From: Tristan Croll (tristan.croll_at_qut.edu.au)
Date: Fri Feb 21 2014 - 01:44:39 CST
I first noticed the problem on an SGI Altix XE cluster, which does have ECC RAM. I honestly can't remember if my local machine is has ECC as well, but I don't think that's the problem. It also arises with and without GPU (mine is a Tesla C2050) - if anything.
Cheers,
Tristan
Tristan Croll
Lecturer
Faculty of Science and Technology
Institute of Health and Biomedical Engineering
Queensland University of Technology
60 Musk Ave
Kelvin Grove QLD 4059 Australia
+61 7 3138 6443
This email and its attachments (if any) contain confidential information intended for use by the addressee and may be privileged. We do not waive any confidentiality, privilege or copyright associated with the email or the attachments. If you are not the intended addressee, you must not use, transmit, disclose or copy the email or any attachments. If you receive this email by mistake, please notify the sender immediately and delete the original email.
> On 21 Feb 2014, at 5:35 pm, "Norman Geist" <norman.geist_at_uni-greifswald.de> wrote:
>
> This is interesting. I've never observed such a runaway behavior while using
> Amber, maybe this is a hint. Additionally, are you guys, having this
> problem, using ECC enabled machines and GPUs.
>
>
> Norman Geist.
>
>> -----Ursprüngliche Nachricht-----
>> Von: owner-namd-l_at_ks.uiuc.edu [mailto:owner-namd-l_at_ks.uiuc.edu] Im
>> Auftrag von Tristan Croll
>> Gesendet: Freitag, 21. Februar 2014 00:52
>> An: namd-l_at_ks.uiuc.edu
>> Betreff: namd-l: RE: Runaway positive feedback when minimizing large
>> systems (was: strange glitches when performing energy minimization in
>> MDFF simulations)
>>
>> Minor error correction: sentence below should read "... had undergone
>> extensive *equilibration*..."
>>
>> -----Original Message-----
>> From: owner-namd-l_at_ks.uiuc.edu [mailto:owner-namd-l_at_ks.uiuc.edu] On
>> Behalf Of Tristan Croll
>> Sent: Thursday, 20 February 2014 9:54 AM
>> To: Bennion, Brian; Norman Geist; namd-l_at_ks.uiuc.edu
>> Subject: namd-l: RE: Runaway positive feedback when minimizing large
>> systems (was: strange glitches when performing energy minimization in
>> MDFF simulations)
>>
>> Hi all,
>>
>> Just wondering if anyone's had a chance to look at this?
>>
>> My own observations from watching a few different-sized simulations
>> minimize in IMD Connect after doing various perturbations to different
>> residues:
>>
>> - For some reason that's not entirely clear (perhaps floating-point
>> errors as the total system energy gets very large?), minimization in
>> general gets more "jumpy" (i.e. atoms move further with each
>> minimization step) as the system size increases;
>>
>> - In some (fairly rare) cases, this leads to a temporary runaway in the
>> algorithm as a small initial deviation is magnified into a very large
>> local disturbance, which can lead to cis/trans or chirality flips
>> before settling;
>>
>> - The "runaway" effect seems to only occur in a few different
>> situations - so far I've only been able to replicate it reliably in
>> solvent-exposed asparagine residues (by changing the length of the
>> side-chain C=O bond) and in histidines (by moving the side-chain
>> nitrogens slightly closer together). Similar scale changes in other
>> residues still lead to transient deviations larger than what one would
>> expect to see in a smaller-scale simulation, but these are usually
>> brought under control quite quickly (I've by no means done an
>> exhaustive check of different residues, though). In both the asparagine
>> and histidine cases, I've seen such runaways arise in re-minimization
>> of structures that had undergone extensive minimization, indicating
>> that structural deviations within (but at the outskirts of) normal
>> ranges can trigger the problem.
>>
>> - I've replicated the same behaviour in NAMD versions 2.8 and 2.9, on a
>> cluster (ibverbs) build and on my local workstation, in both CUDA and
>> non-CUDA builds, and in both CHARMM 27 and 36 forcefields.
>>
>> - Perhaps most strangely, I had more trouble replicating the runaway
>> behaviour in the STMV benchmark than in the ATPase benchmark.
>>
>> Anyway, a fairly straightforward workaround for anyone who's worried is
>> to always do a few thousand steps of backbone-restrained minimization
>> first, even if you're just re-minimizing a previously equilibrated
>> system.
>>
>> Cheers,
>>
>> Tristan
>>
>> -----Original Message-----
>> From: owner-namd-l_at_ks.uiuc.edu [mailto:owner-namd-l_at_ks.uiuc.edu] On
>> Behalf Of Tristan Croll
>> Sent: Thursday, 13 February 2014 9:51 AM
>> To: Bennion, Brian; Norman Geist; namd-l_at_ks.uiuc.edu
>> Subject: namd-l: RE: Runaway positive feedback when minimizing large
>> systems (was: strange glitches when performing energy minimization in
>> MDFF simulations)
>>
>> Hi Brian,
>>
>> I'll go one better. At the link below I've put together a
>> demonstration simulation. It's based on the ATPase benchmark, and the
>> only thing I've done is moved the OD1 atom on ASN33 of segname PSLN by
>> roughly 0.2A.
>>
>> - Untar into a directory of your choice
>> - load f1wations_xplor.psf / f1wations-pme-nve-
>> 200ps_PSLN_33_OD1_moved.pdb in VMD
>> - visualise "segname PSLN and resid 33" and "same residue as within 10
>> of (segname PSLN and resid 33)" in your choice of representation, and
>> center your view on the former
>> - run namd using the config file f1atpase.namd
>> - In VMD, go to Extensions/Simulation/IMD Connect
>> - enter "localhost" in the Hostname box and 2030 in the Port box
>> - after allowing some time for the simulation to set itself up, click
>> Connect
>> - enjoy (?) the show.
>>
>>
>>
>> Cheers,
>>
>> Tristan
>>
>> https://www.dropbox.com/s/zqa9qwy60v2bd4x/f1atpase_min_glitch.tar.gz
>>
>>
>>
>> -----Original Message-----
>> From: owner-namd-l_at_ks.uiuc.edu [mailto:owner-namd-l_at_ks.uiuc.edu] On
>> Behalf Of Bennion, Brian
>> Sent: Thursday, 13 February 2014 2:57 AM
>> To: namd-l_at_ks.uiuc.edu
>> Subject: namd-l: RE: Runaway positive feedback when minimizing large
>> systems (was: strange glitches when performing energy minimization in
>> MDFF simulations)
>>
>> Hello Tristan,
>>
>> In one of you posts you showed a histidine oscillating. can this be
>> replicated with any histidine in your autoimd tests?
>>
>> Does the asp need to be in the center of the box to be show the
>> behaviour?
>> Would you mind posting your autoimd config files and I will have a go
>> at it
>
>
>
> ---
> Diese E-Mail ist frei von Viren und Malware, denn der avast! Antivirus Schutz ist aktiv.
> http://www.avast.com
>
This archive was generated by hypermail 2.1.6 : Wed Dec 31 2014 - 23:22:10 CST