ARGOMENTO:

Ho due cose da segnalare:

1) Nella pagina del progetto delle 'informazioni tecniche' su QMC consiglio di aggiungere tra le applicazioni disponibili anche QMC@HOME-orca-alpha v5.02 dato che ogni tanto arriva qualche sua WU ...

2) Vorrei anche segnalare un piccolo bug di questa applicazione (orca) che appena scaricata, almeno a me, segnava un tempo di elaborazione di 64 ore, mentre ha impiegato meno di mezz'ora...

Problema che ho riscontrato ultimamente, arrivano WU con deadline decrescente (anche di soli pochi minuti) ma tanto da far andare in "waiting to run" un sacco di WU ... da task manager avevo in RAM 12 client, che quasi mi inchiodavano il PC.

Vedete lo stesso problema ?

Questa cosa della deadline decrescente non dovrebbe essere permessa lato server o dovrebbe essere permessa solo qualora gli sviluppatori di BOINC mettessero il client in grado di gestire questa situazione senza far scoppiare la RAM ....

Non ditemi di togliere il flag "leave in memory" perche' l'avevo impostato per non mandare in errore le WU.

EDIT

Controllando meglio, il problema lo sto vedendo sull'unico PC con upgrade alla 6.10.36, negli altri (6.10.18) tutto OK. Che sia un altro regalino della nuova release ?

astroale ha scritto:

Vedete lo stesso problema ?

sul forum del progetto non ho trovato niente

L'ultima news in home è di marzo, volevo solo segnalarlo...

akd ha scritto:

L'ultima news in home è di marzo, volevo solo segnalarlo...

L'importante è che funziona senza pasticci, non come certi progetti pieni di news e che poi ci fanno perdere tempo dietro ai loro casini

Sì certo, però un cenno di vita, anche solo per dire "siamo vivi, sappiamo cosa state macinando" sarebbe gradito...

Per quanto ne so io potrebbero anche essere andati via tutti e noi stiamo elaborando in loop sempre le stesse wu...

akd ha scritto:

Sì certo, però un cenno di vita, anche solo per dire "siamo vivi, sappiamo cosa state macinando" sarebbe gradito...

Per quanto ne so io potrebbero anche essere andati via tutti e noi stiamo elaborando in loop sempre le stesse wu...

Beh, per quello c'e' il forum, l'ultimo intervento di un admin e' abbastanza recente:

qah.uni-muenster.de/forum_thread.php?id=1312&nowrap=true#33283

Ottimo, grazie!

Grazie a simone per le news

Dear all,

and apologies, apologies, apologies - but you probably had enough of those already ... so: a number of quick answers ahead, then a more detailed report in a second posting and feel free to ask more questions! I have reserved some time for Q/A this week, so take your chance =)

adrianwx:
QFA will definitely come back, the problem is that it is at the moment not clear where to put the QFA server up again. The project will continue to be in cooperation with Mike Towler from Cambridge (and in the future probably also with other people from the TCM group of the theoretical physics department there), but I am currently waiting for a new grant and for applications as junior group leader / junior professor, which means I am right now missing the resources to get things running again properly. I could set up something preliminary over here at the Max-Planck-Institute, but this would very likely mean to change a lot of things again in just a few months.

tullio:
The last year of QMC was not the chattiest one, but the scientifically most successfull one (one publication submitted, one publication in preparation, data for two more publications are processed now). With the help of the Cambridge QMC experts during my stay in their workgroup I was able to solve the problems that we run into a lot of times before in the last years (resulting in a lot of new insight, but not really a lot of publications). On the other hand: Not sure if 'too much scientific work' is a good excuse for missing interactions between developers and users when it comes to volunteer computing, so still: I'm sorry! If anyone has the feeling that I should definitely pay attention to his postings in the forum, please feel free to email me: dgd -at - uni-muenster -dot- de

More information in my next posting ...

