MD AUS.ASP Telecon minutes
From TeraGrid Wiki
Meeting minutes of 11/6/08 AUS - MD project telecon
Here is quick recap of what we talked about:
We talked about two scenarios for this project (and we will only pursure the first scenario):
1. User knows which MD code to use for his/her MD science simulations i.e. user has experience with MD codes on HPC machines to some extent. User comes to us for guidance on which machine to use for MD simulations to get the best performance.
2. User knows the MD problem he/she wants to simulate, but do not know which code to use. This becomes an ASTA type work to help user decide which MD code is best suitable for his/her science simulations.
We will look into multiple MD codes (NAMD, AMBER, GROMACS, CHARMM, LAMMPS etc.) eventually, but to start with we will look at AMBER (Ross) and NAMD (Phil), since we already have extensive experience with these codes as well as some priliminary performance/benchmark results.
Eventual plan is to prepare a technical doc that will have benchmark results of these codes on various TG machines (we will stay away from comparing one code to the other from science simulation perspective). This doc is expected to help TRAC reviwers also. This doc will also provide information about compiler used, optimization flags used, any other env variables set, mvapich lib used etc. on various TG machines. This will be presented such that users can install their own version if they want and compare performance results. We will only use released version of codes for our project work.
We will provide IPM results with these codes such that users have idea about performance scaling and impact of communication etc. IPM provides quick and easily understandable info without much effort. We will also pursue profiling with TAU but that will be for more involved understanding of performance and will be used for performance tuning and more directed towards developers than users (as is the case for IPM results). These materials can also be used for MD training classes to inform users about HPC aspect of MD codes.
We (Amit) will contact some well known MD users to get their input on this project idea.
Initially we may put results in TG Wiki and later figure out what is the best place for the technical doc.
Next telecon will be either first or second week of December (same day and time of the week) and will look at some results of AMBER and NAMD and discuss what is the best way to present these results eventually.
Meeting minutes of 12/5/08 AUS - MD project telecon
Here is a quick recap of the meeting minutes. (The whole telecon and the presentation done by Phil were recorded and available at: https://cc.readytalk.com/play?id=gcsa2e31)
1. We discussed what are the right MD benchmarks to run on various TG machines. The thinking was that input cases representing realistic simulations that users make are the appropriate ones as opposed to the benchmarks available on MD codes' webpages. Also we don't need to run same input cases for each of the MD codes we target, since our focus is not to compare codes, but to provide guidance to users regarding how representative input cases perform on differnt machines. Focus is on helping/guiding users (and possibly TRAC reviewers etc.) and not on comparison of codes or rigorously benchmarking one machine against another.
2. We will make all the optimization (both compiler and other environment options etc.) information available to the users via the documentation.
3. We are targetting AMBER and NAMD first, and will follow up with other MD codes (CHARMM, GROMACS, Desmond, LAMMPS) afterwards.
4. IPM and possibly TAU profiled results will be made available as a part of this MD project to MD users.
5. Amit will work with doc group to identify a doc person who will help with the webpage. Wayne and Amit will work with MD experts to figure out how to best present the text, results etc.
6. Phil gave a presentation of NAMD performance on Ranger; if you want to see the slides let Phil or me know. Phil pointed to the following webpage which has description of parameters used for tuning on Ranger: http://mvapich.cse.ohio-state.edu/support/mvapich_user_guide.html
7. Next tele/webcon in late January.
Meeting minutes of 3/12/09 AUS - MD project telecon
AUS-MD project discussion was held in conjunction with AUS technical tele/webpresentation on this day. Phil Blood, Ross Walker, and Wayne Pfeiffer participated on the discussion. The following ABBER webapage was shown : http://ambermd.org/amber10.bench1.html#fix and discussion was how to present data/results/plots for users such that it is useful for them yet do not overwhelm them with too much data
Meeting minutes of 4/9/09 AUS - MD project telecon
This was held as a part of the AUS technical tele/web presentation. Plots of AMBER and NAMD on Ranger and Kraken was presented. Ross and Phil generated the results and Wayne plotted them in various ways. Following is a recap of the discussion: Provide SUs/ns (Shawn Brown). Someone suggested to put conversion factor between multiple macines on the plot. It was suggested that run time options used should be cross-checked with RP site experts (TACC: Hang Liu, Kent) (NICS: Lonnie, Mark). There was discussion about the timestep size being used and to state that on the plots. There was discussion about listing key parameters on the sidebar. There was suggestion about putting clickable links on the plot such that they show what MPI option was used, what config options were used, input parameters for codes etc.