Hitachi Vantara Community : Blog List - Hu's Place

Overcoming CPU Chokepoints For NVMe

community-noreply@hitachivantara.com — Mon, 12 Aug 2019 20:53:00 GMT

Marc Staimer of Dragon Slayer Consulting published a recent article on the CPU chokepoint in servers and controllers for NVMe storage. This supports blogs that I have recently posted on CPU architectural limitations and the need for accelerated compute and other computer architectures.

Marc observes that as more NVMe flash SSDs are required, then the supporting hardware gets increasingly complicated. “It usually means more CPUs, either internal or external ones. The storage can be DAS or shared across NVMe-oF. Either way, more CPUs, drives, drive drawers, switches, adapters, transceivers and cables will be required. The general industry consensus is that scaling capacity and performance using NVMe drives and NVMe-oF just requires more hardware. Storage Class Memory technologies will only exacerbate the CPU chokepoint problem, because their increased performance puts even more load pressure on the CPU.”

“But here's the rub. These systems offer quite noticeable diminishing marginal returns. The hardware grows much faster than the performance gains. This occurs no matter how many CPUs or NVMe flash SSDs are added. Eventually, more hardware means a negative return on overall performance.

The root cause of this NVMe performance challenge isn't hardware. It's storage software that wasn't designed for CPU efficiency. Why bother with efficiency when CPU performance was doubling every 18 to 24 months? Features, such as deduplication, compression, snapshots, clones, replication, tiering and error detection and correction, were continually added to storage software. And many of these features were CPU intensive. When storage software is consuming CPU resources, they aren't available for storage I/O to the high-performance drives.”

While Hitachi Vantara, has not yet delivered NVMe or NVMe-oF in the enterprise storage VSP platform, we have been making changes to the VSP Storage controller in preparation for the introduction of NVMe and NVMe oF when the standards become finalized. I blogged about this a year ago. Essentially, we have rewritten the SVOS (Storage Virtualization Operating System) VSP controller software for NVMe and released it as SVOS RF where the RF stands for Resilient Flash. This systems software was re architected and designed to optimize and scale NVMe performance. The other thing we did in the hardware was to accelerate compute through the offload of some tasks to FPGAs. We also optimized the data path with improved cache algorithms.

The result of these changes helped to accelerate our performance even without NMVe or NVMe oF. Our Flash performance with SAS (Serial Attached SCSI) are comparable to some of the startups who are delivering NVMe storage systems.

Last year’s August 6, 2018, Gartner Critical Capabilities for Solid State Arrays report provides some insight on what our capabilities will be. In terms of performance rating, the VSP F series came in third in front of several vendors that had NVMe. This evaluation did not include the latest SVOS RF and VSP F900/F700/F370/F350 enhancements which were announced in May 2018 because they did not make Gartner’s cutoff date for the 2018 evaluation. These new enhancements featured an improved overall flash design, with 3x more IOPS, lower latency and 2.5x more capacity than previous VSP all flash systems.

The only two vendors ahead of the F series in performance at that time, were the Kaminario K2 and the Pure Storage Flash Blade, none of which have the high reliability, scalability and enterprise data services of the VSP. In fact, the VSP F series placed the highest in RAS (reliability, availability, serviceability) of all 18 products that were evaluated. The Kaminario K2 has a proprietary NVMe-oF host connection which they call NVMeF. One can assume that the performance of the Hitachi Vantara All Flash Arrays, even with SCSI/SAS would be higher if the new models of the VSP and SVOS RF had been included in the evaluation. Here are the Product scores for the High-Performance Use Case for the top three places on a scale from 1 to 5 with 5 being the highest.

Kaminario K2 4.13

Pure Storage FlashBlade 4.08

Hitachi VSP F Series 4.03

Pure Storage M and X Series 4.03

While the standards for NVMe oF are still being worked on and still to be proven, the NVMe standards are pretty close to being finalized, so you can expect to see NVMe coming soon from Hitachi in the near future, and you should expect to see It blow away the competition, since we have already done the ground work to address the choke points that Marc Staimer identifies.

TCP Is A Network Protocol for NVMe

community-noreply@hitachivantara.com — Thu, 08 Aug 2019 21:25:20 GMT

Last year in August, I posted a blog about NVMe, an open standards protocol for digital communications between servers and non-volatile memory storage. It replaces the SCSI protocol that was designed and implemented for mechanical hard drives which processed one command at a time. NVMe was designed for flash and other non-volatile storage devices that may be in our future. The command set is leaner, and it supports a nearly unlimited queue depth that takes advantage of the parallel nature of flash drives (a max 64K queue depth for up to 64K separate queues).