Best wishes & thanks for supporting QMC@HOME,
Martin

PROJECT OVERVIEW I/2011

Dear qmc-at-home supporters,

first of all: thank you very much for your continued support of our research. Our project has again seen some major changes in the last year. This little news posting should give you a short overview of what was done.

* WORKUNITS *
Because of several problems with memory and disk requirements, ORCA calculations are suspended at least for now - though they might come back in the next year if we find a way to make them more user-friendly. So we concentrated on QMC again, where we changed our main application from Amolqc to CASINO (or how our version is called; QASINO), the QMC code of the Cambridge QMC group. This change (together with a lot of stuff we found out about QMC for molecules over the last year) solved almost all our problems ... juhu! (German sound of great happiness.) The bad point: Our screensaver is not working with CASINO, so that over the last year we essentially had no screensaver graphics. I had a preliminary version with the screensaver running, but rather often the graphics got stuck at some point and crashed the workunit. This is very likely because we have not fully ported our screensaver to the updated BOINC graphics system but used a workaround. So full porting is necessary, but work time is scarce ... We are also still missing a GPU version of CASINO (preliminary tests with Amolqc went quite well, but now things have to be done again for CASINO), which should be a great boost in performance (MC problems are rather well suited for GPUs).

* QUANTUM CHEMISTRY *
a) With QASINO we were able to crunch our way through a huge number of calculations on several benchmark sets which now nicely illustrate the potential of QMC for Quantum Chemistry. Most of the data has ended up in one manuscript, that will be the core outcome of my research in Cambridge. This is the abstract:

Accurate Thermochemistry with Fixed Node Diffusion Monte Carlo
(Martin Korth and Mike D. Towler)

"A systematic study of the influence of trial wave function parameters on absolute and relative fixed node diffusion Monte Carlo (FNDMC) energies for molecular systems is presented. Effects resulting from technical parameters up to aspects in the generation of trial wave functions are examined, including the application of pseudo potentials, DFT wave functions, basis sets of increasing size, backflow terms, different Jastrow factor terms and VMC optimization methods. It is found that pseudo potential FNDMC calculations with DFT trial wave functions and mid-size basis sets reach the accuracy of the 'gold standard' CCSD(T)/cc-pVTZ method for the calculation of reaction energies (more than halving the average errors of previously published all-electron FNDMC data with Hartree-Fock trial wave functions)."

Okay, that's a lot of sciencie mojo, it basically says: We are ready to move on to real-life systems!

b) And we paid special attention to respond to a recent publication observing big differences between QMC and 'standard' Quantum Chemistry methods (as this of course would have questioned the reliability of the data presented in the manuscript above):

The lithium-thiophene riddle revisited
(Martin Korth, Stefan Grimme and M.D. Towler)

"A recent study of the interaction of a lithium atom with the thiophene molecule found a large disagreement between high-level coupled cluster (CCSD(T)/AVTZ) and quatum Monte Carlo (fixed-node diffusion Monte Carlo, or FNDMC) calculations. We address this `lithium-thiophene riddle' by analyzing the influence of crucial FNDMC simulation parameters, namely the one-electron models, basis sets, and pseudopotentials used for the generation of the trial wave function. These are shown to have a significant impact on the calculated FNDMC interaction energies and good agreement between CCSD(T) and FNDMC is found when nodal hypersurfaces of sufficient quality are used. Based on our proposed consensus reference value, we go on to benchmark the standard toolbox of lower-level quantum chemistry methods for this model interaction. Newly- eveloped dispersion-corrected DFT methods perform reasonably well despite the partial charge transfer character of the interaction and might well be worthy of further study in larger lithium-thiophene systems."

