Cloud Computing for High-Performance Computing has been researched and tested by various individual and non-it universities as it is technological outsourcing. Cloud computing and information technology itself, such as change the understanding of IT as a service, is in use for a long term. Organizations relates the cloud computing as the possibility of rapid provisioning of IT processes which are flexible, scalable, on-demand as a service. Many start-ups and established organizations already use the dynamic processing power and storage capacity of the cloud, without having to make large investments. But what is the potential lies in the cloud services for complex, data-intensive calculations for science and research ? Can organizations use the services of commercial cloud providers for high-performance computing ? Particularly in HPC often complex and costly infrastructure are necessary. can we can replace it with a service ?
Cloud Computing for High-Performance Computing : Abbreviations
- AMI : Amazon Machine Image
- API : Application Programming Interface
- CCI : Cluster Compute Instances
- CPU : Central Processing Unit
- CRM : Customer Relationship Management
- DSM : Distributed Memory System
- EBS : Elastic Block Store
- EC2 : Elastic Compute Cloud
- ERP : Enterprise Resource Planning
- GFLOPS : Giga Floating Point Operations Per Second
- GPU : Graphics Processing Unit
- HPC : High Performance Computing
- HPCC High Performance Cloud Computing
- IaaS : Infrastructure as a Service
- IPC : Inter Process Communication
- MPI : Message Passing Interface
- MPP : Massive Parallel Processors
- NIST : National Institute of Standards and Technology
- PaaS : Platform as a Service
- PFLOPS : Peta Floating Point Operations Per Second
- PUE : Power Usage Effectiveness
- SaaS : Software as a Service
- SLA : Service Level Agreement
- TCO : Total Cost of Ownership
Cloud Computing for High-Performance Computing : Introduction
Cloud computing and information technology itself, such as change the understanding of IT as a service, is of long term. Organizations relate cloud as the possibility of next flexible IT resource which is scalable and can be used on-demand as a service. Many start-ups and established organizations already use the dynamic processing power and storage capacity of the cloud, without having to make large investments. The aim of this article is to discuss the application potential of cloud computing for high-performance applications and evaluate. First an overview of the different types of cloud services was given, followed by an introduction to high-performance computing. Based on the technical demands of HPC applications, and also the organizational and legal requirements, problems and risks as well as two current implementations briefly presented and subsequently discussed with the potential of cloud computing for HPC applications.
Cloud Computing for High-Performance Computing
The concept of high-performance computing, due to the constantly increasing demand of computing power is not precisely defined. There is no definition on how many operations per second a computer system should perform as a “high performance” machine. Usually the number for multi-processor systems (within ten, hundreds, thousands as well) was used, which are connected to increase performance over a network. The networked operation of multiple processors is often referred to as “parallel computing”.
Typical HPC systems are as defined in networked computing systems. These connected computer units together form a usable computing power, goes beyond the capabilities of a single system. A calculation does not take place only on one physical processor with its associated memory, special it is distributed to as many connected processor units it requires. The parallel processing of a problem, requires that the distributed processes running and communicating one with another to access to a common memory area. This is in practice often implemented using the MPI libraries, which provide standardized methods for inter-process communication. Depending on the application resulting from the IPC, there are special requirements on the network infrastructure. Hundreds or thousands of processes interact with each other in communication, on the one hand, the available network bandwidth will be sufficient not to add the latency in message delivery. This would hinder the parallel processing, thereby reducing the performance of the HPC system.
Critical to the performance enhancement can be used by multiple distributed processors, which also demands the selection of appropriate algorithms, which need to solve a problem such as that performance which scales with the number of processors available.