There are several transports for the NVMe protocol. NVMe by itself can use PCIe (Peripheral Component Interconnect Express), which is a standard type of connection for internal devices in a computer, to transport signals over a PCIe bus from a non-volatile memory storage device (SSD). Hitachi Vantara has implemented NVMe on our hyperconverged, Unified Compute Platform (UCP HC), where internal NVMe flash drives are connected directly to the servers through PCIe. While direct-attached architectures offer high performance and are easy to deploy at a small scale, data services like snapshots and replication will have to be done by the host CPU which adds overhead. If a VM has to access another node to find data, you will need to transfer the data or the application to the same node. For smaller data sets this isn't an issue, but as the workload increases, this negates some of the performance advantages of NVMe. However, you are still ahead of the game compared to SCSI devices and UCP HC with NVMe is a great option for hyperconverged infrastructure workloads.

In my post from last year, I introduced the other transports that enable NVMe to be transported over a Fabric for external attachment (NVMe-oF). These transports included, NVMe-oF using Fibre Channel and NVMe-oF using RDMA over Infiniband, RoCE, or iWARP.

Late last year, another transport was ratified, NVMe-oF using TCP. The value proposition for TCP, is that it’s well-understood, and can use the TCP/IP routers and switches. One of the disadvantages with TCP/IP is congestion. Unlike FC where buffer credits are used to ensure that the target can receive a packet before the packet is sent, The IP layer drops the packet when the network gets congested, and it is up to TCP to ensure that no data is lost, which causes the transport to slow down when the network gets overloaded. While TCP overreacts to congestion, it doesn’t fail; it just slows down. NVMe over TCP is still substantially ahead of SCSI in terms of latency while still behind NVMe over FC and RDMA.

RDMA, provides direct memory access, and will be the choice for high performance, but there will be decisions to be made on the choice of networking protocol; Infiniband, RoCE (RDMA over converged ethernet) or iWARP (Internet Wide Area RDMA Protocol).

While there is still some standardization to be done on NVMe-oF for Fibre Channel, this will probably be the network protocol to be accepted first, since it is more mature than NVMe-oF over TCP, provides flow control through buffer credits, and is a familiar network protocol for storage users. Like TCP, Fibre Channel can use existing routers and switches, with relatively minor changes in software. There will likely be different network protocols depending on use case. Direct attached NVMe over PCIe for hyperconverged and software defined storage, Fibre channel for enterprise storage, TCP for distributed storage and RDMA for high performance storage. SCSI will still be the dominant interface for the next few years. However, NVMe and NVMe-oF will eventually replace traditional SCSI based storage. I would expect the first implementations to be 50% Fibre Channel, 30% TCP, 12% PCIe, and 8% RDMA.

This could change dramatically depending on what the hyper scaling vendors do. This week Amazon acquired E8 Storage, an Israeli Company, that has an end to end 2U NVMe storage system that uses NVMe-oF over TCP. TCP is a logical choice for a cloud company.

Thoughts on HPE’s Acquisition of MapR

community-noreply@hitachivantara.com — Mon, 05 Aug 2019 23:28:40 GMT

In June I posted a blog commenting on the cloud management company MapRclosing its headquarters and laying off 122 workers. At that time Cloudera, another cloud big data management company, announced reduced earnings and reduced outlook which drove its stock down over 38% to around $5. It was only in January of this year when Cloudera and Hortonworks, two of the biggest players in the Hadoop big data space, announced an all-stock merger, which was expected to give new life to these companies in the big data analytics market.

Therefore it was not surprising to hear today’s news of the acquisition of MapR by HPE. With this acquisition HPE said the deal includes MapR’s technology, intellectual property and expertise in artificial intelligence and data management. There will also be a number of employees joining with this acquisition. The MapR technology will be folded into HPE’s Intelligent Data Platform, a set of technologies for data lifecycle management.

Just a few years ago, cloud data management companies like MapR, Cloudera, and Hortonworks were Unicorns, the darlings of Wall Street. Analysts attribute this decline to a natural consolidation of the “surplus of enterprise Hadoop companies” after the hype and frenzy of the VC community reached its peak. The cost of sales and services are very high and the pay per use model puts a squeeze on cash flow. The difficulty of monetizing a business based upon open-source software is challenging. Analysts are also predicting that the Big Data Analytics ecosystem will converge around AWS, Azure, and Google Cloud and many of these smaller companies will be acquired or displaced by the large public cloud vendors.

MapR’s technology will be particularly valuable to HPE in helping customers stitch together data pipelines across multiple on-premises and cloud environments as well as to run multiple workloads in the same environment. According to Patrick Osborne, vice president of big data and secondary storage at HPE is quoted as saying, “ MapR’s distributed file system provides the capability of a data fabric that allows people to manage their analytics on the edge as well as in the core. We didn’t have a technology that would allow customers to do that.”

Although there are many advantages for HPE in this acquisition, they still have to compete with the hyper-cloud vendors in the big data analytics cloud ecosystem, AWS, Azure, and Google cloud who have a clear head start. The recent acquisition of Tableau by SalesForce and Looker by Google, are indicative of a trend by public cloud providers moving to provide end to end big data analytics solutions across multiple clouds. HPE will have to play catchup while integrating the technology, personnel, partners, and customers of MapR. They will also have to solve the revenue problems that plagued MapR.

