Watson Supercomputer with the Communication Network - Assignment Example

IBM Supercomputer, Watson Nanette Stanley CIS312 – Computer Architecture Strayer March 16 PROCESSORS – WATSON USES CLOSE TO 3,000 PROCESSORS TO EXECUTE INSTRUCTION AND PROCESS DATA: 1a. Explain how using 3,000 processors benefits Watson’s computing capabilities. Using many processors in computing will ensure that the machine supports more than one process and ensures those tasks are allocated among them. (Chan, Pacifici & Stadler 1996). Tasks like analysis of natural language, generating hypothesis, scoring and finding evidence, merging and ranking of the hypothesis is shared among the processors. This result to quick finding of solutions to given problem domain. Using that number of processors enabled the supercomputer process more than 500gigabyte of data per seconds. 1b. Determine whether or not doubling, tripling, or quadrupling the number of processors would have an effect. Explain why or why not. Having more than one processor enables the system to handle more than one thread. This will increase the performance of the system that is concurrently running applications. As the number of processors increases then applications with multiple threads will be processed faster. But the processors will be ideal and uneconomical if the application has only one thread (Ondergaard 2011). 2. RAM – WATSON USES 16 TERABYTES OF RANDOM ACCESS MEMORY: 2a. Determine whether large amounts of RAM helps Watson process data pages faster. Watson supercomputers have around 16 terabytes of RAM; this ensures that all information need for processing is stored there. Storing data in RAM reduces access time of data for processing making the machine faster. 2b. Predict the possible effects of doubling the amount of Watson’s RAM. Watson supercomputer will be faster if it’s RAM is double. Considering the kind of work done like knowledge representation, language processing, automated reasoning, machine learning and information retrieval. More RAM will ensure that all this processes are accommodated hence can be accessed fast by the processors, though the overall cost of the hardware will increase. 3. Networking – Watson uses a cluster of servers that requires a communication network. Explain the requirements for the communication network between the servers in terms of throughput, transmission speed, and protocols for reliability. The rate of chanelling message or data packets are delivered successfully over the channel of communication. It is usually measured in bits per second or data packets per time slot. Watson supercomputer used juniper switches with throughput of 10Gbps Ethernet. This enables it to send packets from one server to another at high speed for analysis. This made the machine have few seconds responding to question asked. Transmission speed is the rate at which data is transferred between the servers. Each server of Watson had a 10 Gb fiber optic which was connected to Juniper switch using flat layer2 vLAN. The fiber optic can transmit data at the rate of 1.05Petabits/s. The high transmission speed allows the data to be shared fast among the servers resulting to fast generation of hypotheses and gathering of massive data. (Chan, Pacifici & Stadler 1996). Protocols are system of rules for data exchange within or between computers. Watson computer used Network File System (NFS) protocol to access files from internal customized SONA I/O storage nodes(Ondergaard 2011). The protocol was used to allow data transfer among servers not connected to the internet. This enables the machine to use locally stored pages to answer questions asked by integrating different chunks of information to sensible meaning. 4. Disk storage – Watson analyzes a large number of pages per unit time to extract analytic information. Explain how Watson is able to store large amounts of data. Include the disk space requirements, the access times for reading and writing, throughput, and disk architecture. Watson consumed around four terabytes of disk storage which provided access to more than 200 million pages of both unstructured and structured content. The content was encyclopedia, newswire articles, dictionaries, literary work and many more. The information were obtained mainly from the internet, but during play time the information were stored locally in around 16 terabyte RAM this provided fast access of information for analysis. Around 3,000, POWER7 processor cores which had four threads per core gave Watson 80 Teraflop performance rate. This results to high reading rate from the servers when question was asked. It could process 500 gigabytes of data per second to generate hypothesis, analyze it after gathering enough data. High throughput of around 10Gbps enabled Watson to collect differently analyzed hypothesis from the processor for output at high speed. (Chan, Pacifici & Stadler 1996). The machine was made in such way that, at boot time, around 15 terabytes of information is loaded in the RAM. The information is accessible from all processors at high speed for logical processing. The output is then given inform of speech. ANALYZE THE SOFTWARE COMPONENTS USED TO BUILD WATSON. IN YOUR ANALYSIS ADDRESS THE FOLLOWING COMPONENTS: 5. IBM DeepQA – DeepQA is a massively parallel probabilistic evidence-based architecture that allows Watson to analyze, synthesize evidence, rank, and produce answers. Determine at least five challenges of integrating many components used in the DeepQA architecture (see the DeepQA architecture diagram at http://bit.ly/HGtesV): The components may provide answers that are similar and sometimes closely related or different answers. In case they of different answer it become difficult for the machine to choose from. This may lead to the wrong outcome given. Some components are more or less valuable at some stages of answers ranking, for example, at primary stage, ranking is done making components with minimal impact on this stage to be disproportionately important for giving finer distinction among high ranked components. The importance of the various components may be different for different question types; this may require extra data for the question types making it provide wrong output information. The components are heterogeneous as they are derived from the different variety of algorithms that were developed independently. (Chan, Pacifici & Stadler 1996). These components are not normalized; their distribution sometimes vary in accordance to the characteristics of the question asked. The components usually miss important parts that may not be fixed during training. There is also large imbalance in class. The system may find many answers from different components by searching using the text. However, these answers may be wrong. 