Capitalizing on the New Era of In-memory Computing


Published on

In-memory computing is all set to turn mainstream due to the host of benefits it offers and supportive factors, such as dropping memory prices, availability of more computing power and the growing need to leverage Big Data that requires new methods of processing unstructured information. Companies should use in-memory techniques while developing new analytics applications to take advantages of them and also consider re-engineering legacy systems to prepare them for new world of data, reduce complexity, improve scalability and speed.

  • Be the first to comment

  • Be the first to like this

No Downloads
Total views
On SlideShare
From Embeds
Number of Embeds
Embeds 0
No embeds

No notes for slide

Capitalizing on the New Era of In-memory Computing

  1. 1. Insights Capitalizing on the New Era of In-memory Computing - Girish Khanzode The emergence of intelligent devices, mobility, social media, pervasive networking and analytics has created a digital economy that is fundamentally altering the traditional relationships amongst companies, partners and consumers. Higher growth of internal and external data sources has resulted in a dramatic rise in volume, velocity and variety of unstructured data, often referred to as Big Data. Processing this data effectively and quickly followed by distilling meaningful insights is critical for growth and leadership of companies.
  2. 2. With accelerating market dynamics, the window ofdecision making is narrowing. Users need split-second,correct answers to almost any question, using any datafrom anywhere. A rapid and analytics-driven responsehas become crucial when faced with internal andexternal events. Owing to technological innovationand changing consumer aspirations, managerscannot rely solely on experience or gut feel toarrive at the right decisions. Instead, theyrequire real-time visibility into completetransactional data and analytics processingto take effective measures.Business users expect to rely less on ITresources. There is increasing pressure to reduce ITcosts. Customers now take for granted a satisfyingend-user experience coupled with sophisticatedself-service offerings.Increasing unstructured data is also diminishing theeffectiveness of traditional forecasting models. Onlinereviews and discussions on products now create a fardeeper impact on revenues and product lifecycles thanactual point-of-sale data.In order to take on these new challenges, data mustbe analyzed within expected time windows for relevantdecisions and actions. Consequently, new techniques of dataprocessing have become extremely critical. THE RISE OF IN-MEMORY COMPUTING ERA As predicted by Moore’s Law, memory prices continue to slide coupled with a rise in the number of processors on the chip. DRAM prices are dropping by half every couple of years. Despite these favorable advancements, the single point of latency still continues to be the hard disk even if we consider fast access disks like solid state hard drives (SSD). To overcome this bottleneck, next-generation servers are increasingly relying on directly accessing memory to service I/O requests that were earlier handled by primary storage. Multi-core CPUs are enabling high speed large-scale parallel processing of in-memory data. These factors are propelling in-memory computing since all the data that needs to be processed can now be stored in memory itself. In-memory computing can process massive quantities of real-time events and data in the main memory of the server within seconds to detect correlations and patterns, displaying emerging opportunities and risks, thus expediting informed business decisions. This new computing model has the potential to deeply impact existing IT processes and dramatically shorten batch process execution time, which earlier took days or weeks. It facilitates hosting of both transactional and analytical applications on a single server, requiring smaller disk storage capacities and fewer databases, contributing to IT cost savings. In-memory computing is also extremely effective when handling unstructured data like social media, video or machine data.2 | Infosys
  3. 3. AdvantagesIn-memory computing holds immense potential benefits for data Relational databases must maintain up-to-date data maps hosted onprocessing. Extremely fast computational speeds ensure that results disks, find the right volume when executing each element of a query,are up-to-date and hence more accurate. It enables analysis of more and also update the results in the system, necessitating storage tuninggranular data which would otherwise be obscured in aggregation. IT by database administrators to maintain optimal system levels. Keepingcosts and complexities are lowered when managing heterogeneous data in memory eliminates all these costly efforts and allows softwaredata types and large volumes. Data access speed becomes lightning to easily scale as demand and data volume since movement to main memory from disk storage is avoided, In-memory databases do not have overheads of disk managementresulting in negligible data latency. like on-disk databases. Format changes are performed in memoryTransactional operations become significantly faster since an without data restructuring and data need not be reallocated as itin-memory database requires 5 nanoseconds to access memory grows. Negligible I/O waiting time inside in-memory databases greatlycompared to 5 milliseconds for a disk read operation by a traditional improves and smoothens data throughput. With time, businessesdatabase, making it a million times faster. Data loading from disk- experience growth in data volumes accompanied by increasingbased storage is circumvented, thus simplifying database structures demand for timely access and analysis of the same. The in-memoryand reducing software complexity while also requiring less CPU approach becomes a superior alternative to disk-based systems incomputing power. It is estimated that a typical database management addressing these needs.system based on an in-memory model is around 15 times faster than Big Data metadata is known for multi-dimensionality, flat-schemas,the traditional on-disk one. collections, derivation and recursion. This creates new challenges inIn-memory computing enables parallel processing in the database traditional relational database environments, resulting in complexlayer instead of the application layer, resulting in faster execution. SQL joins and schemas that are difficult to process inside softwareCombining analytics and operations capabilities in a single system program code or stored procedures. These challenges are mitigatedcoupled with reduction in redundant data storage results in in the in-memory approach with use of NoSQL processing techniques.approximately 30% lower costs due to flattened IT infrastructure Relational databases store table rows as blocks on the disk along withcomprising leaner hardware and smaller systems. IT software indexed columns. In-memory databases, instead, operate on datasupport costs are lowered due to reduced extract, transform, and