Within the computing industry, the traditional High Productivity Computing (tHPC) market has acted and continues to act as a generative edge for new technologies and applications. This market area has traditionally been the point where users are pushing advances in system performance and architectures to address problems that range from standard engineering simulations to problems that have hitherto been intractable. In addition, tHPC has acted as a test bed for the IT industry as a whole, where new concepts and technologies are developed and proven and then introduced into larger mainstream markets.
Tabor Research believes that new technologies, methodologies and applications are emerging outside of the traditional HPC market that have the essential characteristics of high productivity computing (requirements for leading edge capabilities, incorporating, testing, and perfecting of new technologies and methodologies, and market creation and expansion.) This new area, which we cleverly call Edge HPC (or eHPC), leverages the experience and technology of the Traditional HPC market, while introducing new areas for innovation. Most importantly we believe that eHPC is at the cusp of significant market generation and growth.
eHPC Drivers
Factors driving the Edge HPC market include:
- Data source explosion – Over the last decade several technologies for acquiring and entering data directly into computer networks have become ubiquitous. For example, internet access allows anyone with a PC to become a data entry source; barcode scanners can provide detailed information on product sales and movement, RFID provides the capability to automatically gather data on the location and movement of anything with a tag; GPS systems actively generate location data which can be stored and transferred, and so on. This list is being expanded as new technologies come on line.
- Processing new data types – The information technology industry has been built around the processing of four basic data types – integer number, floating point number, strings, and logical values. In the final analysis, the entire industry and the trillions of dollars of revenue it has generated is based on our ability to manipulate these four abstracts. Over the last few decades new media data types (e.g., images, audio streams, video streams) have entered and helped to propel the market. However with the exception of some filtering and “photo shop” type processing on images, these new data types have largely been treated as basic large objects (BLOBs) which are only read and written with other data types linked to the BLOB for processing purposes, e.g., searching, sorting, etc. As additional processing techniques continue to evolve we expect that the expanded ability to manipulate these data types will lead to new applications and green field market opportunities.
- Merging of communications and computing – Computing and communications have been connected since the development of the first I/O device. Over time the balance between computational and communications functions associated with computers has steadily shifted towards the communications side. This shift saw major inflection points during the introduction of the World Wide Web and the move to high-bandwidth internet services. This shift combined with continuing increases in the performance and ubiquity of computing/communications networks has created classes of applications ranging from virtualized economies (e.g., eBay), to new social structures (e.g., Facebook, MySpace), to new forms of mass entertainment (i.e. on-line gaming), and so on.
- Complex event architectures – New data sources and faster data communications have combined to enable a new generation of complex event driven applications. These applications accept streaming data from multiple independent sources which are driven by real world events. Data can be generated at irregular intervals, with varying volumes and from geographically dispersed sources. Applications must respond to the incoming data in near-real time. Responses can range from recording data in step with streaming rates, to evaluating system conditions and issuing warning reports and/or system control commands, to recalibrating/reanalyzing external system conditions. In addition to the communications based applications mentioned above, near-real time applications can take the form of in-flight product distributions and logistics, financial tracking, analysis, and trading, national security threat analysis and warning systems, and so on.
None of these factors are new or have gone unnoticed. However, they have combined to create new sets of computational/data and visualization requirements that can be addressed by high productivity computing technologies. Tabor Research believes that a significant “Edge HPC” market currently exists and that this market has strong growth potential over the next five plus years.
High Productivity Computing Definition
Tabor Research defines HPC as the use of servers, clusters, supercomputers, and networked systems – plus associated software, tools, components, storage, and services – for tasks that are particularly intensive in computation, analysis, memory usage, or data management. Within industry, HPC can frequently be distinguished from general business computing in that companies generally will use HPC applications to gain advantage in their core endeavors – e.g., finding oil, designing automobile parts, or protecting clients’ investments – as opposed to non-core endeavors, such as payroll management or resource planning.
At the highest level, Tabor Research divides the HPC market into “Traditional HPC” and “Edge HPC” segments, as follows:
- Traditional HPC – HPC systems used by scientists and engineers in research, development, and production across industry, government, and academia.
- Edge HPC – HPC systems used to address applications that have essential “high productivity computing” characteristics but which lie outside the traditional bounds of scientific and engineering computing. Characteristics of this class of applications include (but are not limited to):
o requirements for leading edge systems performance, or ability to address the most demanding problems.
o requirements for ultra or extreme levels of scalability.
o tendency to incorporate, test, and perfect new technologies and methodologies.
o associated with market creation and expansion.
Classifying Edge HPC Applications
The eHPC market represents a diverse set of users with application requirements for high productivity solutions. These requirements can range from relatively straightforward extensions of traditional HPC applications or workflows into new fields to more abstract requirements for system architectural innovations and/or highly specialization systems configurations or infrastructures. Given this diversity of top level requirements, we believe that the market is best segmented based on the physical and/or logical features that define and drive the applications.
Tabor thus divides the Edge HPC market into four major segments: Complex Event and Business Processing, Process Optimization, Virtual Infrastructure and Environments, and Ultra-scale Computing.
