Term paper on heterogenous computing

Term paper on heterogenous computing.
of 2
All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
Related Documents
  2012EE10452 Heterogenous Computing Gedela Vamsi Krishna 2012EE10452, Group-2 EEL308: Computer Architecture. Email: Abstract   —  Heterogeneous Computing is the use of different kinds of processors in the same machine. This paper discusses the need for Heterogeneous computing and some issues in using Heterogeneous computing. Keywords   —  Heterogeneous Computing, Heterogeneous Systems Architechture I.   I  NTRODUCTION We have got enough performance by using homogenous type of machines in the past in many applications. These applications have different kinds of  parallelism embedded in them like SIMD (Single Instruction, Multiple Data) and MIMD (Multiple Instruction, Multiple Data). But, now a days there is a need for more applications which have very diverse types of embedded parallelism in them. That means the  parallelism has a very less percentage of homogeneity. Homogeneous computing cannot give us the desired speedup anymore. This gives rise to the need of Heterogeneous computing (HC). Heterogeneous computing is the effective usage of a very diverse high performance machines which provide super speed computing. A heterogeneous system includes a combination of heterogeneous machines, communication protocols, programing environments, operating systems and interfaces. This combination results in a positive impact on performance. Heterogeneous computing is very different from network computing and there is a distinction between them. Network computing uses either an Ad-hoc connectivity or clusters of workstations. On the other hand Heterogeneous computing is a technique of solving  problems in computations that have embedded  parallelism. Since heterogeneous computing uses already  proven technology, it has much less risks. Heterogeneous Systems Architecture (HSA) exploits the usage of multiple types of processors like CPUs and GPUs, on the same silicon dye. This can give the user a  best result of both, GPUs which not only renders 3D graphics but also can do a very intense mathematical computation and CPUs which run the operating system and do other traditional tasks. II.   W HY H ETEROGENOUS C OMPUTING ?  A.    Limitations of Homogenous Systems There is continuous quest for higher performance in highly embedded parallelism applications at a reasonable cost. This has exposed several limitations of homogenous systems. Replacing the existing homogenous systems with much more powerful ones is not feasible. HC gives a cost effective solution by usage of the existing systems in an integrated environment. The conventional homogeneous systems specifically use one mode of parallelism, either SIMD, MIMD or vector processing in any given machine. Therefore if a computation requires more than one type of parallelism then the machine spends a lot of time in executing the code for which it is not suited for. Some applications like image understanding need to  process the information at more than one level concurrently but with different types of parallelism at each level. Different types of homogeneous systems have different limitations. For instance, if we consider vector machines, they have highly pipelined arithmetic and logic unit and an interleaved memory which leads to a high  performance in terms of Mflops (Million point operations  per second). The performance decreases very severely if these features are not exploited by the application and its resulting computations.  2012EE10452 Figure 1: Execution of a given code on different systems  [1] Suppose an application code, which has mixed types of parallelism is to be executed (figure 1). Assume this code takes 100 units of time on a serial computing machine. When executed on a vector machine it achieved desired speedup in the vector part of the code but the other part is not much affected by it. Now, when the code is run on a heterogeneous suite, since each part of the code runs the machine type best suited for it, the speedup is likely to be very high.  B.    An HC environment A heterogeneous computing environment contains the following components:    A set of heterogeneous machines,    An intelligent high speed network connection  between the machines    A user friendly programming environment Since all the components in a HC environment are heterogeneous, therefore autonomous, there is a likely improvement in overall system fault tolerance and longevity. III.   M ATCHING AND S CHEDULING  In a heterogeneous system dynamic schemes are necessary to implement mapping (matching and scheduling). Matching refers to the assignment of tasks to each processor and scheduling is computing the order of execution of tasks assigned to each of the processors. A dynamic scheme for both scheduling and matching is needed because the times of arrival of the tasks can be at random and hence some of the machines can go offline and new ones can come online at any given time. The dynamic mapping heuristics can be essentially classified into two types: Online mode and batch mode. A task gets mapped to a machine as soon as it arrives at the mapper in the online mode. Where as in batch mode, the tasks are collected into a set which is examined for mapping at times prior to scheduling, called mapping events. This set of tasks is called the meta task. This set includes the tasks which are newly arrived and also the tasks which are already mapped earlier but their execution did not begin yet. So, in the on-line mode, each task is considered for mapping only once but in batch mode a task is considered at each mapping event until its execution begins. C ONCLUSION  Heterogeneous computing allows for the most efficient use of hardware. It reduces the power consumption and increases the time of execution of computations which consist diverse kinds of parallelisms. A continuous research for building a heterogeneous ecosystem is being conducted by the HSA foundation which is founded by AMD, Samsung, ARM etc. A CKNOWLEDGEMENT  I hereby gratefully acknowledge Prof. Turbo Majumdar for giving me an opportunity to experience the reality of technical research and learning a new technology i.e. Heterogeneous Computing. R  EFERENCES   [1]   R.Fruend and D.Conwell, “Superconcurrency: A for  m of Distributed Heterogeneous Supercomputing,” Supercomputing review, Oct 1990, pp.47-50. [2]   Muthucumaru Maheswaran, Shoukat Ali, Howard Jay Siegel, Debra Hensgen, and Richard F. Freund, “Dynamic Matching and Scheduling of a Class of Independent Tasks onto H eterogeneous Computing Systems”   [3]   Website of HSA Foundation “”   [4]    Newsletter of the IEEE computer technical society technical committee on parellel processing. [5]   V.K. Prasanna Kumar, “Parellel Algorithms and Architechtures for Image U nderstanding”Academic Press, Boston, 1991.   [6]   Ashfaq A. Khokhar, Viktor K. Prasanna Kumar, Muhammad E. Shaaban, and Cho Li Wang, “ Heterogeneous Computing: Challenges and Opportunities.”   [7]   Ashfaq A. Khokhar et. al , “Heterogeneous Computing: Problems and Issues ,”  IEEE CS Press, Los Alamitos, Calif., Order No. 2167, 1991, pp 301-308. [8]   George Kyriazis, “ Heterogeneous System Architecture: A Technical Review” AMD, Rev 1.0, 30 th  Aug 2012. [9]   “ ”   [10]   “ -computing/what-is-heterogeneous- computing”  
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks

We need your sign to support Project to invent "SMART AND CONTROLLABLE REFLECTIVE BALLOONS" to cover the Sun and Save Our Earth.

More details...

Sign Now!

We are very appreciated for your Prompt Action!