While Hybrid cloud provides an opportunity to augment public cloud offerings, HPE and MapR will have upfront development and support costs which will impact cash flow. Customers want to move to the cloud faster than MapR can allow them. These customers do not have the luxury of waiting for and trialing CDP while there are other options that are available today.

Hitachi Vantara customers have several options for accelerating their movement to the cloud and big data analytics for structured and unstructured data. One approach is to develop a data lake with Pentaho and other best of breed data ingestion and data orchestration tools for big data analytics that can span multiple cloud delivery platforms with a common meta data catalog and schema. Pentaho’s low code approach can simplify and accelerate the implementation of big data analytics. Hitachi Vantara has taken this approach internally for our enterprise data that need to reside within our private cloud

Another option from Hitachi Vantara is to use our REAN Cloud ,a global Cloud Systems Integrator (CSI), Managed Service Provider (MSP) and Premier Consulting Partner in the Amazon Web Services (AWS) Partner Network (APN) and Microsoft's Azure Silver Partner membership. REAN Cloud offers consulting & professional services, including cloud strategy, assessment, cloud migration, and implementation to realize our customers’ vision. REAN Cloud provides a REAN Cloud Accelerated Migration Program (RAMP) which can accelerate the migration to public cloud from a matter of weeks to days with their automated services and migration consulting expertise. Migration to the hyper cloud vendors enables the use of their menu of analytics tools. REAN Cloud incudes 47Lining an AWS Advanced Consulting Partner with Big Data Competency designation. 47Lining develops big data solutions and delivers big data managed services built from underlying AWS building blocks like Amazon Redshift, Kinesis, S3, DynamoDB, Machine Learning and Elastic MapReduce.

A full transition to the cloud has proved more challenging than anticipated and many companies are looking to hybrid cloud solutions to transition to the cloud at their own pace and at a lower risk and cost. Companies are looking for DataOps tools and platforms, and systems integrators that can help them create data lakes and deliver big data analytics in a timely manner. They want proven vendors who will be with them for the long term and who already have the platforms and services for hybrid cloud and big data analytics that can work within the ecosystem of public and private clouds.

All comments are my own and should not be considered to reflect the opinions of Hitachi Vantara.

Mainframe FICON Connection for Analytics

community-noreply@hitachivantara.com — Mon, 05 Aug 2019 19:51:20 GMT

On the occasion of the 50^thanniversary of the Apollo lunar landing, I posted a blog about my experience with IBM mainframes in support of North American Rockwell’s work on the Lunar command and service modules. The architecture for the first S/360 mainframe which was introduced in 1964, is still alive and well today. While the first mainframes had less processing power and memory than today’s iphone, technology has kept apace and today’s mainframe (IBM Z Systems) still plays a central role in the daily operations of most of the world's largest corporations. In banking, finance, health care, insurance, utilities, government, and a multitude of other public and private enterprises, the mainframe computer continues to be the foundation of modern business. And as such mainframes generate a lot of valuable data for analytics.

However, combining mainframe data with other forms of data is difficult since mainframe data is formatted in EBCDIC while open systems data is coded in ASCII. ASCII and EBCDIC are two character encoding standards. Therefore, in order to combine data from mainframes with data from open systems, there has to be a translation of the data format. However, the bigger problem is in the transmission of the data between mainframes and open systems in order to combine the data for analysis. Normally this would have to be over TCP/IP using FTP, File Transport Protocol.

Although the core data may reside on the mainframe, the bulk of data today is unstructured data residing on open systems and now IoT and mobile devices. AI and ML programs are primarily built for open systems. Therefore, data is usually sent to an open systems server for translation and analysis in combination with other data, and most of the data is sent over FTP from the mainframe. This presents a problem since FTP ports are wide open and there are insufficient controls and restrictions on what types of data can or cannot be sent via FTP credentials in the clear. While there are secure transfer protocols like SFTP and FTPS, clients do not like to use them due to the overhead and costs associated with encryption. There have been many recent reports of data breaches related to FTP. FTP was never designed with security in mind and because of that, it’s become one of the favorite venues for hackerslooking to get into a corporate network.

In order to solve this problem, Hitachi Vantara and Luminex have partnered on a Mainframe Data Integration (MDI) platform that leverages the Luminex mainframe channel I/O interface to securely share and transfer data between mainframes and distributed systems environments using the FICON channel. Since FICON channels are specifically designed and optimized for the purpose of moving data from and to the mainframe. MDI provides a faster, more secure, cheaper (less CPU) and easy (native) platform for connecting mainframe data vs. TCP/IP. A financial customer was able to reduce their fraud detection investigation from 50 days down to 5 days and instead of spending 90% of their time on data collection and only 10% on analyzing the data, they can now spend 80% of their time on investigations, improving the quality of results.

