Information Technology Reference
In-Depth Information
investigation. For this reason, we have developed the HPCaaS model and
implemented its software libraries, which provide high-level functions for
obtaining HPC resource from IaaS clouds.
11.4.3 The HPCaaS Model: Providing an HPC Service
We have proposed an enhanced cloud service model by including the
HPCaaS layer (see Figure 11.3), to abstract HPC resources, including both the
hardware (networks, storage, and servers [physical/virtual]) and the software
(operating systems, middleware, and user-level HPC applications). This
implies that these HPC resources can be exposed to the cloud community
as service software libraries. These software libraries are (1) a set of virtual
machine images with prebuilt HPC applications and (2) a set of program
scripts that can be used to create, manage, and terminate a computer cluster.
They support accessing HPC applications and transferring data between
users and the computer cluster.
11.4.3.1 Providing an HPC Service Library for Amazon EC2
For each of the IaaS cloud service providers, the corresponding HPCaaS
services are grouped together and packaged as a service library. In this case,
the HPC service software library for Amazon EC2 is implemented. Currently,
we have provided its implementation on the Linux platform.
11.4.3.2 Management and Application Access Scripts
To implement the cluster management and HPC-application accessing
scripts of the HPC service software library, the EC2 command line tools
(Kay 2012) are used. The only dependencies are Perl and CURL, which are
normally included in many Linux distributions. Due to the use of the EC2
command line tools, the HPCaaS model is compatible with Amazon-like
clouds that support the same API. Examples of such clouds include private
clouds OpenStack (OpenStack Project 2012), Eucalyptus (Nurmi et al. 2009),
and research cloud NeCTAR (Kirby 2012).
Through the command line tools, we have implemented a set of bash shell
scripts to handle features such as HPC cluster creation and termination,
HPC-application job submission, and retrieval of results of HPC-application
execution (Wong and Goscinski 2012). A list of selected bash scripts is shown
in Table 11.1.
11.4.3.3 AMIs with Prebuilt HPC Applications
When accessing any HPC application in the proposed cloud framework,
an HPC cluster must be created and started on EC2, where two pieces of
information are required: (1) an AMI for creating instance(s) and (2) the
Search WWH ::
Custom Search