Scientific Cloud Computing: Early Definition and Experience
Lizhe Wang, Jie Tao, Marcel Kunze
Institute for Scientific Computing, Research Center Karlsruhe
Hermann-von-Helmholtz-Platz 1, 76344 Eggenstein-Leopoldshafen, Germany
Alvaro Canales Castellanos, David Kramer, Wolfgang Karl
Department of Computer Science, University Karlsruhe (TH)
76128 Karlsruhe, Germany
Abstract
Cloud computing emerges as a new computing paradigm which aims to provide reliable, customized and QoS guar-anteed computing dynamic environments for end-users. This paper reviews recent advances of Cloud computing, identifies the concepts and characters of scientific Clouds, and finally presents an example of scientific Cloud for data centers.
1. Introduction
Cloud computing emerges as a hot topic from the late of 2007 due to its abilities of offering flexible dynamic IT infrastructures, QoS guaranteed computing environments and configurable software services. As reported in Google trends (Figure 1), Cloud computing (blue line), which is enabled by Virtualization technology (yellow line), has out-paced Grid computing [7] (red line).
Figure 1. Cloud computing in Google trends Currently numerous projects from industry and academia have been proposed, for example, RESER-VOIR project [23] - IBM and European Union joint research initiative for Cloud computing, Amazon Elastic
Compute Cloud [16], IBM’s Blue Cloud [14], scientific Cloud projects such as Nimbus [17] and Stratus [26], OpenNEbula [19].
There are still no widely accepted definition for Cloud computing albeit Cloud computing practice has attracted much attention. Several reasons has lead into this situation:
• Cloud computing involves researchers and engineers
from various backgrounds, e.g., Grid computing, soft-ware engineering, data storage. They work on Cloud computing from different viewpoints.
• Technologies which enable the Cloud computing are
still evolving and progressing, for example, Web 2.0 and SOA.
• Existing computing Clouds still lack large scale
de-ployment and usage, which would finally justify the concept of Cloud computing.
In this paper, we try to give an early definition of “Cloud computing” based on recent advances from academia and industry as well as our experience. This paper also in-troduces a proof-of-concept computing Cloud – Cumulus, which is deployed at our site. This paper is organized as follows. Section 2 introduces the current projects of Cloud computing. Section 3 defines the concept of Cloud comput-ing. Cumulus project, our experience of Cloud computing, is presented in Section 4. Section 5 concludes the paper.
2. Recent advances of Cloud computing
This section discusses several projects which are cur-rently devoted to Cloud computing.
2.1. Globus virtual workspace service and
Nimbus
A virtual workspace [10, 9] is the abstraction of an ex-ecution environment that can be made dynamically avail-The 10th IEEE International Conference on High Performance Computing and Communications
978-0-7695-3352-0/08 $25.00 © 2008 IEEE DOI 10.1109/HPCC.2008.38
825
The 10th IEEE International Conference on High Performance Computing and Communications
978-0-7695-3352-0/08 $25.00 © 2008 IEEE DOI 10.1109/HPCC.2008.38
825
The 10th IEEE International Conference on High Performance Computing and Communications
978-0-7695-3352-0/08 $25.00 © 2008 IEEE DOI 10.1109/HPCC.2008.38
825
able to authorized clients by using well-defined proto-cols. The abstraction captures resource quota assigned to such execution environments during deployment (such as CPU or memory) as well as software configuration aspects of the environment (such as operating system installation or provided services). The workspace service allows a Globus Toolkit client to dynamically deploy and manage workspaces.
The virtual workspace services consist of the following interfaces [10, 9]:
• The Workspace Factory Service has one operation
called “create”. Create has two required parameters: workspace metadata and a deployment request for that metadata.
• Once created, a workspace is represented as a WSRF
resource and can be inspected and managed through operations of the Workspace Service.
• The Group Service allows an authorized client to
man-age a group of workspaces as a whole.
• The Status Service offers the interface through which
a client can query the usage data the service has col-lected about it.
Based on Globus virtual workspace services, a cloudkit named Nimbus [17] has been developed to build scientific Clouds. With Nimbus client, users could:
• browse virtual machine images inside the cloud, • submit their own virtual machine images to the clouds, • deploy virtual machines, and
• query virtual machine status, and finally access the
vir-tual machine.
2.2. OpenNEbula
OpenNEbula (former GridHypervisor) is a virtual infras-tructure engine that enables the dynamic deployment and allocation of virtual machines in a pool of physical re-sources. OpenNEbula extends the benefits of virtualization platforms from a single physical resource to a pool of re-sources, decoupling the server not only from the physical infrastructure but also from the physical location [19].