I will be presenting this along with other topics at a SHARE Lunch and Learn in Pittsburgh this week.

LNL: [15] The Connected Mainframe

Room: Room 317-318
Session Number: 25817

Thursday, August 08, 2019: 12:45 PM - 1:45 PM

Luminex will also be presenting the MDI

Adventures in Mainframe Data Integration:

How MDI is Changing the Value and Economics of the Mainframe

Tuesday, August 6

2:15 PM - 3:15 PM

Room: 403-404

Research in Regenerative Medicine

community-noreply@hitachivantara.com — Thu, 01 Aug 2019 12:32:12 GMT

This week I am at the Hitachi Central research Lab in Kokubunji, Japan, to attend Hitachi’s annual Kenpatsu. This is an event where the Central research lab provides an update to the global Hitachi organizations on the many projects that they are working on. Since Hitachi is a technology company with extensive R&D capabilities, you can imagine how exciting this was for me to attend especially with the explosion of technology that is going on today. The main focus for Hitachi’s research is on Social Innovation, making the world a smarter, healthier, and safer world.

One project that was especially interesting for me (considering my advanced age) was the work that was being done in regenerative medicine. This work is being done in our Kobe Lab, in research partnership with Kyoto University and Sumitomo Dainippon Pharma Co., Ltd. Kyoto University is where Dr. Shinyu Yamanaka was awarded the Nobel Prize for the discovery that mature cells could be converted to stem cells for use in regenerative medicine.

Regenerative medicine is a branch of research in tissue engineering and molecular biology which deals with the "process of replacing, engineering or regenerating human cells, tissues or organs to restore or establish normal function”.

The Hitachi Kobe lab is developing equipment to automate the production of Induced pluripotent stem cells or iPS cells which were first generated by Dr. Yamanaka. In 2006, Dr. Yamanaka established that by introducing a small number of genes into ordinary human somatic (differentiated) cells, these pluripotent cells can differentiate into any type of cell in the body and proliferate indefinitely in culture. The process of changing a cell from a differentiated to a pluripotent state is called reprogramming. The method developed by Dr. Yamanaka has been shown to be highly reproducible, relatively simple, and is considered a major scientific breakthrough. Currently, cell cultures are produced by hand, but only experts are capable of producing medical-grade quality cells. If medical-grade cells are only able to be cultured by certain skilled people, regenerative medicine will not become generally available. Hitachi would like to change that by developing automated culture equipment capable of the stable mass production of iPS cells.

The prior alternative was to use human embryonic stem (ES) cells which were produced by removing cells from a 6-7 day old embryo and growing them in culture. While embryonic stem cells are natural, induced pluripotent stem cells can be generated using cells from an adult body, such as skin, which are plentiful and harmless to remove. As this does not require the destruction of an embryo, it avoids many of the ethical issues that surround human ES cells. Furthermore, unlike human ES cells, it is possible to derive patient-specific iPS cells and induce them into differentiated cells of various types, which can then be transplanted back into the patient without risk of immune rejection.

Hitachi began the research at their Center for Exploratory research in Saitama. Once they established the feasibility of building a machine to automate the development of iPS cells, Hitachi established the Hitachi Kobe Laboratory ("Kobe Lab") within the Kobe Biomedical Innovation Cluster ("KBIC"), where many people are carrying out cutting-edge advanced research in medical treatment, and moved the research team outside of the company in preparation for the social implementation phase. Hitachi felt that if they were to just stay within Hitachi, it would be difficult to develop a truly useful automated cell culture equipment. Hitachi believes in co-creation, that conducting R&D in the KBIC is the best environment to create truly useful automated cell culture equipment for the field of regenerative medicine. Hitachi Kobe Laboratory joined the Kobe Biomedical Innovation Center in the Kobe Biomedical Innovation Cluster.

Hitachi is developing automated culturing technology and process based on this research and clinical work that is capable of cultivating large quantities of high-quality medical grade cells for widespread use in regenerative medicine in the future. Hitachi has just overcome the first hurdle. This was to process cells with automated equipment that were of the same quality as those processed by expert human technicians. The next step is to be able to develop automated culture equipment capable of the stable mass production of cells of a quality exceeding that of the experts. The quality of the cultured cells is extremely important in regenerative medicine, and the "fight against bacteria" is a major issue. There are microorganisms in the air, and if just one of these bacteria enters the culture fluid, their presence will increase exponentially and destroy the human cell culture in an instant. In this event, those cells cannot be introduced to the body. This is a very difficult process and maintaining the sterility of the equipment is key when culturing cells.

