Using an In-Memory Data Grid for Near Real-Time Data Analysis

By Nicole Hemsoth

August 6, 2012

by Dr. William Bain, ScaleOut Software, Inc.


In today’s competitive world, businesses need to make fast decisions to respond to changing market conditions and to maintain a competitive edge. The explosion of data that must be analyzed to find trends or hidden insights intensifies this challenge. Both the private and public sectors are turning to parallel computing techniques, such as “map/reduce” to quickly sift through large data volumes.

In some cases, it is practical to analyze huge sets of historical, disk-based data over the course of minutes or hours using batch processing platforms such as Hadoop. For example, risk modeling to optimize the handling of insurance claims potentially needs to analyze billions of records and tens of terabytes of data. However, many applications need to continuously analyze relatively small but fast-changing data sets measured in the hundreds of gigabytes and reaching into terabytes.  Examples include clickstream data to optimize online promotions, stock trading data to implement trading strategies, machine log data to tune manufacturing processes, smart grid data, and many more.

Over the last several years, in-memory data grids (IMDGs) have proven their value in storing fast-changing application data and scaling application performance.  More recently, IMDGs have integrated map/reduce analytics into the grid to achieve powerful, easy-to-use analysis and enable near real-time decision making. For example, the following diagram illustrates an IMDG used to store and analyze incoming streams of market and news data to help generate alerts and strategies for optimizing financial operations. This article explains how using an IMDG with integrated map/reduce capabilities can simplify data analysis and provide important competitive advantages.

Real-Time Analytics Engine 


What is an In-Memory Data Grid?

By storing fast-changing data within a middleware software tier, IMDGs enable applications to seamlessly scale performance by adding servers that access and update a shared, memory-based data set.  To maximize scalability, IMDGs automatically load-balance data across servers on which the grid is hosted. They also redundantly store data on multiple servers to ensure high availability in case a server or network link fails. Additional capabilities, including eventing and distributed locking, make IMDGs a powerful data storage platform.

IMDGs typically integrate their data storage model with object-oriented programming languages, such as Java and C#. They store data as a collection of objects which are accessible either by specifying an identifying key or by querying object properties. The IMDG’s built-in parallel query mechanism can quickly scan a large data set for objects whose properties match a query specification. This provides an important tool for identifying data to be reviewed or analyzed. The following diagram illustrates the use of parallel query for selecting stock history data.

In Memory Data Grid 

Using an IMDG for Analytics

Without a doubt, the field of data analytics has gained a powerful new tool with the “map/reduce” analysis model, which has recently surged in popularity as open source solutions such as Hadoop have raised awareness. In fact, the roots of the map/reduce pattern date back to pioneering work in the 1980s which originally demonstrated the power of data-parallel computing.

Map/reduce implementations take many forms and are offered as components in several competing frameworks. Nearly all of these solutions are aimed at accelerating data analysis for disk-based data. With some data sets reaching petabytes in size, the benefits are often measured in reducing batch job processing times from hours to minutes for these “big data” analyses.

However, the overhead (and complexity) of disk-based map/reduce platforms is too high for applications which must quickly analyze fast-changing data sets measured in hundreds of gigabytes or terabytes. (Estimates by some analysts indicate that as much as sixty percent of data sets are smaller than ten terabytes.) In many situations, an answer in hours or minutes is not acceptable.  For example, an e-commerce Web site may need to monitor online shopping carts to see which products are selling. A financial services company might need to hone its equity trading strategy as it optimizes its response to fast-changing market conditions.

To address this challenge, leading-edge IMDGs have incorporated map/reduce analytics engines, transforming them from just scalable, memory-based data stores into parallel computing platforms for analyzing data and providing fast, near real-time results. IMDGs leverage the grid’s automatic load-balancing to minimize data motion and speed up analysis. Instead of migrating data into memory from disk, an IMDG analyzes data in place. Results also are stored and combined in memory, minimizing file I/O to calculate the final results. By eliminating these overheads, IMDGs dramatically reduce network usage and thereby shorten analysis time.

Moreover, by simplifying the programming model, IMDGs offer another advantage over popular, disk-based map/reduce platforms. Instead of requiring the application developer to create a key space for identifying objects to be analyzed, they make use of object-oriented query specifications to select objects.  Also, both the analysis (“map”) and merge (“reduce”) codes can be structured as straightforward, object-oriented methods written as if to be executed on a single workstation. These capabilities shorten design time and enable analysis applications to be quickly developed and revised.

