Meaningful announcement, or
Locked
Meaningful announcement, or
 
|
political bullshit?
============================================================ FRIAM Applied Complexity Group listserv Meets Fridays 9a-11:30 at cafe at St. John's College lectures, archives, unsubscribe, maps at http://www.friam.org |
Re: [FRIAM] Meaningful announcement, or
|
|
Douglas Roberts wrote:
> political bullshit? > > http://www.kob.com/article/stories/S1381807.shtml?cat=500 > Curious to hear technical details... UNM and Encanto are apparently already connected by high speed fiber. http://newmexicosupercomputer.com/encanto4.html ..extending that to other colleges would be a big investment. ============================================================ FRIAM Applied Complexity Group listserv Meets Fridays 9a-11:30 at cafe at St. John's College lectures, archives, unsubscribe, maps at http://www.friam.org |
Re: [FRIAM] Meaningful announcement, or
|
Administrator
|
In reply to this post by Douglas Roberts-2
No, just computer architectural ignorance.
I think building a traditional "super computer" is likely an error. Maybe we could get Roger Fry and the Connection Machine gang to give it another shot. We should be thinking thousands of processors with at least mesh memory, if not hyper-cube. But even better would be gigabit networking or more to more than a thousand nodes within the state, each running "interesting architectures" including massive GPU farms. Most of the current thinking is your basic application server on steroids. Google will eventually beat all these efforts because they are thinking plumbing/networking with scalable data stores (NoSql). Just one smart node is not the way to go. State wide, easy access networking with massive, low latency storage is the first step. The institutions will add the surprises. -- Owen On Jan 24, 2010, at 9:19 PM, Douglas Roberts wrote: political bullshit? ============================================================ FRIAM Applied Complexity Group listserv Meets Fridays 9a-11:30 at cafe at St. John's College lectures, archives, unsubscribe, maps at http://www.friam.org |
Re: [FRIAM] Meaningful announcement, or
|
|
In reply to this post by Marcus G. Daniels
Depends. Being connected is one thing. Having a reasonable cost structure for bandwidth is another.
On Sun, Jan 24, 2010 at 9:55 PM, Marcus G. Daniels <[hidden email]> wrote:
-- Doug Roberts [hidden email] [hidden email] 505-455-7333 - Office 505-670-8195 - Cell ============================================================ FRIAM Applied Complexity Group listserv Meets Fridays 9a-11:30 at cafe at St. John's College lectures, archives, unsubscribe, maps at http://www.friam.org |
Re: [FRIAM] Meaningful announcement, or
|
|
In reply to this post by Owen Densmore
As long as we're dreaming, let's bring back Time Warp. One that works, this time.
--Doug
On Sun, Jan 24, 2010 at 10:01 PM, Owen Densmore <[hidden email]> wrote:
-- Doug Roberts [hidden email] [hidden email] 505-455-7333 - Office 505-670-8195 - Cell ============================================================ FRIAM Applied Complexity Group listserv Meets Fridays 9a-11:30 at cafe at St. John's College lectures, archives, unsubscribe, maps at http://www.friam.org |
Re: [FRIAM] Meaningful announcement, or
|
|
In reply to this post by Owen Densmore
Owen Densmore wrote:
> We should be thinking thousands of processors with at least mesh > memory, if not hyper-cube. That's fine if it is on a single piece of silicon or on the same board. Latency across sites will be on the order of milliseconds just from the speed of light. Compare to nanoseconds for intra-chip communication. Even state-of-the-art interconnects for installations like Encanto are still just on the order of microseconds within the facility. > Google will eventually beat all these efforts because they are > thinking plumbing/networking with scalable data stores (NoSql). Hmm, I think their application toolkits will get good at use cases where there are medium and high latencies to deal with along with medium individual bandwidths, e.g. JavaScript within a web browser and delays from communication over the internet. And they'll get better and better at managing millions of such workloads. That's completely different from high performance computing and scientific workloads. Marcus ============================================================ FRIAM Applied Complexity Group listserv Meets Fridays 9a-11:30 at cafe at St. John's College lectures, archives, unsubscribe, maps at http://www.friam.org |
Re: [FRIAM] Meaningful announcement, or
|
Administrator
|
On Jan 25, 2010, at 12:10 AM, Marcus G. Daniels wrote:
