Microsoft is aiming to have its first cluster version of Windows ready in time for a supercomputing conference this fall.
Windows for supercomputers likely out by fall
About The Author
Eugenia Loli
Ex-programmer, ex-editor in chief at OSNews.com, now a visual artist/filmmaker.
Follow me on Twitter @EugeniaLoli
“That WAS and IS for your beloved NT !!!!”
So, no other operating system ever crashes? A BSOD is just a kernel panic.
“NT(now as XP) lacks good mem/resource support, just try open your NT machine, keep using it for all kinds work, you will see the mem leak problem”
This statement barely even parses… and isn’t correct.
Rather than attack every OS from every angle (as everyones else seems to be doing quite well), let’s have a look at clusters from a practical point of view for a second.
Clusters in Use
————————-
Today, clusters are deployed within academics and research, oil and gas companies, finance companies, life sciences companies, animation/movie production, engineering/manufacturing/defense and other areas.
According to reports and surveys, the average size of a cluster varies between 16 and 64 nodes (depending on report). The average cluster is a Dual P4 with 2Gb of RAM or thereabouts, has 100Mbit or Gigabit Ethernet. In short : The Top500 list of clusters are not representing the vast majority of clusters in R&D and industry.
Cluster Software
————————-
From my own companies experience (we’re a developer of software for clusters, mainly being used in oil and gas but also life sciences), there is a steadily growing list of off the shelf software, tailored for clusters. This is true for industries such as engineering/manufacturing/defense, but also high performance visualization (our industry), finance and oil and gas (again, our industry). Oil companies rely on fairly huge clusters for Prestack Depth Migration, but other than that there’s a growing list of software that’s being run on clusters also within oil and gas.
Cluster Software Development
———————-
A handful of software companies are offering development tools, languages and full environments for development and deployment of software for clusters. Companies such as our own offers toolkits that allows transparent development and deployment on standalone and cluster environments (in other words, use your $2000 laptop for development, then scale up on your cluster). Companies such as Platform Computing, ArcSoft and others are very well positioned if and when a Windows HPC platform becomes usable and attractive.
GeoPolitical
———————-
In many parts of the world, Linux knowledge and experience is scarce. Windows knowledge, on the other hand, is ‘abundent’ and easily available. A number of companies (and governments) acknowledge this, and for that reason welcomes Microsoft to the high performance computing arena. This despite the market monopoly Western countries are accusing Microsoft of. In short, whereas Linus offers in terms of customization (every country’s their own OS at little or no cost), Microsoft ‘counters’ in terms of market acceptance at a significant cost.
Conclusion : There isn’t one
————————
Interestingly enough, there is no such thing as a conclusion on this issue. There’s, however, an evolution. For everyone who has followed the ‘Beowulf’ grow and prosper beyond the initial goals, it is now clear that the cluster industry is a multi billion one, regardless of ‘target industry and use’.
In that context, there is little evidence that a Microsoft HPC edition – when it debuts – will not be a popular alternative in most, if not all, the industries and areas I have mentioned.
Put your money where your mouth is
————————–
Our company has been asked, and we have approached Microsoft, to supply our software for Microsoft HPC clusters. We welcome that with open arms, not because Linux isn’t up for the task, not because Windows is *more* up to it, but simply for the following reason: We would like to see a proper alternative to Linux in this space. More choice means more competition means better solutions.
That said, there are issues that Microsoft HPC Server will have to do well for it to ‘survive’ in a cluster world. These are partly related to performance (network, threading, memory management), but more importantly related to management.
So, to both sides of this: Start writing your own software for clusters, make sure they run on both OS’es, and we’ll talk in two years time to sum up our experience.
Good luck.
CEO