I'm assuming some things here, but I'll give it a shot anyhow

When you say you want to let the explorers talk to each other, you mean that "if the ab initio structure
is too close to another" try again.

pseudo code

1. (Local machine) does an ab initio and comes up with a structure
2. (Local machine) ask the server "is this structure/or is this structure too close to" (though the later may be a sticking point) a structure already doing a "full atom relax".
3 (BOINC) network is unavailable go to 6. (for dial up users, net and server outages)
4. (Server) YES go to step 1
5. NO (server) updates database of structures to include new structure
6. (Local machine) do a "full atom relax"

/pseudo code

Problems:
1. size of DB on server
2. net traffic to server
3. the amount of the same results may not warrant it

If my assumptions are correct it doesn't seem overly hard to me.

Though of course I'm not sure if you are talking about the CPU's talking during the full atom relax stage, or even talking during the ab initio stage?

If this is all wrong, could you give us some information about what you would really like to do with the interactions between CPUs?

edit: formatting

---

I'm assuming some things here, but I'll give it a shot anyhow

When you say you want to let the explorers talk to each other, you mean that "if the ab initio structure
is too close to another" try again.

pseudo code

1. (Local machine) does an ab initio and comes up with a structure
2. (Local machine) ask the server "is this structure/or is this structure too close to" (though the later may be a sticking point) a structure already doing a "full atom relax".
3 (BOINC) network is unavailable go to 6. (for dial up users, net and server outages)
4. (Server) YES go to step 1
5. NO (server) updates database of structures to include new structure
6. (Local machine) do a "full atom relax"

/pseudo code

Problems:
1. size of DB on server
2. net traffic to server
3. the amount of the same results may not warrant it

If my assumptions are correct it doesn't seem overly hard to me.

Though of course I'm not sure if you are talking about the CPU's talking during the full atom relax stage, or even talking during the ab initio stage?

If this is all wrong, could you give us some information about what you would really like to do with the interactions between CPUs?

edit: formatting

That is the basic idea--there are many possible variations; for example, you could consider the energy distribution and density of similar structures in addition to the distance to the closest one. In addition to your Problems 1 and 2, there is the time required for the very large number of similarity comparisons (your machine would be idle during this time).

Another possibility we have been considering is to generate a uniformly spread out set of ab intio structures using our in house machines, and then send them out to BOINC for relaxing. The advantage is that communication is easy on our local clusters. The disadvantage is that it will take a large amount of time to do the ab initio runs locally, even though they are individually very quick, and that the movies would not be as much fun to watch since most things happen in the ab initio part.

A compromise would be to carry out a first set of runs using BOINC as you are all doing, and then develop a shorthand description of the oversampled and undersampled regions of the landscape based on these results. Then, in a second set of runs, this information would be sent out to all processors and used in your steps 4-5. We already tried an early version of this approach in the 1n0u bab runs, and it looked pretty promising.

---

Let me first say thank you for your speedy and in-depth reply.

I've been digesting this information for quite some time, and truth be known, I'm probably saying things you've already thought of :)

Another possibility we have been considering is to generate a uniformly spread out set of ab intio structures using our in house machines, and then send them out to BOINC for relaxing. The advantage is that communication is easy on our local clusters. The disadvantage is that it will take a large amount of time to do the ab initio runs locally, even though they are individually very quick, and that the movies would not be as much fun to watch since most things happen in the ab initio part.

My gut feeling is that you will probably have to go this way to get an adequate spread of the searchers, and a run of these might be beneficial, it has in it's favour that it may show you where (and why) your code is biased toward particular areas. I don't discount the possibility ( probability ) that you are already aware of the code's limitations and just have to power though them, I can imagine many reasons why this might be the case.

A compromise would be to carry out a first set of runs using BOINC as you are all doing, and then develop a shorthand description of the oversampled and undersampled regions of the landscape based on these results. Then, in a second set of runs, this information would be sent out to all processors and used in your steps 4-5. We already tried an early version of this approach in the 1n0u bab runs, and it looked pretty promising.

That does sound good, but if you have a really large problem with oversampling, in nth set of runs, assuming that you would continue them, the speed of the results and size of the "shorthand description" may put a lot of strain on your server, though only time will tell.


edit: formatting. it doesn't like

---

I've been thinking and doing a little research on the subject here's what I've come up with so far:


Have you had a chat to these people The
Computer Science and Mathematics Division (CSM) at Oak Ridge National
Laboratory's

Though they may say something along the lines of:
Ditch BOINC, use a mixed peer-to-peer network, use the newer processors like a node, type things, but I'm sure they could restrain themselves if you told them that, this couldn't happen whilst keeping a dedicated and growing BOINC participant base.


They should come up with many ideas that would help speed things up, and it doesn't seem like it will cost you any $.

The Computer Science and Mathematics Division (CSM) is ORNL's premier source of basic and applied research in high-performance computing, applied mathematics, and intelligent systems. Basic and applied research programs are focused on computational sciences, intelligent systems, and information technologies.

Our mission includes working on important national priorities with advanced computing systems, working cooperatively with U.S. Industry to enable efficient, cost-competitive design, and working with universities to enhance science education and scientific awareness. Our researchers are finding new ways to solve problems beyond the reach of most computers and are putting powerful software tools into the hands of students, teachers, government researchers, and industrial scientists.

Edit: What is wrong with the url BBCode?

---

Have you had a chat to these people The
Computer Science and Mathematics Division (CSM) at Oak Ridge National
Laboratory's

Edit: What is wrong with the url BBCode?

They probably don't like the included line breaks.

Have you had a chat to these people The Computer Science and Mathematics Division (CSM) at Oak Ridge National Laboratory's

---

They probably don't like the included line breaks.

You are indeed right, thanks.

---