The following diagram illustrates a map/reduce analysis of stock trading strategies across a set of stock histories held in the IMDG. A parallel query selects stocks for analysis, and the IMDG analyzes the stocks and merges the results using the supplied methods:

 Running Map/Reduce on an IMDG 

Running Map/Reduce on an IMDG

ScaleOut Grid Computing Edition (GCE) from ScaleOut Software is an example of an IMDG with an integrated data analytics engine. Using it as an example, the following steps demonstrate how an IMDG performs a map/reduce data analysis:

  • The data set to be analyzed in the IMDG originates from one of two sources. In many cases, especially those with tight latency requirements, the application continuously updates the grid as data flows through for processing. Alternatively, the application may stage the data set in the grid from persistent storage via a bulk loading operation. In either case, the IMDG holds the data, creates replicas for high availability, and load-balances it across servers to avoid hot spots.
  • ScaleOut GCE allows a query specification to be written either in Java using filter methods or in C# using the Microsoft language integrated query (LINQ) mechanism. This query specification selects the data to be analyzed, for example, ticker symbols, sales data, machine data, etc.
  • In ScaleOut GCE, the analysis and merge methods can be written either in Java or C#. Since GCE holds the objects to be analyzed or merged in memory, these methods are written without the need to use grid APIs. The analysis method specifies the analysis logic for a single data object selected by the query specification. For example, it might calculate stock trading profits for one company’s recent history of stock prices. The merge method combines the results of analyzing multiple objects and is repeatedly executed as necessary to merge all results. In the above example, it might calculate the average return for stock trades spanning many companies.
  • Using a special API in GCE called “invoke” and supplying the query specification and both the analysis and merge methods, the application starts a map/reduce computation called a “parallel method invocation” (PMI). GCE automatically performs the query, analysis, and merge steps in parallel across all grid servers using a multi-threaded computation engine and then returns the final, merged result back to the application. PMI operations can be performed repeatedly to provide a continuous stream of results. Because GCE avoids batch scheduling and keeps the overhead for starting and running the analysis low, it returns results with minimum latency for near real-time performance.

When using an IMDG, all computations are performed “in-place,” reducing data motion which is the enemy of high performance for map/reduce. Also, the IMDG leverages its features for maximizing scalability and high availability, such as partitioning, peer-to-peer architecture, and load-balancing. In addition, GCE implements special features for ensuring the high availability of map/reduce computations.

Lowering the Complexity Barrier

The map/reduce programming model has generated widespread interest in large part due to the popularity of the Hadoop open source software stack. However, Hadoop introduces a complex programming model and deployment architecture which must be thoroughly understood for Hadoop to be used effectively. For example, applications need to be written to fit Hadoop’s specific parallel execution model, incorporating several specialized elements such as record readers, mappers, combiners, and reducers. The number and interaction of these elements impact performance and require tuning. Beyond this, Hadoop’s execution environment, including the HDFS file system, job tracker (that is, the batch scheduler), and task trackers on each execution node must be deployed and managed. It may take a seasoned Java developer with knowledge of parallel programming weeks to become proficient with Hadoop. These complexities create a steep learning curve which impedes rapid adoption.

In contrast, the IMDG-based approach to map/reduce data analysis eliminates much of Hadoop’s complexity. Its object-oriented approach offers a simpler parallel execution model that reduces development time and eliminates the need for tuning. The user invests much less time in learning the model and focuses more on the analytical challenges of the business problem. Learning curves are flattened, and productivity is increased.

Delivering High Performance

To see the performance benefits of using an IMDG with integrated map/reduce, consider a real-world financial analysis application that  compares various stock trading strategies based on historical market data stored in the IMDG. This application makes use of the IMDG’s analytics engine to perform a map/reduce analysis across all grid servers and merge the results. Each stock history is stored as a separate object within the IMDG, and specific stock histories are selected for analysis using a parallel query. The analysis method evaluates a set of trading strategies across a single stock history, and the merge method combines the results across two stocks. The analytics engine repeatedly executes these methods to analyze all selected stocks and merge the results.

Performance measurements were made for this application using ScaleOut GCE’s IMDG to evaluate throughput scaling as the number of stock histories and grid servers was proportionally increased. As the graph below illustrates, the IMDG delivers linearly scalable throughput (shown as the red line in the graph). An alternative implementation of this application was measured using Hadoop’s map/reduce environment. Hadoop provided linear scaling with about 16X lower throughput (shown as the blue line in the graph) due to significant overhead introduced by file I/O and batch scheduling. By staging the stock history data in the IMDG instead of the Hadoop file system (HDFS), Hadoop’s throughput was increased by about 6X (shown as the green line), although it was still significantly below the IMDG’s throughput due to file I/O between the map and reduce phases.