>> Google will eventually beat all these efforts because they are >> thinking plumbing/networking with scalable data stores (NoSql). > Hmm, I think their application toolkits will get good at use cases > where there are medium and high latencies to deal with along with > medium individual bandwidths, e.g. JavaScript within a web browser > and delays from communication over the internet. And they'll get > better and better at managing millions of such workloads. That's > completely different from high performance computing and scientific > workloads. Good point, the architecture for large scale computing is likely differ between application: scientific/engineering, visualization, large graph computations, simulations and so on. Which of us have an idea of specific "super computing" architectures? What would your "three wishes" be for such a system? Doug: I know your work with clusters would qualify. If you had a mega buck or two, how'd you spend it? -- Owen ============================================================ FRIAM Applied Complexity Group listserv Meets Fridays 9a-11:30 at cafe at St. John's College lectures, archives, unsubscribe, maps at http://www.friam.org |
Re: [FRIAM] Meaningful announcement, or
|
|
Owen,
HPC system requirements are pretty much 100% application-dependent. Large scale ABM simulations, like EpiSims require clusters with fast interconnects, as Marcus indicated. Other HPC apps have different requirements. E-Commerce needs large memory + lots of CPU horsepower for database transaction processing, but not necessarily low-latency MPP interconnect.
Other apps might have low-bandwith, high-latency behaviors that cloud-style HPC could satisfy. BTW, I got a case of the Willies looking at the NMCAC page. The have a pretty capable cluster, but the organization looks like it was patterned after the worst of the bureaucracies that LANL and the State of New Mexico have demonstrated. Look at
for an example of what I mean. --Doug -- Doug Roberts [hidden email] [hidden email] 505-455-7333 - Office 505-670-8195 - Cell
On Mon, Jan 25, 2010 at 9:17 AM, Owen Densmore <[hidden email]> wrote:
============================================================ FRIAM Applied Complexity Group listserv Meets Fridays 9a-11:30 at cafe at St. John's College lectures, archives, unsubscribe, maps at http://www.friam.org |
Re: [FRIAM] Meaningful announcement, or
|
|
In reply to this post by Marcus G. Daniels
LAVA started to bid on developing/providing these "gateways" over a year
ago and dropped out after discovering how dysfunctional the whole deal was (State procurement, NMCAC, etc.). The spec was well motivated technically but had become a bit of a nightmare patchwork of requirements that did not seem to acknowledge some real-world constraints. Essentially they wanted two types of "visualization" gateways: One type would be suitable for large installations and provide high quality 3D visualizaition and teleconferencing, the other would be suitable for smaller installations ( like a small classroom or meeting room). There were a number of requirements that made a "reasonable" installation totally unreasonable (like 3D teleconferencing at 1080p resolution). The low-end 3D systems they were clearly looking for were 3D Ready DLP TVs which we already had lots of experience with. There were two implicit requirements that these devices could not meet and we could not get clarification on whether their shortomings were acceptable. The shortcomings were: 1) while nominally 1920x1080, as TVs using rear-projection DLP technology and fixed optics, there is an "overscan" such that when used as a computer monitor (2D or 3D), a few dozen pixels splash outside the viewable area. Particularly aggravating when working with WinDoze as the little funky task bar gets clipped, no matter where you put it 2) the smaller systems specced a *pair* of these devices and it was strongly implied that they run in 3D together (as a single system). This is a nice thought and in principle not that hard, excepting that the rear-projected DLP is a single-chip system with a color wheel and no facility for syncing any of that from the outside. We had already run a pair of these side-by-side and the result was never going to achieve what they implied they wanted. Not being veteran bidders on gov't contracts, we finally threw our hands up and went home. The contract was finally let to some group from out of state called IOSYS and from what they are delivering, the state/NMCAC clearly dropped many of the requirements (explicit or implicit) that were confounding us (maybe even based on our various unacknowledged responses outlining the problems with the spec we saw?). The results are probably reasonable but far from what they were