Clinical research to confirm safety in humans commenced in 2013. For safety and other reasons, there is no fixed date, but researchers aim to make medical applications available as soon as possible. According to current research findings reported from Japan and overseas, iPS cells are capable of differentiation into the constituent cells of a wide range of tissues and organs, including nerves, cardiac muscle, and blood. (Think of the possibility of replacing brain cells that were damaged by Alzheimer) However, organs are more complex because of their three-dimensional (3D) structure. Small livers have been reported but there are as yet no reports of large 3D, functional organs of human size. This is an area that requires a combination of iPS cell technologies with 3D printers, biomaterials, and other technologies. This could be the next challenge for the Kobe Lab.

This was just one of many projects that I was able to hear about at our Hitachi Research Lab. If you are interested in hearing and seeing more of what Hitachi is researching for Social Innovation, you don't have to go to Kobe or Kokubunji. You can see them by visiting our NEXT 2019 event in Las Vegas, October 8-10 at the MGM Grand. You can click here to register.

Interrogare And demandi Realize DataOps Advantage with Hitachi Vantara

community-noreply@hitachivantara.com — Tue, 30 Jul 2019 11:45:10 GMT

Interrogare, a 20 year old market research company in Germany, conducts more than two million interviews each year across a range of industry sectors and focus areas, including product pricing, employee engagement, brand awareness and customer loyalty. Legacy tools and labor intensive manual processes were making it increasingly difficult to not only gather, store and analyze survey data but also to share findings with clients. According to their executives there were a lot of inefficient and expensive manual stepsCreating a centralized view of different data sources could take several weeks.

In 2015Interrogare decided to found demanti as a company to build the technical foundation for a fully automated market research solution to streamline the process and incorporate new data sources as part of its customer insights activities. The customer insights managers at demandihelp organizations to maximize the value of business information by creating 360-degree customer views. These views combine insights from enterprise resource management (ERP) platforms, customer relationship management (CRM) solutions and satisfaction surveys.

Jens Adams, the CEO of demandi described the challenges: “We needed to be able to ingest data from web analytics and cloud-based platforms, such as Salesforce. We also wanted to improve our visualization capabilities, so we could present research findings in different formats according to client preferences,”

demandi began by implementing the open-source version of Pentaho but found it difficult as their only source of knowledge was from internet articles and e-books. But everything changed when they met the Hitachi Vantara team at a conference. Shortly afterwards, in 2017, demandi deployed the enterprise version of Pentaho Data Integration (PDI). “Hitachi Vantara was so enthusiastic about how Pentaho could help transform our business. We felt that the team understood our challenges and we knew it was the right solution for us. With Hitachi Vantara, we can tap into powerful data extraction, transformation and loading capabilities,” says Adam.

Hitachi Vantara’s extensive data management expertise helped to accelerate and optimize the deployment and configuration of the new solution. “The best practices shared by the Hitachi Vantara team were invaluable. They helped us establish a stable operational environment and maximize the ‘drag and drop’ data management features of Pentaho.” Hosted on Amazon Web Services (AWS), demandi’s infrastructure made good use of Pentaho’s rich library of prebuilt connectors during implementation. “The Amazon Redshift and S3 connectors gave us a head start on integrating the new solution with our cloud environment,” says Adam. “Pentaho is a very open system, which is extremely important for us.”

This was a classic example of a DataOps implementation effort. DataOps, at a high level, is the process of delivering the right data to the right place at the right time. There are many tools available to deliver DataOps processes. In fact, there are often too many options, which can easily add to confusion and unnecessary delays and restarts. The DataOps advantage that Hitachi Vantara delivered was the data management expertise that our personnel were able to deliver and the tried and tested library of prebuilt connectors in the enterprise version of Pentaho.

Another DataOps advantage that was cited by Mr. Adams was Pentaho’s unique Metadata injection capability. Metadata injection streamlines the initial implementation and accelerates the onboarding of new clients and the loading of new data. For example, you might have a simple transformation to load transaction data values from a supplier, filter specific values, and output them to a file. If you have more than one supplier, you would need to run this simple transformation for each supplier. Yet, with metadata injection, you can expand this simple repetitive transformation by inserting metadata from another transformation that contains the ETL Metadata Injection step. This step coordinates the data values from the various inputs through the metadata you define. This process reduces the need for you to adjust and run the repetitive transformation for each specific input.

The Hitachi Vantara DataOps advantage has helped demandi differentiate its services in an industry where many companies are only just starting to think about new technologies. For example, it can now use intelligence captured in Pentaho to help clients track trends and real-time key performance indicators (KPIs). With this capability, the company can gain insights into key business processes and functions, such as product returns, purchase volumes and customer satisfaction.

“We’ve transformed how we manage data and how we take our services to market,” says Adam. “With smarter data integration and ingestion, demandi and Interrogare can position themselves as a disruptor in the market research industry.”

Please click on this demandi link to see their DataOps story.

