From: Aron Broom (broomsday_at_gmail.com)
Date: Fri May 04 2012 - 16:33:39 CDT
Ok, Thanks a lot for that clarification. I had wondered why at some
timesteps they were not equal, but by the time the simulation finished they
were. So I've been doing things in a reasonable manner it seems, and more
importantly the multiple walker method is quite resilient to crashing and
asynchrony, splendid.
Thanks,
~Aron
On Fri, May 4, 2012 at 5:30 PM, Giacomo Fiorin <giacomo.fiorin_at_gmail.com>wrote:
> Hi Aron, if you're in doubt, have a look at the .txt files that are
> generated, it should clarify which state files are the most recent.
>
> Each replica/walker should be restarted *from its own most recent state
> file*, with no exceptions. Just keep running each one as a regular NAMD
> simulation, they will communicate with one other on their own.
>
> In any case, if you have e.g. replicaID = "R001" in the configuration file
> of one replica/walker, and you feed to it the state file from the walker
> "R002" it should give you an error. So you should see for yourself which
> one is the right way :-)
>
> If one walker has been offline for a long time (e.g. because it crashed
> and you're now restarting it), there will be of course a sudden jump in its
> PMF, because at that moment it reads all the hills that the other walkers
> have been accumulating in the meantime. So it makes no sense to compare
> the PMFs before and after a restart, if one replica/walker has been
> off-sync for a while.
>
> What makes sense instead is to compare the PMFs written by different
> replicas/walkers within minutes from each other. Those should be
> essentially the same: if they turned out wildly different, please send me
> the details so that I can try to figure it out.
>
> Bests
> Giacomo
>
>
> On Fri, May 4, 2012 at 4:54 PM, Aron Broom <broomsday_at_gmail.com> wrote:
>
>> sorry, just as a second follow-up question. If what I've said in the
>> previous message is true, is it also possible to restart a walker that has
>> crashed while the other walkers are still running, and they will again
>> start picking up the new hills?
>>
>> Thanks,
>>
>> ~Aron
>>
>> On Fri, May 4, 2012 at 4:52 PM, Aron Broom <broomsday_at_gmail.com> wrote:
>>
>>> I've been running a few preliminary experiments with the multiple
>>> walkers MetaDynamics, and I'm wondering about restarts. I've had some
>>> issues with a few nodes crashing, and have had to restart some of the
>>> walkers. What I've noticed is that the pmfs for different walkers at the
>>> same timestep don't match once there has been some kind of crash, which I
>>> guess makes sense, as there is no inefficiency of synchronizing them, so
>>> they are all semi-independent (or maybe I'm confused on that).
>>>
>>> Anyway, my question then is, if all my walkers have stopped (having
>>> finished whatever short chunk of time they get run for, or all having
>>> crashed, or some combination of the two), and I want to continue with the
>>> next chunk of simulation, should I take the *.colvars.state file from the
>>> most progressed walker (the one with the most hills) and replace all the
>>> other walkers with that, so that they now start at the most advanced stage?
>>>
>>> Thanks,
>>>
>>> ~Aron
>>>
>>> --
>>> Aron Broom M.Sc
>>> PhD Student
>>> Department of Chemistry
>>> University of Waterloo
>>>
>>>
>>
>>
>> --
>> Aron Broom M.Sc
>> PhD Student
>> Department of Chemistry
>> University of Waterloo
>>
>>
>
-- Aron Broom M.Sc PhD Student Department of Chemistry University of Waterloo
This archive was generated by hypermail 2.1.6 : Mon Dec 31 2012 - 23:21:31 CST