Again some sci-speak, basically saying: We know how to do it right now! (And sorry to Caffarel et al. for this very, very simplistic description of the underlying scientific problem to which both research groups have contributed.)

b)
So what will be next? The above mentioned benchmarks actually delivered a lot more results, some of which are definitely worth publishing, so that two more manuscripts will likely follow. And then? Somewhere above I said; we can now move on to 'real-life' systems - how does that work? I do not want to reveal too much, but will roughly cover one point:

Real-life systems (a protein-ligand complex, an enzyme within a catalytic cycle) are far to large to be treated with QMC (in fact every 'quantum computational' method using the computational resources available today. But there is one way around this: You can use QMC for the inner, important part of system and a lower-level (faster, less accurate) method for the outer (less important) part. Techniques like this are called QM/MM methods: QM for a (inner) quantum mechanical method and MM for an (outer) molecular modeling methods - the later one being based on a 'classical view' of what happens in a molecule, commonly used for instance for protein folding, where things have to be faaast). Our QM method will be QMC, combined with MM methods to form the new QMC/MM method, that will allow us to push QM/MM techniques to higher accuracy and (probably even more important) open them to massively parallel computations. For this I joined the work group of Walter Thiel at the MPI 'für Kohlenforschung' (Carbon or coal research - a name related to the history of this place in the middle of the Ruhrgebiet, the industrial powerhouse of Germany from before the 'Kaiserzeit' to the 70s energy crisis). Professor Thiel is one of the leading scientists in the field of QM/MM development and application.

* PROJECT STAFF *
Uh, just me again; Robert is snowed under with money in a private company and Waldemar has turned his attention to another project at least for a while. Why is noone willing to invest money into a 10-20 TFlops volunteer supercomputer? See how little money is around in science ... I should add: if you are doing Volunteer Computing, which is still not really accepted in a lot of scientific communities. But I guess you simple have other problems if you have convinced your government to build you a new supercollider ... probably even more severe problems ... =)

* PROJECT STATE *
Yeah, I guess we will never leave beta. Simply missed it at some point. are we possible the most stable beta project ever? Maybe having the screensaver available for CASINO would be a good point?

* PROJECT FUTURE *
Things I will work on in the next year are:
- Re-open QuantumFIRE alpha for QMC in physics as soon as (financially) possible
- CASINO screensaver & GPU app ...
- QMC/MM development and application to enzymatic reactions!!!

* LAST BUT NOT LEAST *
If you have any questions - please post them!

Thanks again for supporting QMC-at-HOME!
Martin

E' normale che WU da 70 ore siano a 0% dopo un ora di elaborazione?

E' possibile, se non addirittura molto probabile :P

Aspetta che le ore comincino a salire (che arrivino a 10 tipo..) e vediamo se la percentuale sale.

Nosf3r ha scritto:

E' normale che WU da 70 ore siano a 0% dopo un ora di elaborazione?

Se l'incremento percentuale non è lineare è molto probabile una cosa del genere

Infatti dopo un paio d'ore ora è al 6% grazie per la delucidazione!

Ottimo

ciao a tutti
ho un problema abbastanza banale ma che su questo progetto è insormontabile...
In questo progetto ho il nick "gualo" e non trovo da nessuna parte dove modificarlo in "gualo [Lombardia]".
La home page non esiste e le uniche info che riesco a visualizzare ma non a modificare son le WU e i pc che elaborano questo progetto...
avete un qualche consiglio da darmi?

ciao e grazie

Penso che devi cambiare il browser, per questo progetto io uso google chrome.

gualo ha scritto:

ciao a tutti
ho un problema abbastanza banale ma che su questo progetto è insormontabile...
In questo progetto ho il nick "gualo" e non trovo da nessuna parte dove modificarlo in "gualo [Lombardia]".
La home page non esiste e le uniche info che riesco a visualizzare ma non a modificare son le WU e i pc che elaborano questo progetto...
avete un qualche consiglio da darmi?

ciao e grazie

Io non ho problemi a visualizzare questo progetto.


Comunque se ti può servire questo è il link diretto al tuo account: qah.uni-muenster.de/home.php

Inserisci mail e pass ed entri nel tuo account

a me in IE mi compare questa pagina e basta



o quando clicco su Return to QMC@HOME main page

Mai avuto problemi, uso firefox.

Daniele69 ha scritto:

a me in IE ...cut...

"IE"? Che sarebbe?