The AI Revolution Requires Accelerated Compute

community-noreply@hitachivantara.com — Wed, 24 Jul 2019 18:36:30 GMT

We are in the midst of a new technology revolution that is described as the AI revolution. It is different from all previous revolutions like the industrial revolution or the information revolution in that it is not based on improving our human efforts based on explicit human knowledge. It goes beyond that by provided machines with the ability to learn and develop tacit knowledge, the intuitive know how, that is in the human brain. The AI revolution can provide super human artificial intelligence that could provide incalculable benefits for society.

AI or Machine Learning requires the computation of an enormous amount of data and is very compute intensive. One of the limitations for AI and Machine Learning is the limitations of today’s computers. CPUs are built on the Von Neumann architecture which connects a processing unit to a memory over a bus. The Von Neumann architecture can only process one instruction at a time in sequence. The speed and capabilities of this type of processor used to double every two years as researchers packed more transistors on to a microchip per Moore’s Law. Unfortunately, this is no longer the case and companies are scrambling to find different ways to improve processing speeds to support the AI revolution.

One approach that is gaining a attention today is through “Accelerated Computing” (AC). IDC defines AC as the practice of offloading key workloads to silicon subsystems like high-speed GPUs (Graphics Processing Unit) and low latency FPGAs (Field Programmable Gate Arrays). These multi-chip configurations are increasingly targeting the unstructured data workloads leveraged by artificial intelligence, advanced data analytics, cloud computing and scientific research.

CPUs, GPUs and FPGA process tasks in different ways: A typical CPU is optimized for sequential serial processing and is designed to maximize the performance of a single task within a job, like transaction processing. GPUs, on the other hand, use a massively parallel architecture aimed at handling multiple functions at the same time. As a result, GPUs are 50 to 100 times faster than CPUs in tasks that require multiple parallel processes such as machine learning and big data analysis. While CPUs and GPUs execute software, FPGAs are hardware implementations of algorithms, and hardware is always faster than software. However, FPGAs do not handle floating point which is used for intensive signal- and image-processing applications. While FPGAs can be reprogrammed, it requires a special hardware description language which differs from normal programming languages in that they are able to accommodate parameters including propagation delays and also signal strengths.

All three types of processors could be used in combination. The FPGA could forward incoming data at high speeds, while the GPU would handle the heavy algorithmic work. CPUs would play a management role, interpreting the results of the GPU and sending the “answer” to the user. Such a combined system would play to the strengths of each type of processor while maximizing system efficiency. Since the FPGA would have fewer responsibilities, it could be smaller and less difficult to design and therefore cheaper and faster to implement. Accelerated Computing can create powerful compute engines out of standard CPUs, GPUs and FPGAs.

Hitachi has been using Accelerated Compute for some time, since the introduction of the HNAS, high performance NAS controller, over a decade ago. HNAS combines the use of FPGAs to accelerate data movement while a CPU handles the data management. (refer to my previous blog post Solving The Von Neumann Bottleneck With FPGAs).

Last year we introduced an Accelerated Compute model of our Unified Compute Processor, the Hitachi Advanced Server DS225. The Hitachi Advanced Server DS225 delivers unparalleled compute density and efficiency, to meet the needs of the most demanding high-performance applications in the data center. DS225 takes full advantage of the ground-breaking Intel Xeon Scalable Processor family in combination with NVIDIA Tesla GPUs. By combining the Intel processors with up to four dual-width 300W graphic accelerator cards and up to 3TB memory capacity in a 2U rack space package, this server stands ready to address the challenging compute demands of the AI Revolution.

Super Computers are also turning to accelerated compute. In June of 2018, the Summit computer at the United States Department of Energy's Oak Ridge National Laboratory (ORNL) topped the supercomputing list with a sustained theoretical performance of 122.3 petaflops on the High Performance Linpack test used to rank the Top500 supercomputing list. This surpassed the Sunway TaihuLight system at the National Supercomputing Center in Wuxi, China, which is capable of 93.01 petaflops.

Unlike earlier supercomputers, the Summit Computer uses standard components and software. designed by IBM and NVIDIA. Summit has a 4,608 node hybrid architecture, where each node contains multiple IBM POWER9 CPUs (2/Node) and NVIDIA Volta GPUs (6/Node) all connected together with NVIDIA’s high-speed NVLink. Each node has over half a terabyte of coherent memory (high bandwidth memory + DDR4) addressable by all CPUs and GPUs plus 800GB of non-volatile RAM that can be used as a burst buffer or as extended memory. To provide a high rate of I/O throughput, the nodes are connected in a non-blocking fat-tree using a dual-rail Mellanox EDR InfiniBand interconnect. The operating system is Red Hat Enterprise Linux (RHEL) version 7.5.

Supercomputers with accelerated computing are breaking the boundaries around many sciences. The Summit computer will be used in several studies, including the following:

Astrophysics: With 100 time more compute power than was previously available scientist will be able to build higher resolution models to study things like super novas for clues on how heavy metals were seeded in the universe.

