yup. It pre-existed it by over a year. They do compliment each other. Closer binding in the press release is an artifact of too many PR people in the mix.
Our UD arrangement was (and is) an exceptional deal that likely will never happen again. That helps with the bottom line.
As to markets for these cycles, it is hard to tell. We certainly went down this path for a while and there are un-announced projects that will extend this idea very very far. I don't think that at the moment there is a market, but the problems are technological (security being the big one -- if you're a company doing something secret -- drugs or security; you need absolute assurance that the distributed environment doesn't leak information. None of the distributed environments that I know about are there yet.)
As to the unpredicability of distributed cycles, this is one of those problems that the central limit theorm comes to play. The average available horsepower on any of these large distributed systems is pretty static (sans stupid things like 9/11).
Write me at Purdue and I'll be glad to discuss this further. I can be found through directory [dot] purdue [dot] edu.
David Moffett, AVP Research Computing, Purdue University.
Second, the hardware is only in Research Computing for a maximum of two years.
Third, we did a small study about what it would save to de-power the CDROM, Zip, Sound Card and floppy and found that the cost of an undergraduate doing that was far more than the $0.08-$0.12 that disconnecting these would save (both consumed and A/C used) over the 2 years that we would have them.
Good question; one we certainly thought about and we used a couple of people for an afternoon answering for all sorts of configurations.
David Moffett, AVP Research Computing, Purdue University.
It is currently RedHat 7.3 (with patches, etc). We're talking actively about what to do next since that is about to leave active support. RH's big push to put all new cool features in their less-free Advanced Server will price us out that environment.
Debian and Suse are both under active consideration with trial clusters being brought up using each. In late January, visit our web site and find out which way we went.
David Moffett, AVP Research Computing, Purdue University
The blend of resources is a result of the PR engine that all this material is filtered through. Sorry for the confusion.
Recycled Clusters, aka 'Scrap Metal', are all running Linux. They will run Linux or one of the BSDs for the forseeeable future. The raw economics of having a thousand of anything running something that is not Free gets horrible really fast.
The Suns, on the other hand, run Solaris 9. The Starfire interconnect support software for Linux is a ways off!;-)
What ties them all together is a common job queueing system and many of the same application codes.
David Moffett, AVP Research Computing, Purdue University
First the Virgina Apple Cluster will fight an uphill battle to get research codes ported to the G5/MacOS X platform. Purdue's Research Computing doesn't have the staff that doing such a large porting effort would require. We are friends with Apple and talk with them regularly -- I am by no means slamming them. We have a 10 system cluster of Apple XServes within Research Computing and are working with them on products that have not yet been annouced.
Secondly, all these recycled clusters and the Sun gifts total well below $1m. If you'd like to give Purdue the $5m-$6m that the Apples cost (plus some salary money to help with porting) I'd be glad to talk with you directly. So far, I simply don't have the money that buying and owning such a cluster would require.
We are working toward having a very diverse collection of resources that meet different computational needs. The 5 machine Sun F6800 cluster with each machine having 24 CPUs and 192GB of RAM being the new top and thousands of machines under United Devices management being the bottom with many kinds of resources in the middle.
David Moffett, AVP Research Computing, Purdue University.
havent they heard of a little device called an AM radio?
Yup, we have. We've also heard of FM radio where the game is locally broadcast on
WAZY.
Alas, broadcast radio is a one way medium. PDAs are two-way.
We're "cooking" up things that your transistor radio couldn't dream of... like allowing you to order food from your seat. "I'd like my wings with EXTRA hot sauce" and you won't miss a play while it is delivered to you.
Football in the modern era is about convience and positive user experiences. PDAs are just one more way that we can make the experience a novel, exceptional and over-all pleasant time. Along the way, we're learning a lot about 802.11 performance in large venues, how safety & security can use this technology to enhance safety and how to deploy such resources quickly & get them back just as quickly.
David Moffett, AVP Research Computing, Purdue University
A rendering job is submitted to
PBS just like all the thousands of other high performance computing jobs that research computing processes on a typical day. It is scheduled with a normal scheduler (like
Maui) and managed like a normal research job. And yes, it has a priority that can be bumped up or down.
The blending of resources between research and instruction is just one of the outcomes of the High Performance Classroom. The local paper wrote this: Powerful gifts make tough questions easier to answer. Most universities keep their Instruction and Research separated. We at Purdue don't do that.
David Moffett, AVP Research Computing, Purdue University
Net is that we ARE going to use the Dells. Purdue has partnered with
United Devices to capture the unused cycles in the 2100ish machines in the 60 instructional labs. The UD install team was at Purdue 10/30 & 10/31 working with research computing to do just that.
Bigger picture this is about leaving no cycle un-consumed when we're in need of cycles. It is clear that no matter how large Research Computing's budget is, there will be needs that exceed the supply. Recycling, UD, Gifts, etc are some of the ways we're going after the chronic under-supply of resources.
David Moffett, AVP Research Computing, Purdue University
What the High Performance Classroom is about (in part) is not wasting the time of the students taking the class. They can do Hello World in the first 45 minutes, not the first 3 hrs and move on to more real things. Visualization is important NOW because the datasets are growing faster than anyone can comprehend the numbers.
