon also believes that Oracle will deliver advanced real-time analytics that can be applied to Systems of Intelligence to optimize performance, balance workloads against business objectives, and recover from failure. This will help remove real-time decisions from operators and enable them to focus on the strategic decisions in managing Systems of Intelligence environments.

Oracle customers wanting to implement this reference strategy should ensure Oracle is committing the necessary resources and priority to this program, and that the timescale for delivery is well inside the timescales required by enterprise IT and the business.

5.0 The Business Case  for Systems of Intelligence

This research focuses on the IT business case for the first stage of moving to Systems of Intelligence. This requires speeding up the elapsed time for current Systems of Record by 50% or more. In addition, this speed-up “makes room” within current budgets to enable further investment, particularly for analytics and AI software. Future Wikibon research will take a broader view of the total business case for moving to Systems of Intelligence.

5.1 The IT Business Case for Enabling Systems of Intelligence

Figure 3 below shows the business case for lowering latency and “making room” in existing Systems of Record. Offloading the IO load and reducing the IO latency with “UniGrid” systems reduces latency and reduces CPU cycles. These CPU savings are reflected in lower Oracle Database costs, lower CPU costs, and reduced environmental and operational administration. In addition, there are significant savings in database and systems administration costs. There are some additional costs in the IO subsystem required to achieve the overall savings, but these are easily offset by the infrastructure savings elsewhere.

There are other features in the Exadata X7 that help with offloading IO work. These are discussed in Oracle’s literature on the Exadata X7.

Our recent discussions with Oracle users indicate that Oracle business practices still make it difficult to achieve database license savings.  However, most felt that there was a good chance that the licenses saved could be applied to future analytic licenses.  In our financial analysis below, we are assuming that the savings can be used to deploy advanced Oracle features for implementing real-time analytic systems.

Business Case for Offloading
Figure 3 – The Business Case for Lowering Execution Latency of Existing Systems of Record by Offloading IO and Reducing IO Latency
Source: © Wikibon 2017

Most Oracle users we interviewed agreed that the savings on operational costs would be 30% or more by moving the responsibility of updating Oracle hardware and software to Oracle.  They also agreed that the most important benefit of a strategy of outsourcing non-differentiated IT work to Oracle was a faster time to value of new initiatives.

Reducing transaction times and lowering licensing costs could deliver business savings from increased productivity of operators in the line of business, and increased customer satisfaction. Strategically, however, the returns on using the “room” created by improving the Systems of Record can be used to introduce real-time advanced analytics and artificial intelligence systems.  The next section gives an example of the potential savings of True Systems of Intelligence. Future Wikibon research will look further into this.

5.1 The Business Case for Systems of Intelligence and Real-time Re-pricing

The major strategic business advantage of using “UniGrid”  systems to lower elapsed times in Systems of Record is to make room for real-time Digital Differentiation. One example is dynamic real-time pricing of goods and services. Figure 4 below is an illustration comparing the traditional sequential multi-step business process with a real-time single-step real-time decision. Practical experience is that after an initial period of doubt and concern about the automation (the user says he must personally check every machine decision), the user become rapidly confident with the automation and will accept without review.

Figure 4: Impact of Continuous and Individual Re-pricing
 Source: © Wikibon 2018

An early example of this is the dynamic pricing of taxis or ride sharing, first introduced by Uber. If there is an imbalance in the supply of taxis, and prices remain fixed, wait-time for taxis increase, and user frustration increases.  What Uber found is that raising prices reduces demand and increases supply, and service levels remain reasonable. And, of course, Uber and their drivers increase their revenues.

The same principle could be applied by cities to all vehicle drivers. Reducing demand by increasing the cost of entering a city would improve congestion and increase productivity. The utilization of the road systems is increased by moving demand to less busy times. The same principle can be applied to bridge and road tolls.

Another example is individualizing prices according to demand and the ease of doing business with a customer. If a customer has a high chance of returning goods or a high probability of making making unfavorable social media comments, it would make sense to reduce discounts given to these customers. And of course one could optimize the supply chain by dynamic price changes to (say) the color of an item.

All of these examples require access to large amounts of information on all aspects of supply and demand, sophisticated models that analyze this data in real-time, and the ability to apply those changes to the Systems of Record. The business benefit is Digital Differentiation, which at best provides sustainable differentiation, and at worst ensures competitiveness.

The business cases for these Systems of Intelligence will be investigated in detail in future Wikibon research.

6 Conclusions

The scope of this research is limited to large enterprises which currently have strong and ongoing investments in Oracle Database products. Wikibon will be investigating other approaches to implementing UniGrid and Systems of Intelligence in future research.

Wikibon believes that it is possible to add real-time analytics to existing systems of record, and create Systems of Intelligence in a demanding Oracle environment. In addition, Oracle’s cloud strategy offers a unique capability of developing a multi-cloud strategy with the same system and software architecture in every individual cloud.

Over and above the compelling infrastructure benefits, our discussions with executives adopting these sorts of offerings indicate that the main strategic reason for adoption is to enable faster time-to-value for new applications and enhancements to existing applications. Wikibon will be actively researching the potential business value of fully implementing Systems of Intelligence in future research in 2018.


Wikibon strongly recommends that CTOs, CIOs and senior business executives should define a Systems of Intelligence strategy and migration plan and utilize the six principles for developing and deploying Systems of Intelligence described in this research guide. Wikibon also strongly recommends that any Systems of Intelligence project focus on tangible Digital Differentiation through automation, and avoid the trap of thinking that just making managers smarter about current processes with additional reports can create Digital Differentiation.

Book A Briefing

Fill out the form , and our team will be in touch shortly.
Skip to content