Materials: Studying the behavior of sub-atomic particles to develop new materials for energy storage, conversion and production.

Cancer Surveillance: Acomprehensive view of the U.S. cancer population at a level of detail typically obtained only for clinical trial patients.This will help to uncoverhidden relationships between disease factors such as genes, biological markers and environment.

Systems Biology: Using a mix of AI techniques researchers will be able to identify patterns in the function, cooperation and evolution of human proteins and cellular systems. These patterns can collectively give rise to clinical phenotypes, observable traits of diseases such as Alzheimer’s, heart disease or addiction, and inform the drug discovery process.

Accelerated Computing will become more ubiquitous as demand for AI and machine learning continues to increase. Accelerated Computing will need to fill the demand for intensive compute power until Quantum computers become available for commercial use. When that happens, Artificial Intelligence will take an exponential step forward providing social innovations which will vastly improve our lives and society.

Reduce Shrinkage with Live Face Matching

community-noreply@hitachivantara.com — Tue, 23 Jul 2019 21:04:30 GMT

This week I had a chance to meet with Chris Henderson a Biometric specialist who recently joined Hitachi Vantara in Australia. Chris was no stranger since we had worked together in the past and he has been working on some biometric projects for Hitachi Australia around Finger Vein identification and Live Face Matching (LFM) systems as a contractor.

Chris has been working on a Live Face Matching system with a large retailer in Australia to reduce shrinkage using video analytic technology from Hitachi. Shrinkage is a term used in the retail industry for the discrepancy between the dollar amount of the book inventory and the physical inventory in the store. This discrepancy is caused by errors, spoilage and theft. The biggest cause for shrinkage is shoplifting, followed by employee theft. In Australia, shrinkage amounted to a loss of $4.7 USD billion last year with specialty store like electronics, hit the hardest. I looked up the numbers for the United States where Forbes reported $46.8 billionin losses last year. Shrinkage losses have a direct hit on profit and are a special concern in the retail industry where margins are razor thin.

Most retail stores have CCTV cameras in the store which are used after the fact to identify suspicious activity and prosecute shoplifters. Unfortunately, this does not prevent shop lifting activity. This is where a live face-matching system, can help. The Forbes article noted the following about shop lifting:

41% of retailers surveyed reported increases in overall inventory shrink.
The average cost per shoplifting incident doubled to $559.
60 percent of known shoplifters were detected entering at least two separate locations of the same retail chain.
20% 0f known shoplifters visited three or more locations of the same retail store.

A Live Face Matching system that is connected to all the stores locations can share the templates of know shoplifters and alert the security personnel as soon as they enter the store and allow security personnel to pre-empt the shoplifter, by providing closer attention or customer care. This is safer and less likely to cause an incident or risk of litigation, than confronting a shoplifter after the act.

A large store in Australia was dealing with frequent occurrences of antisocial behavior and theft. Retail theft was amounting to about 2% of turnover which is the same worldwide. Their network of security cameras was not deterring these incidents. The store deployed a solution that combined an advanced Live Face Matching system from Hitachi with advanced video data analytics and facial recognition technology. During the first three months the store achieved significant success in identifying and apprehending a number of repeat offenders News of this activity spread throughout the community which deterred other thieves.

With video systems, there are privacy concerns. Depending on the jurisdiction, a person visiting or entering a store automatically opts in to being videoed. A Face matching system does not store a picture. It stores data about the face like the distance between the eyes and creates a template of data points. The number of data points and the algorithms that are used are proprietary to the vendor and are stored encrypted. In that sense the data stored is anonymous. Templates of offenders who have been observed shop lifting in the past are entered in a data base and an alert is given when a match occurs as the offender enters the store. The data base is shared across the store’s locations, since offenders often target other locations of the same store where they are familiar with the store’s method of operations.

Live Face Matching from Hitachi analyzes live video to recognize registered individuals for security or operational purposes. Highly accurate and able to run on a variety of different camera feeds, LFM is a powerful tool for law enforcement, corporate security, identity-based operations, and customer services. High accuracy, frame rate capacity, and affordability make LFM a clear choice for your facial recognition solution.

Lessons from Apollo 11

community-noreply@hitachivantara.com — Wed, 17 Jul 2019 22:51:10 GMT

July 20^th, 1969, marks the 50^thanniversary of the lunar landing of Apollo 11. This brings back a lot of great memories for me as I was on the IBM team that was supporting the IT requirements for North American Rockwell, of Downey, California, the manufacturer of the Apollo Command and Service modules. When I was a senior at the University of California Berkeley, in 1961, I was inspired by President John F. Kennedy’s announcement of the dramatic and ambitious goal of sending an American safely to the Moon and back before the end of the decade. The fact that the goal was achieved in 9 years was nothing short of amazing since most of the technology to achieve this did not exist when the goal was stated.