 Throughput Comparison 

In Summary

With the ever increasing explosion in data for analysis and the need for fast insights on emerging trends, IMDGs offer a highly attractive platform for hosting map/reduce analysis. By simplifying the development model, IMDGs shorten the learning curve in developing analysis codes and eliminate the tuning steps required by more complex platforms. Because IMDGs run the analysis on data already staged in memory and load-balanced across grid servers, file I/O is eliminated and data motion is minimized. IMDGs also provide the infrastructure needed to automatically run analysis code on all grid servers in parallel and then combine the results with minimum latency. The net result is that by using an IMDG, application developers can easily analyze fast-changing, memory-based data and discover data patterns and trends that are vital to a company’s success.


Dr. William L. Bain is founder and CEO of ScaleOut Software, Inc. Bill has a Ph.D. in electrical engineering/parallel computing from Rice University, and he has worked at Bell Labs research, Intel, and Microsoft. Bill founded and ran three start-up companies prior to joining Microsoft. In the most recent company (Valence Research), he developed a distributed Web load-balancing software solution that was acquired by Microsoft and is now called Network Load Balanc­ing within the Windows Server operating system. Dr. Bain holds several patents in computer architecture and distributed computing. As a member of the Seattle-based Alliance of Angels, Dr. Bain is actively involved in entrepreneurship and the angel community.

Subscribe to HPCwire's Weekly Update!

Be the most informed person in the room! Stay ahead of the tech trends with industy updates delivered to you every week!

Geospatial Data Research Leverages GPUs

August 17, 2017

MapD Technologies, the GPU-accelerated database specialist, said it is working with university researchers on leveraging graphics processors to advance geospatial analytics. The San Francisco-based company is collabor Read more…

By George Leopold

Intel, NERSC and University Partners Launch New Big Data Center

August 17, 2017

A collaboration between the Department of Energy’s National Energy Research Scientific Computing Center (NERSC), Intel and five Intel Parallel Computing Centers (IPCCs) has resulted in a new Big Data Center (BDC) that Read more…

By Linda Barney

Google Releases Deeplearn.js to Further Democratize Machine Learning

August 17, 2017

Spreading the use of machine learning tools is one of the goals of Google’s PAIR (People + AI Research) initiative, which was introduced in early July. Last week the cloud giant released deeplearn.js as part of that in Read more…

By John Russell

HPE Extreme Performance Solutions

Leveraging Deep Learning for Fraud Detection

Advancements in computing technologies and the expanding use of e-commerce platforms have dramatically increased the risk of fraud for financial services companies and their customers. Read more…

Spoiler Alert: Glimpse Next Week’s Solar Eclipse Via Simulation from TACC, SDSC, and NASA

August 17, 2017

Can’t wait to see next week’s solar eclipse? You can at least catch glimpses of what scientists expect it will look like. A team from Predictive Science Inc. (PSI), based in San Diego, working with Stampede2 at the Read more…

By John Russell

Microsoft Bolsters Azure With Cloud HPC Deal

August 15, 2017

Microsoft has acquired cloud computing software vendor Cycle Computing in a move designed to bring orchestration tools along with high-end computing access capabilities to the cloud. Terms of the acquisition were not disclosed. Read more…

By George Leopold

HPE Ships Supercomputer to Space Station, Final Destination Mars

August 14, 2017

With a manned mission to Mars on the horizon, the demand for space-based supercomputing is at hand. Today HPE and NASA sent the first off-the-shelf HPC system i Read more…

By Tiffany Trader

AMD EPYC Video Takes Aim at Intel’s Broadwell

August 14, 2017

Let the benchmarking begin. Last week, AMD posted a YouTube video in which one of its EPYC-based systems outperformed a ‘comparable’ Intel Broadwell-based s Read more…

By John Russell

Deep Learning Thrives in Cancer Moonshot

August 8, 2017

The U.S. War on Cancer, certainly a worthy cause, is a collection of programs stretching back more than 40 years and abiding under many banners. The latest is t Read more…

By John Russell

IBM Raises the Bar for Distributed Deep Learning

August 8, 2017

IBM is announcing today an enhancement to its PowerAI software platform aimed at facilitating the practical scaling of AI models on today’s fastest GPUs. Scal Read more…

By Tiffany Trader

IBM Storage Breakthrough Paves Way for 330TB Tape Cartridges

August 3, 2017

IBM announced yesterday a new record for magnetic tape storage that it says will keep tape storage density on a Moore's law-like path far into the next decade. Read more…

By Tiffany Trader

AMD Stuffs a Petaflops of Machine Intelligence into 20-Node Rack

August 1, 2017

With its Radeon “Vega” Instinct datacenter GPUs and EPYC “Naples” server chips entering the market this summer, AMD has positioned itself for a two-head Read more…