Complex Event Processing
Complex event processing (CEP) applications are driven by continuous data feeds generated by real world events such as: electronic trading on stock markets, security monitoring systems, sensor based inventory tracking systems, and so on. Data may be streamed into the system from multiple independent sources, and data may dramatically decrease in value over time. Data volumes can vary significantly from moment to moment. CEP solutions often involve networks of: sensors, multiple communicating servers, and control devices. Applications operate in near real-time, with events initiated from real world occurrence often setting off a chain of response and control events throughout the system network.
CEP applications fall into the eHPC realm when:
- Requirements to maintain low response latencies in the face of growing data volumes exceed the capabilities of standard systems.
- eHPC system capabilities allow for the expansion of CEP based applications into new market domains.
- System scale requirements exceed the capabilities of standard systems.
Tables 1 provides examples of CEP domains and applications.
Table 1
Examples of Complex Event Processing Applications
Domain | Applications Type |
Civil Infrastructure/Utilities | Delivery network monitoring |
Pipeline monitoring | |
Computer Systems | Intrusion detection |
Network monitoring | |
Financial Services | Event alert |
Market pricing | |
Digital trading | |
General Business | Environmental monitoring |
In store monitoring | |
Real time supply chain | |
Health Informatics | Disease tracking |
Patient monitoring | |
Military Operations | Battlefield monitoring |
Shared battle space awareness | |
Target detection | |
National/Civil Security | Environmental monitoring |
Signal intelligence | |
Area surveillance | |
Telecom | Network traffic routing |
Transportation | “In-flight” asset tracking |
Process Optimization
The Process Optimization (PO) application profiles mirror traditional HPC workflows. These applications make use of technology above and beyond standard enterprise solutions, either in architecture, software, or system management. PO applications have one or more of the following properties:
- Data intensive – The majority of the applications’ time may be spent streaming data from disk to memory to processor and back to disk for database centric applications, or from data source to processor to disk for event centric applications. In addition, both data transfers and total database volumes may be significantly larger that those found in common business applications.
- Computationally intensive – The majority of the applications’ time may be spent performing arithmetic or logical operations in such areas as: statistical analysis and modeling, process simulations, pattern matching, and so on.
- Operational time processing – Applications are operating in synchronization with the operational timing of the organization.
Table 2 presents a list of example applications and domains that we see as fitting into the Process Optimization segment at this time.
Table 2
Examples of Process Optimization Applications
Domain | Applications Type |
Military Operations | “Sense and Respond” logistics |
Business Intelligence | Data mining, database search |
Civil Infrastructure/Utilities | Anomaly management |
Computer Systems | Anomaly management |
Financial Services | Capital budgeting |
Combinatorial auctions | |
Derivatives pricing | |
Legal compliance | |
General Business | Distribution resource planning |
Facility location planning | |
Resource scheduling | |
Vehicle routing | |
Military Operations | Asset tracking |
Distribution resource planning | |
Facility location planning | |
Spares management | |
National/Civil Security | Seismic activity monitoring |
Weapons and delivery systems planning | |
Telecom | Anomaly management |
Transportation | Real time route planning/rerouting |
Other | Complex text and image matching |
Text classification and filtering |
Virtual Infrastructure and Environments
Virtual Infrastructure and Environments (VIE) applications implement computer network based business and social structures. They also hold the promise of extending these structures through synthetic realities ala Second Life . These structures range from on-line gaming environments, to multi-person/system training environments, to virtual economies, to virtual social environments. The applications fall into the eHPC market based on:
- Requirements for real time or near real time response rates.
- High productivity computing requirements to address continually expanding message rates and data volumes.
- The ability to expand the VIE application domain based on advances in computer system capabilities.
Table 3 provides a list of example VIE domains and applications.
Table 3
Examples of Infrastructure and Environments Applications
Domain | Applications Type |
Virtual Civil Infrastructure | Internet commerce |
Consumer products | On-line gaming |
Social networks | |
B to B and B to C | Virtual economies |
B to B | Virtual offices |
Ultra-Scale
One eHPC feature that appears across multiple application spaces is the requirement for “Ultra-scale Computing capabilities.” Ultra-scale computing systems are specially designed and/or configured to effectively manage node counts that significantly exceed those provided by industry standard products.
Currently ultra-scale applications generally appear as service layers to the internet. The primary example of this application is internet search engines. Applications can be both data intensive (e.g., map and satellite photo applications) and/or compute intensive (e.g., search applications). This segment is currently represented by a small number of very large sites.
Table 4 provides a list of example Ultra-scale domains and applications.
Table 4
Examples of Ultra-scale Applications
Domain | Applications Type |
Internet data processing | Data aggregation |
Data dissemination | |
Search | |
Other | Other |
Conclusions
Tabor Research believes that over the last few years a number of technology and market factors have combined to create new market opportunities outside the boundaries of the traditional HPC market. We believe this “Edge HPC” market is currently generating significant revenues and has strong growth potential. Over time, we expect it to exceed the tHPC market due to the scope of domains it will impact.