OpenNEbula contains one frontend and multiple back-ends. The frontend provides users with access interfaces and management functions. The backends are installed on Xen servers, where Xen hypervisors are started and virtual machines could be backed. Communications between fron-tend and backends use SSH. OpenNEbula gives users a sin-gle access point to deploy virtual machines on a locally dis-tributed infrastructure.
Figure 2. OpenNebula architecture
2.3. Amazon Elastic Compute Cloud
Amazon Elastic Compute Cloud (EC2) [16] is a Web ser-vice that provides resizable compute capacity in the cloud. It is designed to make Web-scale computing easier for de-velopers. Amazon EC2’s simple Web service interface al-lows users to obtain and configure capacity with minimal friction. It provides users with complete control of ing resources and lets them use Amazon’s proven comput-ing environment. Amazon EC2 reduces the time required to obtain and boot new server instances to minutes, allow-ing users to quickly scale capacity, both up and down, as computing requirements change. Amazon EC2 changes the economics of computing by allowing users to pay only for capacity that they actually use.
With Amazon EC2 users can:
• create an Amazon Machine Image (AMI) containing
the applications, libraries, data and associated config-uration settings, or use Amazon’s pre-configured, tem-plated images to get up and running immediately;
• upload the AMI into Amazon Simple Storage Service
(S3). Amazon EC2 provides tools that make storing the AMI simple. Amazon S3 provides a safe, reliable and fast repository to store user’s images;
• use Amazon EC2 Web service to configure security
and network access;
• choose the type of instance users want to run;
• start, shutdown, and monitor as many instances of
user’s AMI as needed, using the web service APIs;
• pay for the CPU time and bandwidth that user actually
consume.
826 826 826
2.4. Discussion
We have studied the solutions for network configuration, data management, virtual machine infrastructure deploy-ment inside the cloud.
Nimbus & Globus virtual workspace provide three net-work configurations:
• public mode picks a public IP address from a pool for
virtual machine,
• private mode picks a private IP address from a pool for
virtual machine, and
• advisory mode gives a static IP address for virtual
ma-chine
The solutions are sometimes however beyond of some user scenarios. For example, a data center might employ a cen-tral DHCP service, which allocates dynamic IP addresses for all virtual machines. Globus virtual workspace in addi-tion requires to contact all the backends of the local infras-tructure. Sometimes a computer center might employ a lo-cal virtualization management system, like VMware Infras-tructure [29], to manage local hosting resources. It would pay off , in our viewpoint, that Globus virtual workspace talk with a local management system. The same scenario happens when Globus Toolkit works together with local re-source scheduler like OpenPBS [20], or Condor [12].
OpenNEbula employs NIS (Network Information Sys-tem) to manage a common user system and NFS (Network File System) for virtual machine image management. How-ever it has been widely recognized that NIS has a major security flaw: it leaves users’ password file accessible by anyone in the entire network. To employ OpenNEbula in professional way, it is better to merge OpenNEbula with some modern infrastructure solutions, e.g., LDAP [11] and Oracle Cluster File System [21].
3. Cloud computing: definition,
characteriza-tion and Enabling technologies
3.1. Functionalities
Computing clouds render users with services to access hardware, software, and data resource; Furthermore, some configurable integrated platforms for users could be sup-ported:
• HaaS: Hardware as a Service
Hardware as a Service was coined possibly at 2006. As the result of rapid advances in hardware virtualiza-tion, IT automavirtualiza-tion, and usage metering and pricing, users could buy IT hardware - or even an entire data
center/computer center - as a pay-as-you-go subscrip-tion service. The HaaS could be flexible, scalable and manageable to meet your needs [2].
• SaaS: Software as a Service
Software or application is hosted as a service and pro-vided to customers across the Internet. This mode eliminates the need to install and run the application on the customer’s local computer. SaaS therefore alle-viates the customer’s burden of software maintenance, and reduce the expense of software purchases by on-demand pricing.
• DaaS: Data as a Service
Data in various formats, from various sources, could be accessed via services to users on the network. Users could, for example, manipulate remote data just like operate on local disk; or access data in a semantic way on the Internet.
Based on the support of HaaS, SaaS, and DaaS, Cloud computing thereafter delivers Platform as a Service (PaaS) for users. Users thus can on-demand subscribe a computing platform with requirements of hardware configuration, soft-ware installation and data access demands. Figure 3 shows the relationship between above services.
PaaS
DaaS SaaS
HaaS
Figure 3. Cloud functionalities
3.2. Key features
Cloud computing distinguishes itself from other ing paradigms, like Grid computing [7], Global comput-ing [6], Internet Computcomput-ing [13], in followcomput-ing aspects:
• User-centric interfaces
Cloud services could be accessed with user-centric in-terfaces, which means:
– The Cloud interfaces do not force users to change their working habits, e.g., developing language, compiler, operating system, and so on.
– The Cloud client which is required to be in-stalled locally is lightweight, for example, Nim-bus Cloudkit client size is around 15MB. – Cloud interfaces are location independent and
can be accessed by some well established inter-faces like Web service and Internet browser.
827 827 827
• On-demand service provision
Computing Clouds provide resources and services for users on-demand. Users can customize required com-puting environments later on, for example, software installation, network configuration, as users normally own “root” privilege.
• QoS guaranteed offer
The computing environments provided by computing Clouds can guarantee QoS for users, e.g., hardware performance like CPU bandwidth and memory size.
• Autonomous System
The Computing Cloud is an autonomous system and managed transparently to users. Hardware, software and data inside clouds can be automatically reconfig-ured, orchestrated and consolidated to a single plat-form image, finally rendered to users.
3.3. Enabling technologies
A lot of enabling technologies contribute to the Cloud computing, here we identify several state-of-the-art tech-niques:
• Virtualization
Virtualization technologies multiplex hardware and thus provide flexible and scalable platforms. Vir-tual machine techniques, such as VMware [29] and Xen [1], offer virtualized IT-infrastructures on-demand. Virtual network advances, such as VPN [5], support users with a customized network environment to access cloud resources.
• Serviceflow and workflow orchestration
Computing clouds offer a complete set of service im-ages on-demand, which could be composed by ser-vices inside the Cloud. Cloud should be able to au-tomatically orchestrate services from different sources and of different types to form a serviceflow or work-flow for users.
• Web service and SOA
Cloud services are normally exposed as Web services, which follow industry standards, like WSDL [28], SOAP [24] and UDDI [18]. The services organization and orchestration inside clouds could be managed in a Service Oriented Architecture (SOA). A set of Cloud services furthermore can be included a SOA, make themselves available on various distributed platforms and can be accessed across networks.
• Web 2.0 and Mashup
Web 2.0 describes the trend in the use of World Wide Web technology and web design to enhance creativity,
information sharing, and, most notably, collaboration among users. These concepts have led to the devel-opment and evolution of Web-based communities and hosted services [4].
Mashup is a Web application that combines data from more than one source into a single integrated storage tool [3]. SmugMug [25] is an example of Mashup, which is a digital photo sharing website, allowing the upload of an unlimited number of photos for all ac-count types, providing a published API which allows programmers to create new functionality, and support-ing XML-based RSS and Atom feeds.
Hypervisor Xen Hypervisor Xen Hypervisor Xen OpenNEbula frontend Virtual Workspace Service
Globus VM Host VM Host VM VM Host VM VM SSH
Oracle File System
virtual network domain physical network domain Access point
Figure 4. Cumulus architecture • World-wide distributed storage system
A Cloud storage model should foresee:
– A network storage system, which is backed by distributed storage providers (e.g., data centers), offers storage capacity for users to lease. The data storage could be migrated, merged, and managed transparently to end users for whatever data formats. Examples are Google File Sys-tem [8] and Amazon Elastic Storage [16].
828 828 828
– A distributed data system which provides data sources accessed in a semantic way. Users could locate data sources in a large distributed environ-ment by the logical name instead of physical lo-cations. Virtual Data System (VDS) [27] could be good reference.
4. Cumulus: a scientific Cloud for data center
Cumulus is an on-going project of Cloud computing at our site. We design Cumulus in a layered architecture (see also Figure 4):
• Globus virtual workspace service resides on the access
point of Cumulus, accepts users’ requirements of vir-tual machine operation.
• The OpenNEbula works as Local Infrastructure
Vir-tualization Manager (LIVM). The frontend of Open-NEbula works on the Cumulus access point and get messages from Globus virtual workspace service.
• OpenNEbula frontend communicates to its backends
and Xen hypervisors on the hosts via SSH for virtual machine manipulation.
Virtual machines stay in a separate network domain. Network solution for virtual machines could be:
• Virtual machines could start with Xen virtual network
interface and configure like physical machine. For ex-ample, virtual machines might be configured with dy-namic IP addresses by listening to a center DHCP ser-vice in the network domain.
• Virtual network technologies could be used for virtual
machine network space. For example, VNET [15] ties virtual machines together efficiently and makes them appear to users.
To reach a productional quality, we build the local infras-tructure using IBM Bladecenter as backend and Oracle VM server [22] as operating system. All the hosts and virtual machines are backed by Oracle Cluster File System [21]. Virtual machine images and templates are stored in a Oracle File System, which could be mounted by all hosts. Applica-tion level software are also saved in the Oracle data server. Virtual machines could thus automatically mount software installation packages required by users. The results so far look promising.