By Tiffany Trader

Cray Moves to Acquire the Seagate ClusterStor Line

July 28, 2017

This week Cray announced that it is picking up Seagate's ClusterStor HPC storage array business for an undisclosed sum. "In short we're effectively transitioning the bulk of the ClusterStor product line to Cray," said CEO Peter Ungaro. Read more…

By Tiffany Trader

Nvidia’s Mammoth Volta GPU Aims High for AI, HPC

May 10, 2017

At Nvidia's GPU Technology Conference (GTC17) in San Jose, Calif., this morning, CEO Jensen Huang announced the company's much-anticipated Volta architecture a Read more…

By Tiffany Trader

How ‘Knights Mill’ Gets Its Deep Learning Flops

June 22, 2017

Intel, the subject of much speculation regarding the delayed, rewritten or potentially canceled “Aurora” contract (the Argonne Lab part of the CORAL “ Read more…

By Tiffany Trader

Reinders: “AVX-512 May Be a Hidden Gem” in Intel Xeon Scalable Processors

June 29, 2017

Imagine if we could use vector processing on something other than just floating point problems.  Today, GPUs and CPUs work tirelessly to accelerate algorithms Read more…

By James Reinders

Quantum Bits: D-Wave and VW; Google Quantum Lab; IBM Expands Access

March 21, 2017

For a technology that’s usually characterized as far off and in a distant galaxy, quantum computing has been steadily picking up steam. Just how close real-wo Read more…

By John Russell

Nvidia Responds to Google TPU Benchmarking

April 10, 2017

Nvidia highlights strengths of its newest GPU silicon in response to Google's report on the performance and energy advantages of its custom tensor processor. Read more…

By Tiffany Trader

Russian Researchers Claim First Quantum-Safe Blockchain

May 25, 2017

The Russian Quantum Center today announced it has overcome the threat of quantum cryptography by creating the first quantum-safe blockchain, securing cryptocurrencies like Bitcoin, along with classified government communications and other sensitive digital transfers. Read more…

By Doug Black

HPC Compiler Company PathScale Seeks Life Raft

March 23, 2017

HPCwire has learned that HPC compiler company PathScale has fallen on difficult times and is asking the community for help or actively seeking a buyer for its a Read more…

By Tiffany Trader

Trump Budget Targets NIH, DOE, and EPA; No Mention of NSF

March 16, 2017

President Trump’s proposed U.S. fiscal 2018 budget issued today sharply cuts science spending while bolstering military spending as he promised during the cam Read more…

By John Russell

Leading Solution Providers

Groq This: New AI Chips to Give GPUs a Run for Deep Learning Money

April 24, 2017

CPUs and GPUs, move over. Thanks to recent revelations surrounding Google’s new Tensor Processing Unit (TPU), the computing world appears to be on the cusp of Read more…

By Alex Woodie

CPU-based Visualization Positions for Exascale Supercomputing

March 16, 2017

In this contributed perspective piece, Intel’s Jim Jeffers makes the case that CPU-based visualization is now widely adopted and as such is no longer a contrarian view, but is rather an exascale requirement. Read more…

By Jim Jeffers, Principal Engineer and Engineering Leader, Intel

Google Debuts TPU v2 and will Add to Google Cloud

May 25, 2017

Not long after stirring attention in the deep learning/AI community by revealing the details of its Tensor Processing Unit (TPU), Google last week announced the Read more…

By John Russell

MIT Mathematician Spins Up 220,000-Core Google Compute Cluster

April 21, 2017

On Thursday, Google announced that MIT math professor and computational number theorist Andrew V. Sutherland had set a record for the largest Google Compute Engine (GCE) job. Sutherland ran the massive mathematics workload on 220,000 GCE cores using preemptible virtual machine instances. Read more…

By Tiffany Trader

Six Exascale PathForward Vendors Selected; DoE Providing $258M

June 15, 2017

The much-anticipated PathForward awards for hardware R&D in support of the Exascale Computing Project were announced today with six vendors selected – AMD Read more…

By John Russell

Top500 Results: Latest List Trends and What’s in Store

June 19, 2017

Greetings from Frankfurt and the 2017 International Supercomputing Conference where the latest Top500 list has just been revealed. Although there were no major Read more…

By Tiffany Trader

IBM Clears Path to 5nm with Silicon Nanosheets

June 5, 2017

Two years since announcing the industry’s first 7nm node test chip, IBM and its research alliance partners GlobalFoundries and Samsung have developed a proces Read more…

By Tiffany Trader

Messina Update: The US Path to Exascale in 16 Slides

April 26, 2017

Paul Messina, director of the U.S. Exascale Computing Project, provided a wide-ranging review of ECP’s evolving plans last week at the HPC User Forum. Read more…

By John Russell

  • arrow
  • Click Here for More Headlines
  • arrow
Share This