grouped into columns that require fewer I/O operations. Columnarload (ETL) processes between systems. In addition, on-the-fly data is of the same type and can be easily and efficiently compressed,aggregation eliminates the need for manual data query tuning and which makes it a highly optimal alternative for analytical aggregation efforts. This approach avoids data searches involving scanning of index entriesCompanies want to micro-segment customers, requiring the analysis and navigation of indirect references to find the disk pages where theof granular data as well as the factors that influence buying patterns, data resides. In-memory databases scan columns and jump memoryin order to improve marketing efficacy. However, traditional data pointers to prepare the desired data. Thus, queries run significantlywarehousing methods focus on data aggregation and preparation faster than on disk-based relational databases even though theyprior to the data analysis process. The resulting loss in data granularity use optimization techniques like caching of rows in memory afterdue to anticipated data usage assumptions made before preparation reading from the disk.can result in ineffective decisions and missing patterns. Infosys | 3
  4. 4. Challenges Real-time data access can increase its vulnerability or can introduce new risks, especially in the financial domain. For example, a stock trader may execute trades in real-time before a compliance system can kick in. Or a production analysis report that is used to make a decision might provide completely different conclusions when prepared again a few minutes later. Companies will need to set up different sets of policies for real-time analytical reporting involving what-if scenarios and for time-logged compliant production reports.4 | Infosys
  5. 5. Application AreasFollowing are some of the potential areas that can benefit immensely from in-memory technologies:Personalized Incentives candidates and then execute these trades in seconds. If data is fetched from disks using traditional techniques, the system might give lossesPoint-of-sale retailers and loyalty program distributors can offer more due to time delays. In-memory systems offer the speed and scalabilityattractive real-time discounts to long-term customers based on their critical to succeed in these operations.individual purchase history. Analytics of this nature requires a largeamount of data mining with quick results. E-commerce sites canleverage in-memory databases and parallel processing algorithms to Next Generation Analyticshandle a far greater load with improved site response performance In order to improve quality of decisions, data has to be processed atto enhance customer satisfaction and lure new customers, resulting a higher frequency. Using traditional analytics, mining of terabytesin higher revenues. of data will require days to identify useful results and trends. In- memory analytics can perform the same operation in hours, evenOptimized Pricing minutes, making in-memory technology an integral part of high- performance analytics in future. Speedy execution will bring hugeLarge retail chains can advertise the most competitive rates for their competitive advantage to businesses that need to make correct andgoods by closely monitoring their inventory and carefully tracking rapid decisions driven by the latest trends.consumption patterns and trends, to be able to optimally order itemsfrom suppliers. This process can be further streamlined by trackingtruck routes and traffic data in real-time in order to meet dynamic Risk Managementdemand. Software systems for this level of sophistication require Fraud detection is critical for companies engaged in high-valueabsolutely latest data and the capability to process it in real-time in financial transactions. It is imperative to raise an alert when aorder to offer products at the lowest cost while avoiding overstocking transaction is about to happen, as delays could prove disastrous. Sinceor running out of fast moving goods. detection algorithms must process terabytes of financial transactions in a few seconds, in-memory systems become essential.High Frequency TransactionsTrading firms deploy sophisticated high-frequency trading platformsthat automatically and swiftly make buy and sell decisions that aredriven by price pattern changes in order to survive and make profits.A delay of even a few microseconds can degrade these systems’performance. Analytical algorithms must scan long-term price pointsof hundreds of stocks to identify promising long/short selling stock Infosys | 5
  6. 6. Conclusion In-memory computing is all set to turn mainstream due to the host of benefits it offers and supportive factors, such as dropping memory prices, availability of more computing power and the growing need to leverage Big Data that requires new methods of processing unstructured information. Companies should use in- memory techniques while developing new analytics applications to take advantages of them and also consider re-engineering legacy systems to prepare them for new world of data, reduce complexity, improve scalability and speed.6 | Infosys
  7. 7. About the Author Girish Khanzode Products & Platforms Innovator for Futuristic Technologies, Infosys Labs Girish is a veteran in Enterprise Software Product design and development with more than 20 years of professional experience. He has built and led large product engineering teams to deliver highly complex products in multiple domains, covering entire product life cycle. Currently, he is engaged in innovating and building the next generation products and platforms in emerging new technology areas like Enterprise Data Security and Privacy, Collaboration technologies, Digital Workplace, Social Analytics, Smart Cities, Big Data and Internet of Things. Girish holds M. Tech. degree in Computer Engineering and a bachelor’s degree in Electrical Engineering. Infosys | 7
  8. 8. About InfosysInfosys partners with global enterprises to drive their innovation-led growth.Thats why Forbes ranked Infosys #19 among the top 100 most innovativecompanies. As a leading provider of next-generation consulting, technologyand outsourcing solutions, Infosys helps clients in more than 30 countriesrealize their goals. Visit and see how Infosys (NYSE: INFY),with its 150,000+ people, is Building Tomorrows Enterprise® today.For more information, contact© 2013 Infosys Limited, Bangalore, India. All Rights Reserved. Infosys believes the information in this document is accurate as of its publication date; such information is subject to change without notice.Infosys acknowledges the proprietary rights of other companies to the trademarks, product names and such other intellectual property rights mentioned in this document. Except as expressly permitted,neither this documentation nor any part of it may be reproduced, stored in a retrieval system, or transmitted in any form or by any means, electronic, mechanical, printing, photocopying, recording orotherwise, without the prior permission of Infosys Limited and/ or any named intellectual property rights holders under this document.