I was in complete awe as I witnessed the video of the landing on my 17 inch Zenith television. To commemorate this event, I took pictures of this historic event on TV with my Kodak camera. This has become all the more amazing to me, over the years, considering the types of systems that the industry had to work with in those day.

I was hired by IBM in 1967, as one of the many recruits for the launch of the new family of mainframe computer systems called the System 360 (S/360). It was the first family of computers designed to cover the complete range of applications, from small to large, both commercial and scientific. Customers could purchase a smaller system with the knowledge they would always be able to migrate upward if their needs grew, without reprogramming of application software or replacing peripheral devices. S/360 was one of the technologies which helped and benefitted from this race to the Moon.

The largest S/360 was the model 91 which could do 16.6 million instructions per second and had up to 8 MB of memory. The state of the art storage system at that time was the 2314 which had removable disk packs. It provided a storage capacity of 29.2 MB per pack or 233 MB in the eight-pack facility and had a data rate of 310 KB/S. Compare that with the processing power and memory in your iPhone.

There will be many programs being aired this week to commemorate this achievement. However, you will not be able to see the video of the landing that I viewed on television back in 1969. NASA has lost the original moon walk tapes. A team of retired NASA employees and contractors tried to find the tapes in the early 2000s but were unable to do so. The researchers concluded that the tapes containing the raw unprocessed Apollo 11 SSTV signal were erased and reused by NASA in the early 1980s, following standard procedure at the time. This has given conspiracy theorists reason to doubt the validity of the Moon landing.

So much for data retention policies. However, I am not about to cast stones. When I went to search for the photos that I took 50 years ago to post in this blog, I could not find them in the shoe boxes in my closet. (If only I had an iphone back then to records these once in a life time events. I would be able to find them in the cloud)

This story has several lessons.

The most important is that we need to set bold goals to galvanize people into action and accelerate results
Have a data retention policy
Conspiracy theories occur when government agencies lose things
Clouds are better than shoe boxes for storing valuable documents.

The 5G Requirement for Data Storage

community-noreply@hitachivantara.com — Tue, 16 Jul 2019 23:29:40 GMT

This year we have seen some of communications service providers (CSP) like ATT, China Mobile, and Telefonica begin the rollout of 5G services. 5G stands for fifth-generation cellular wireless. 5G is fundamentally different than 4G, the previous generation. 5G tops out at 10 Gbps which is 100 times the theoretical top speed of 4G. Its network latency of 1 ms is almost 50 times lower than 4G; and it will support 1 million devices per sq km vs 2000 for 4G, with 99.999% availability of the network. 5G will generate data at an unprecedented velocity and volume and fuel a wide range of data-driven services and digital business models.

5G connectivity will provide seamless connectivity to sensors in virtually everything from heavy machinery to wearables for prescriptive maintenance, fraud detection, and security. Its ability to support a massive number of devices in a small area will enable smarter cities, factories, utilities, and smart agriculture. Ultra-reliable, low latency communications, will transform industries like critical infrastructure, autonomous vehicles, and real time healthcare. It will make it possible for surgeons to perform remote life-saving surgeries over 5G networks

5G is expected to enable billions of connected devices powering IoT. Statista.com estimates that there will be 75.44 billion connected devices by 2025, generating tons of data. While many applications like smart homes will be connected to the cloud, other applications that require real-time analysis and control of IoT devices will generate huge amounts of data that will be too large to transport, store, and analyze in the cloud in time to be useful. This demand for high volume, real time processing, can only be met by processing on the edge. Real time 5G IoT applications will require systems sitting on the edge, processing data and connected to backend repositories in the enterprise or cloud.

This requires two types of storage systems, one for the edge and one for the backend repositories.

On the edge, the storage system will need to be a low-cost, high performance storage systems that can manage very high volumes of data. In my previous post I described what Telefonica calls their 5G Storage use case. This 5G storage is direct attached storage that is managed by Hitachi Content Platform Anywhere Edge. The direct attached storage reduces, latency and cost, while the Hitachi Content Platform Edge provides a bottomless virtual storage capacity for the local file system.

For the backend repository, you can connect to a cloud if you can afford the latency of connecting to the cloud. If you cannot afford the latency you will need to connect to a scalable high performance storage system that can handle the massive amount of data that will be generated by 5G applications. The ingest or transfer of data from the edge to the core will require a high performance, storage system with scalable connectivity and bandwidth. Low latency solid state devices and NVMe will provide optimum performance. Capacities will need to scale to multiple petabytes with dedupe, compression and automated tiering to reduce cost. A lot of the data management functions within the storage controllers should be offloaded to FPGAs to reduce controller latencies and cost. Storage virtualization should be available to ensure seamless migration for large data stores.

The data demands, driven by 5G’s connected solutions, will require the need for high performance, high end storage systems.5G will be the catalyst for IoT solutions that will use data, networks, analytics, and reporting to create a smarter, healthier, safer society.