[gmx-users] Replica exchange simulations more than number of processor

Szilárd Páll pall.szilard at gmail.com
Tue Apr 18 23:00:55 CEST 2017 

BTW, forgot to mention that you may not loose much _aggregate_
performance by over-subscribing cores. I have not tried recently, but
last time I checked even with with multiple (single-threaded) ranks
per core you *should* get about the same aggregate performance. Just
turn off pinning (which will be done by default) and preferably run #
ranks multiple of the core count and it should run fine.

--
Szilárd


On Sat, Apr 15, 2017 at 3:03 PM, Мижээ Батсайхан <b.mijiddorj at gmail.com> wrote:
> Dear gmx users,
>
> Thank you very much for your useful discussions.
>
> Best regards,
> Mijee
>
> On Sat, Apr 15, 2017 at 7:00 PM, <
> gromacs.org_gmx-users-request at maillist.sys.kth.se> wrote:
>
>> Send gromacs.org_gmx-users mailing list submissions to
>>         gromacs.org_gmx-users at maillist.sys.kth.se
>>
>> To subscribe or unsubscribe via the World Wide Web, visit
>>         https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users
>> or, via email, send a message with subject or body 'help' to
>>         gromacs.org_gmx-users-request at maillist.sys.kth.se
>>
>> You can reach the person managing the list at
>>         gromacs.org_gmx-users-owner at maillist.sys.kth.se
>>
>> When replying, please edit your Subject line so it is more specific
>> than "Re: Contents of gromacs.org_gmx-users digest..."
>>
>>
>> Today's Topics:
>>
>>    1. Re: Replica exchange simulations more than number of
>>       processor (Thomas Piggot)
>>    2. Re: Replica exchange simulations more than number of
>>       processor (Christopher Neale)
>>
>>
>> ----------------------------------------------------------------------
>>
>> Message: 1
>> Date: Fri, 14 Apr 2017 20:06:29 +0100
>> From: Thomas Piggot <t.piggot at soton.ac.uk>
>> To: gmx-users at gromacs.org
>> Subject: Re: [gmx-users] Replica exchange simulations more than number
>>         of processor
>> Message-ID: <339c3321-fd12-7318-43b6-b41433f582fd at soton.ac.uk>
>> Content-Type: text/plain; charset=gbk; format=flowed
>>
>> Have you thought about a more selective scaling of the interactions
>> (i.e. HREX with only scalings applied for certain bits of the protein,
>> not the complete thing as in REST2) so as to reduce the protein
>> hydrophobicity at the lower lambda's? I've seen this selective scaling
>> work well in other types of system to counteract broadly similar types
>> of sampling issues at lower lambda's.
>>
>> Just a thought, it might not work well for what you are doing,
>>
>> Cheers
>>
>> Tom
>>
>> On 14/04/17 19:27, Christopher Neale wrote:
>> > In REST2, the protein becomes way too hydrophobic as lambda is
>> decreased. Just run a standard simulation at 300K, another at 600K and then
>> another one at 300K with REST2 lambda=0.5 and you'll see what I mean.
>> >
>> > REST has a hidden barrier corresponding to the point at which water
>> becomes a theta solvent and the exchange is very poor across that boundary,
>> with globular replicas staying at higher lambda and extended replicas
>> staying at lower lambda. However, with REST2, there is no such hidden
>> barrier because water never gets anywhere near being a theta solvent. So in
>> REST2 you get good exchange for all the wrong reasons.
>> >
>> > Work continues....
>> >
>> >
>> > ________________________________________
>> > From: gromacs.org_gmx-users-bounces at maillist.sys.kth.se <
>> gromacs.org_gmx-users-bounces at maillist.sys.kth.se> on behalf of Thomas
>> Piggot <t.piggot at soton.ac.uk>
>> > Sent: 14 April 2017 14:16:26
>> > To: gmx-users at gromacs.org
>> > Subject: Re: [gmx-users] Replica exchange simulations more than number
>> of processor
>> >
>> > Hi Chris,
>> >
>> > This is interesting. Do you have an idea why the sampling at low lambda
>> > values isn't very good?
>> >
>> > Cheers
>> >
>> > Tom
>> >
>> > On 14/04/17 17:40, