Comparisons between some clusters nt compared to Linux
Performance of the NT cluster is quite good when compared to the Linux cluster [16]. The paper by Luecke, Raffin and Coyle of Iowa State University shows that the NT cluster can outperform the linux cluster (of equal processors) in the following tested areas:
Time taken for an MPI synchronization barrier,
Scalability due to some problems in the Linux software producing bad results,
Right shifting where each processor gets a message of size n from its neigh- bor,
A naive broadcast,
A binary tree broadcast,
and All to all MPI coma nd
As they concluded from these tests the NT cluster seems to be able to actually outperform that of the Linux cluster and hence should remove any doubts as to the ability of the NT operating system being able to handle clustering. The NT cluster used was one with the HPVM program allowing it to be clustered with up to 128 processors. This shows that the operating system implemented on the cluster can have an effect on the overall performance of that cluster. MacOS compared to Linux The MacOS has some benefits over the Linux operating system in that it has some third party software for numerical calculations that runs better on the Macintosh G3 than on the Unix workstations [13]. Another benefit is the amount of software that is available for the MacOS being quite large with programs like Microsoft Word that aren’t on Linux. The other main reason is that the Mac operating system is more user friendly and is easier to use than Linux for the novice user that might want to use the cluster. The Mac cluster is also low maintenance with only a single library and utility needed to run the cluster.
Linux is free and is therefore very suitable for low cost clusters. Another ben- efit is that for users that are already accustomed to the Unix operating system or even the Linux operating system, the use of Linux would be more suitable. Linux is also more adjustable to the wishes of the users. Linux clusters can run on different architectures and then make advantage of the specific hardware, where as the MacOS can’t.
Summary and Conclusions
Each operating system has its own benefits with the Linux operating systems main benefits are that it is free and that it can run on different architectures. This allows for cheap clusters and as Linux is the most common operating system that is used today in clusters, there is a good track record for its use. The Windows operating systems on the other hand, have ease of use as their main goal, with effort made in the making the setup of the clusters as simple as possible. The HPVM software allows the cheaper version of the Windows operating system, Windows NT, to be used in clustering. This makes the cost of this form of cluster a lot cheaper than it otherwise would be and also enables Windows NT to be used as the operating system in clusters that can rival that of supercomputers. The Mac OS is very easy to use for setting up a cluster and also provides great power associated with the Macintosh computers. The Solaris Operating system is aimed at providing simple effective clustering. The ultimate goal for this operating system is to provide an all in one system that caters for all uses of clusters.
In general the Linux operating system is useful for very cheap clusters, while the NT and Mac OS clusters are more expensive they provide great ease of use. The Solaris also has a Java based console that enables easy management of the cluster. The Solaris system is designed to be parallel with lots of communication between the nodes and this is where other systems can falter. The MacOS is suitable for small groups with limited resources because of its ease of use in setting up, use and management. The best operating system for a cluster is thus the one that is best suited to the users needs.
References
[1] High performance supercomputing at mail order prices. Web site. http://www.ncsa.uiuc.edu/access.html.
[2] Introduction to the cesdis beowulf project. http://www.beowulf.org/intro.html.
[3] Linux high avalability how-to. http://metalab.unc.edu/pub/Linux/ALPHA/linux-ha/High-Availability- HOWTO-5.html. 9
[4] Mosix application environments. http://www.mosix.org/txt what it can do.html.
[5] Site of other appleseed clusters. Web site. http://exodus.physics.ucla.edu/appleseed/appleseedsites.html.
[6] Sun clusters - a white paper. http://www.sun.com/software/white-papers/wp- sunclusters/sunclusterswp.pdf.
[7] What is mosix. http://www.mosix.org/txt what is.html.
[8] High Performance Cluster Computing: Architectures and Systems,. Pren- tice Hall PTR, 1999.
[9] Windows nt server enterprise edition — technical details clustering architecture. Web Site, April 12 1999. http://www.microsoft.com/ntserver/ntserverenterprise/techdetails/prodarch/clustarchit.asp.
[10] Introducing windows 2000 clustering technologies. Web site, February 2 2000. http://www.microsoft.com/WINDOWS2000/library/howitworks/cluster/introcluster.asp.
[11] Christa Anderson. Big clusters solve big problems. Windows NT Magazine US Print, June 1999.
[12] A. Chien, M. Lauria, R. Pennington, M. Showerman, G. Iannello, M. Buchanan, K. Connelly, L. Giannini, G. Koenig, S. Krishnamurthy, Q. Liu, S. Pakin, and G. Sampemane. Design and evaluation of an hpvm-based windows nt supercomputer. the international journal of high- performance computing applications. The International Journal of High- Performance Computing Applications, 13(3):201–219, Fall 1999.
[13] Victor K. Decyk, Dean E. Dauger, and Pieter R. Kokelaar. How to build an appleseed: A parallel macintosh cluster for numerically intensive com- puting. Web site and in Conference. http://exodus.physics.ucla.edu/appleseed/appleseed report.html and In- ternational Conference on Numerical Simulation of Plasma in Banff, Al- berta, Canada May 2000.
[14] Louis A. Giannini and Andrew A. Chien. A software architecture for global address space communication on clusters: Put/get on fast messages. In Proceedings of HPC-7 ’98.
[15] Roman Koniuk and Chris Stewart. The d’artagnan cluster. web site. http://www.students.yorku.ca/ kipper/dartagnan/dart.html.
[16] Glenn R. Luecke, Bruno Raffin, and James J. Coyle. Comparing the com- munication performance and scalability of a linux and an nt cluster of pcs, a sgi origin 2000, an ibm sp and a cray t3e-600,. The Journal of Perfor- mance Evaluation and Modelling for Computer Systems (PEMCS), March 2000.