asking for and the price seems really high (now that the bid is awarded) but nobody in their right mind could have jumped through all the hoops they had to with a very limited promise. Winning the contract did not guarantee you would sell one much less the implied 22 systems. I understand the final system costs $38K/installation. It consists of 2 Mitsubishi 3D DLP TVs, a Windoze Box, an Access Grid compatible Teleconferencing system and ParaView (open source) software for scientific visualization. No 3D teleconferencing. They will be demo'ing it across 8 of the gateways this afternoon. The only installations I know of for sure are SFCC and UNM... the others surely include NMSU and NM Tech... Eastern and Western are a couple of other likely suspects. Maybe UNMLA and Highlands? Decent performance of the system would require good bandwidth to the sites... I assume these first eight sites were chosen because they already had decent connectivity. - Steve > Douglas Roberts wrote: >> political bullshit? >> >> http://www.kob.com/article/stories/S1381807.shtml?cat=500 >> > Curious to hear technical details... UNM and Encanto are apparently > already connected by high speed fiber. > > http://newmexicosupercomputer.com/encanto4.html > > ..extending that to other colleges would be a big investment. > > > ============================================================ > FRIAM Applied Complexity Group listserv > Meets Fridays 9a-11:30 at cafe at St. John's College > lectures, archives, unsubscribe, maps at http://www.friam.org ============================================================ FRIAM Applied Complexity Group listserv Meets Fridays 9a-11:30 at cafe at St. John's College lectures, archives, unsubscribe, maps at http://www.friam.org |
Re: [FRIAM] Meaningful announcement, or
|
|
In reply to this post by Douglas Roberts-2
Douglas Roberts wrote:
As long as we're dreaming, let's bring back Time Warp. One that works, this time.I used to think I needed to spend all of my time developing time-travel so I could go back in time to do all the things I didn't get done while I was wasting my time on time-travel. But then I realized it was going to take more than one lifetime for me to build a time machine so I thought maybe putting my time into developing cloning technology so I could work on the problem in parallel would be the answer. Then I realized that I wouldn't be able to decide how to distribute the work among myselves so I decided to work on the problem of parallel universes so that rather than breaking the problem up into smaller pieces and then putting the pieces back together, I could instead pursue all of approaches to the problem including the extremely low-probability but high-payoff strategies. Then I realized that in the multiverse, this is already happening and one (many) of my alternate selves has already cloned himself so he could do the bruteoforce parallel solution to developing a time machine so he (they) could go back in time to solve whatever the original problem they didn't have time to solve in real time. And then I realized that there had to be many more versions of me in the multiverse that had gone through this gedankenexperiment and realized the futility and silliness of it all. Somewhere, some version of me is actually doing productive work. - Steve
============================================================ FRIAM Applied Complexity Group listserv Meets Fridays 9a-11:30 at cafe at St. John's College lectures, archives, unsubscribe, maps at http://www.friam.org |
Re: [FRIAM] Meaningful announcement, or
|
|
There was a nice SF story based on this concept. The little boy who buried his prize possessions was terminally ill with no funding for the cure. Wish I could remember title and author: anyone? Thanks Robert PS SF in this case stands for Science Fiction and not any of the notable US cities. On 1/25/10 10:35 AM, Steve Smith wrote: Douglas Roberts wrote:As long as we're dreaming, let's bring back Time Warp. One that works, this time.I used to think I needed to spend all of my time developing time-travel so I could go back in time to do all the things I didn't get done while I was wasting my time on time-travel. But then I realized it was going to take more than one lifetime for me to build a time machine so I thought maybe putting my time into developing cloning technology so I could work on the problem in parallel would be the answer. ============================================================ FRIAM Applied Complexity Group listserv Meets Fridays 9a-11:30 at cafe at St. John's College lectures, archives, unsubscribe, maps at http://www.friam.org |
«
Return to Friam
|
1 view|%1 views
| Free forum by Nabble | Edit this page |