5. Conclusion
This paper reviews the recent advances of Cloud comput-ing and presents our early definition of Cloud computcomput-ing, its interfaces, and its features. We also discuss our experience of building a scientific Cloud for a data center.
References
[1] P. Barham, B. Dragovic, K. Fraser, S. Hand, T. L. Har-ris, A. Ho, R. Neugebauer, I. Pratt, and A. Warfield. Xen and the art of virtualization. In Proceedings of
the 19th ACM Symposium on Operating Systems Prin-ciples, pages 164–177, New York, U. S. A., Oct. 2003.
[2] Here comes HaaS [URL].
http://www.roughtype.com/archives/2006/03/here comes haas.php access on June 2008.
[3] Web 2.0 definition [URL].
http://en.wikipedia.org/wiki/mashup (web application hybrid)/, access on June 2008.
[4] Web 2.0 definition [URL]. http://en.wikipedia.org/wiki/web 2/, access on June 2008.
[5] B. Gleeson etc. A framework for ip based virtual private networks. Rfc2764, The Internet Engineering Task Force, Feb. 2000.
[6] G. Fedak, C. Germain, V. N´eri, and F. Cappello. Xtremweb: A generic global computing system. In
Proceedings of the 1st IEEE International Symposium on Cluster Computing and the Grid, pages 582–587,
2001.
[7] I. Foster and C. Kesselman. The grid: blueprint for
a new computing infrastructure. Morgan Kaufmann,
1998.
[8] S. Ghemawat, H. Gobioff, and S. Leung. The google file system. In Proceedings of the 19th ACM
Sympo-sium on Operating Systems Principles, pages 29–43,
2003.
[9] K. Keahey, K. Doering, and I. Foster. From sandbox to playground: dynamic virtual environments in the grid. In Proceedings of the 5th International Workshop on
Grid Computing, pages 34–42, 2004.
[10] K. Keahey, I. Foster, T. Freeman, and X. Zhang. Vir-tual workspaces: achieving quality of service and quality of life in the grid. Scientific Programming, 13(4):265–275, 2005.
[11] V. A. Koutsonikola and A. Vakali. Ldap: Frame-work, practices, and trends. IEEE Internet Computing, 8(5):66–72, 2004.
[12] M. Litzkow, M. Livny, and M. W. Mutka. Condor -a hunter of idle workst-ations. In Proceedings of the
8th International Conference on Distributed Comput-ing Systems, pages 104–111, 1988.
829 829 829
[13] M. Milenkovic, S. H. Robinson, R. C. Knauerhase, D. Barkai, S. Garg, V. Tewari, T. A. Anderson, and M. Bowman. Toward internet distributed computing.
IEEE Computer, 36(5):38–46, 2003.
[14] IBM Blue Cloud project [URL]. http://www-03.ibm.com/press/us/en/pressrelease/22613.wss/, ac-cess on June 2008.
[15] A. I. Sundararaj and P. A. Dinda. Towards virtual net-works for virtual machine grid computing. In
Pro-ceedings of the 3rd Virtual Machine Research and Technology Symposium, pages 177–190, 2004.
[16] Amazon Elastic Compute Cloud [URL]. http://aws.amazon.com/ec2, access on Nov. 2007.
[17] Nimbus Project [URL].
http://workspace.globus.org/clouds/nimbus.html/, access on June 2008.
[18] OASIS UDDI Specification [URL]. http://www.oasis-open.org/committees/uddi-spec/doc/tcspecs.htm, ac-cess on June 2008.
[19] OpenNEbula Project [URL]. http://www.opennebula.org/, access on Apr. 2008. [20] OpenPBS [URL]. http://www.pbsgridworks.com/,
ac-cess on Nov. 2007.
[21] Oracle Cluster File System [URL]. http://oss.oracle.com/projects/ocfs/, access on June 2008.
[22] Oracle Virtual Machine [URL].
http://www.oracle.com/technologies/virtualization/index.html/, access on June 2008.
[23] Reservoir Project [URL]. http://www-03.ibm.com/press/us/en/pressrelease/23448.wss/, access on June 2008.
[24] Simple Object Access Protocol (SOAP) [URL]. http://www.w3.org/tr/soap/, access on Nov. 2007. [25] SmugMug [URL]. http://www.smugmug.com/, access
on June 2008.
[26] Status Project [URL]. http://www.acis.ufl.edu/vws/, access on June 2008.
[27] Virtual Data System [URL].
http://vds.uchicago.edu/, access on Nov. 2007. [28] Web Service Description Language (WSDL) [URL].
http://www.w3.org/tr/wsdl/, access on Nov. 2007. [29] VMware virtualization technology [URL].
http://www.vmware.com, access on Nov. 2007.
830 830 830