Beyond that, the HPC is about teaching new skills to a data centric way of doing science and engineering.
David Moffett, AVP Research Computing, Purdue University
Slashdot Mirror
David
Our UD arrangement was (and is) an exceptional deal that likely will never happen again. That helps with the bottom line.
As to markets for these cycles, it is hard to tell. We certainly went down this path for a while and there are un-announced projects that will extend this idea very very far. I don't think that at the moment there is a market, but the problems are technological (security being the big one -- if you're a company doing something secret -- drugs or security; you need absolute assurance that the distributed environment doesn't leak information. None of the distributed environments that I know about are there yet.)
As to the unpredicability of distributed cycles, this is one of those problems that the central limit theorm comes to play. The average available horsepower on any of these large distributed systems is pretty static (sans stupid things like 9/11).
Write me at Purdue and I'll be glad to discuss this further. I can be found through directory [dot] purdue [dot] edu.
David Moffett, AVP Research Computing, Purdue University.
Second, the hardware is only in Research Computing for a maximum of two years.
Third, we did a small study about what it would save to de-power the CDROM, Zip, Sound Card and floppy and found that the cost of an undergraduate doing that was far more than the $0.08-$0.12 that disconnecting these would save (both consumed and A/C used) over the 2 years that we would have them.
Good question; one we certainly thought about and we used a couple of people for an afternoon answering for all sorts of configurations.
David Moffett, AVP Research Computing, Purdue University.
Debian and Suse are both under active consideration with trial clusters being brought up using each. In late January, visit our web site and find out which way we went.
David Moffett, AVP Research Computing, Purdue University
Recycled Clusters, aka 'Scrap Metal', are all running Linux. They will run Linux or one of the BSDs for the forseeeable future. The raw economics of having a thousand of anything running something that is not Free gets horrible really fast.
The Suns, on the other hand, run Solaris 9. The Starfire interconnect support software for Linux is a ways off! ;-)
What ties them all together is a common job queueing system and many of the same application codes.
David Moffett, AVP Research Computing, Purdue University
First the Virgina Apple Cluster will fight an uphill battle to get research codes ported to the G5/MacOS X platform. Purdue's Research Computing doesn't have the staff that doing such a large porting effort would require. We are friends with Apple and talk with them regularly -- I am by no means slamming them. We have a 10 system cluster of Apple XServes within Research Computing and are working with them on products that have not yet been annouced.
Secondly, all these recycled clusters and the Sun gifts total well below $1m. If you'd like to give Purdue the $5m-$6m that the Apples cost (plus some salary money to help with porting) I'd be glad to talk with you directly. So far, I simply don't have the money that buying and owning such a cluster would require.
We are working toward having a very diverse collection of resources that meet different computational needs. The 5 machine Sun F6800 cluster with each machine having 24 CPUs and 192GB of RAM being the new top and thousands of machines under United Devices management being the bottom with many kinds of resources in the middle.
David Moffett, AVP Research Computing, Purdue University.
Yup, we have. We've also heard of FM radio where the game is locally broadcast on WAZY.
Alas, broadcast radio is a one way medium. PDAs are two-way.
We're "cooking" up things that your transistor radio couldn't dream of... like allowing you to order food from your seat. "I'd like my wings with EXTRA hot sauce" and you won't miss a play while it is delivered to you.
Football in the modern era is about convience and positive user experiences. PDAs are just one more way that we can make the experience a novel, exceptional and over-all pleasant time. Along the way, we're learning a lot about 802.11 performance in large venues, how safety & security can use this technology to enhance safety and how to deploy such resources quickly & get them back just as quickly.
David Moffett, AVP Research Computing, Purdue University
A rendering job is submitted to PBS just like all the thousands of other high performance computing jobs that research computing processes on a typical day. It is scheduled with a normal scheduler (like Maui) and managed like a normal research job. And yes, it has a priority that can be bumped up or down.
The blending of resources between research and instruction is just one of the outcomes of the High Performance Classroom. The local paper wrote this: Powerful gifts make tough questions easier to answer. Most universities keep their Instruction and Research separated. We at Purdue don't do that.
David Moffett, AVP Research Computing, Purdue University
Good Question!
Net is that we ARE going to use the Dells. Purdue has partnered with United Devices to capture the unused cycles in the 2100ish machines in the 60 instructional labs. The UD install team was at Purdue 10/30 & 10/31 working with research computing to do just that.
Bigger picture this is about leaving no cycle un-consumed when we're in need of cycles. It is clear that no matter how large Research Computing's budget is, there will be needs that exceed the supply. Recycling, UD, Gifts, etc are some of the ways we're going after the chronic under-supply of resources.
David Moffett, AVP Research Computing, Purdue University
everyone (including you) must start somewhere.
What the High Performance Classroom is about (in part) is not wasting the time of the students taking the class. They can do Hello World in the first 45 minutes, not the first 3 hrs and move on to more real things. Visualization is important NOW because the datasets are growing faster than anyone can comprehend the numbers.
Beyond that, the HPC is about teaching new skills to a data centric way of doing science and engineering.
David Moffett, AVP Research Computing, Purdue University