Information Technology Reference
In-Depth Information
IV.
P
UBLIC
C
LOUD
S
ERVICES FOR
HPC
There are many public cloud computing services on offer. Many of those serve a variety
of tasks such as enabling users to: create hundreds, even thousands, of virtual machines (e.g.,
virtual servers) at the click of button; acquire unlimited space to save their data; access the
required amount of software functionality for office-related tasks (e.g., word processing,
spreadsheets etc.) and business-related operations (e.g., CRM). If fact, there is now a potential
to create one's entire IT infrastructure in the cloud. A Californian startup company by the
name of 3Tera has developed software, named AppLogic, that can automate the creation of
complex corporate systems. The software provides a simple graphical user interface which a
designer can use to drag and drop icons representing traditional components (e.g., databases,
routers, firewalls, cables etc.) onto a page in a Web browser. When the designer has the
arrangement in place, he or she can then click a button and the software creates the virtual IT
infrastructure in a matter of minutes (Carr, 2009).
Cloud computing for high performance computing (HPC) has not been on the priority list
of many cloud providers. However, this attitude is now changing as cloud providers have
begun to realize the business potential of this area of computing. Currently, there are few
cloud providers that can declare themselves as having cloud solutions for HPC. However, two
cloud providers, namely Amazon and Microsoft, have two cloud platforms that have the
potential to provide the high performance computing required by the scientific community.
The first product is Elastic Compute Cloud (EC2) from Amazon and the second is Azure
from Microsoft. Amazon's EC2 is part of Amazon Web Services (AWS). It offers a simple
web service interface that allows users to request, monitor, and manage any number of VM
instances running on physical compute nodes in their datacenters. The system therefore
enables users to scale capacity, both up and down, as their computing requirements change.
Compute nodes are organized into CPU capacity (several CPUs), RAM, and local storage.
The user can programmatically start and stop compute node instances to deal with increasing
or decreasing demand. EC2 is now increasingly been targeted by the scientific community for
HPC purposes and is proving to be useful (see The MathWorks, 2008; Rehr et al, 2008).
Azure is a complex beast. It has three components: Windows Azure, SQL Azure and
Azure AppFabric (formerly known as .Net Services). Windows Azure is designed to provide
developers with on-demand compute and storage to host, scale, and manage Internet or cloud
applications. SQL Azure is developed to extend the capabilities of SQL Server, Microsoft's
database managing system (DBMS), into the cloud as a Web-based distributed relational
database. AppFabric is a set of integrated technologies designed to help developers connect
applications and services between Windows Azure and on-premises deployments (see Foley,
2009; Bailey, 2009). Microsoft is devoting increasing sources behind this platform such as
adding more data centers. This has made Azure attractive for some HPC projects as
demonstrated by some of Microsoft's case studies (Microsoft, 2009a).
An increasing number of small HPC providers are now migrating some of their services
to the cloud. For example, SGI, an HPC solution provider, recently launched a cloud-based
platform named 'Cyclone'. Cyclone is expected to, initially, support five technical domains:
computational fluid dynamics, finite element analysis, computational chemistry and materials,
computational biology, and ontologies (Leung, 2010). Another small HPC solution provider
to enter the cloud market is Penguin Computing with its Penguin on Demand (POD). POD,
