Carolyn, How much time do you need between now and 3/31/16? The Allocations Committee meets in March to make decisions about Q3 and Q4. So for big requests we provide part now and part through the Committee decision in March. Regards, Ray ----------------------------------------- Ray Bair Argonne National Laboratory and the University of Chicago On 1/11/16, 8:37 AM, "[email protected] on behalf of [email protected]" <[email protected] on behalf of [email protected]> wrote:
Hello,
A change in allocation has been requested:
Requester: cphillips (Carolyn Phillips) Project: Metadynamics_LC Title: Using Parameter Free Metadynamics to calculate LC elastic constants Description: We will be using a parameter-free metadynamics method for an atomistic model of a liquid crystal 4-n-pentyl-4’-cyanobiphenyl (5CB), as well as the homologous series (6CB,7CB,8CB).
Each nCB molecule will be simulated via a coarse-grained united-atom approximation in a classical N-body Molecular Dynamics code. We will be using NAMD + PLUMED2. Simulations will be used to measure the elastic constants of each liquid crystal system using metadynamics.
NAMD is a well-parallelized stable MD code with a history of use by other users on Blues. Equilibrium simulations of the correct scale of the model described above parallelized across on 4x16-core nodes required ~300 MB of memory and ~36 hours to implement 10 million time steps. However most simulations will also be paired with PLUMED2, a parallelized version of PLUMED1.3 that can be used to implement metadynamics. Metadynamics is not known to perform well in parallel, due to a relatively high frequency of communication across all nodes, so I do not expect to use more than 1-4 nodes for individual "walkers" (simulation implementations). Instead production run simulations will likely use multiple (~20) "walkers" each deployed on 1-4 nodes, thus requiring ~80 nodes at a time. We will be evaluating trade-offs in length of simulation, number of nodes per implementation, and number of walkers to achieve convergence. Thus a better approximation of the allocation required to complete this project will result from the initial scoping stage of this project.
Previous MD + metadynamics simulations have been performed on a different model of the system using a different MD code on Midway at UChicago. Small runs using NAMD on Blues and Fusion have already been issued to gain familiarity.
Expected number of Project Members: 2 Current: undetermined amount Justification: A mix of simulation types are currently being run.
- Large simulations that are scaled over four nodes at a time using NAMD - Smalls simulations being run over a single node but also using multiple walkers to empirically determine what walker scheme generates the best convergence.
Using these two schemes I have used 100K this year.
Soon I will need to run Large Simulations over 4 nodes with multiple walkers.
Requested: 500000
A specific reason has been given: Testing a scheme for achieving convergence using many walkers and a small system plus equilibrating a large system that will be used to extract data has used 100K already this year.
This needs to be approved and the final allocation amount decided upon.
Thank You, The LCRC Accounts System