6. SUSE Linux Enterprise Server 11 O / S – IBM decided to use this operating system instead of a Windows based or Apple OS X Lion. 6a. Explain the reasons for using SUSE Linux as the operating system for Watson. SUSE Linux supports so many third-party ISV applications. Watson was to run so many applications. e.g. use for generating hypotheses, speech analyses, ranking and many others making it a perfect OS(Ondergaard 2011). Most hardware also supported the OS. This made it suitable for operating Watson which were built from the integration of different hardware. It also supports different processor architectures like 32-bit, 64-bit and others which were used in the development of Watson. Since Watson was to obtain information from different sources, SUSE Linux could support physical, virtual and cloud environments. 6b. Determine if you would use a different operating system for Watson and explain why or why not. I find using SUSE Linux the best operating system to be used Watson. It interoperates with Window and UNIX. His make it easy for it to be integrated into mixed infrastructures, enabling them adds new applications as run existing one. SUSE Linux can be used with different processors unlike, for example, windows which needs either 64-bits or 32-bits processor. This will increase the portability of Watson, which will be applied in different fields. 7. Apache Hadoop framework – This framework allows for building applications for distributed processing. Determine the challenges of data accessibility to a clustered environment in terms of: 7a. Reliability –Some time the node may not analyze data correctly this leads to the wrong conclusion drawn from the data. Incorrect data analysis is normally, as result failure of system, server, OS. 7b. Scalability – Data in any analysis process is always growing as it allows it to be analyzed in real-time. This reduce the speed of analysis since as the data to be analyzed grows the cluster’ 7c. Data errors and failures – data sent to node for review may have errors, or the node itself fails to process the data. This reduces the speed of analysis. The challenge can be reduced by replicating the same data in different nodes to allow study made in case error occurs in the node. 7d. High-availability –some time the cluster node may fail. This reduces it availability and makes analysis slower. It is usually reduced by ensuring that the study is made on different nodes at the same time. 8. Apache UIMA framework – This is the Unstructured Information Management framework that enables applications to be decomposed into components. Explain the reasons for the Java-based UIMA framework in Watson, particularly when it comes to data flow between Watson components. UIMA allows decomposition of the application into components, which implements interfaces defined by it. The component also provides via XML files a self-describing metadata. They are usually written in either C++ or Java; the data can flows between the components. It also provides capabilities to wrap components of Watson as networks service. (Chan, Pacifici & Stadler 1996). Scaling to large volumes is made by replication process pipelined over network cluster nodes. This framework support running and configuring pipelines of Watsons’ Annotator components. The components actually do the work of analyzing information that is unstructured. These annotators are used to feed the Watson so as to improve it answering capability (Tsalolikhin 2012). The frameworks also have a simple server that accepts REST request and output annotation. The results are used for other web services which were required by the machine. 9. Watson has been applied to other environments such as the health care and communications ecosystems: 9a. Identify the components that allow Watson to be usable in these systems (i.e. exchangeability of components). Watson has larger information retrieving components. The information can be used to provide useful data. It is then used in decision making in application environment. It also has machine learning component. The components is able to build more knowledge, which is used in future to solve similar problem domain. Automated reasoning part make the application to reason and make the decision appropriately. 9b. Determine if Watson can be applied to any IT ecosystem. Explain why or why not. It is possible to use Watson in Information Technology field(Tsalolikhin 2012). It can be used to source information about different items to be used in building given software. Watson use can advise on budget, time and schedule that will maximize output while minimizing the cost in production of the software. 10. Watson applies natural language processing to solve problems and answer questions. Determine whether computer systems that support this type of processing can be considered to be equal or equivalent to humans’ ability to think. Natural language processing has machine learning capability just like human who learn from their environment. This allows the system to build it knowledge from past experience and use it to make future decisions. Machine learning algorithms take what is important from input data to build logical hypotheses. When a problem is presented the machine is in a position to “think” just like human and provide the right answers to the problem in accordance to the environment of the problem. This is done by collecting evidence from the hypotheses learned by the system in the past. When natural language improves in understanding the system learns from information stored. It then applies them in real world environment (Tsalolikhin 2012). Combining natural language with natural language generation the system becomes more and more like human by giving and receiving instruction given either by other system or human being. References Chan, M. C., Pacifici, G. & Stadler, R. (1996). Prototyping network architectures on a supercomputer. Yorktown Heights, N.Y.: IBM T.J. Watson Research Center. S\Ondergaard, A. S. (2011). Fra Jeopardy til sundheds-it: Om IBMs supercomputer Watson. K\Obenhavns Universitet. Tsalolikhin, A. (2012). System administration of the IBM Watson supercomputer. Linux Journal, 2012 (216), p. 1. Read More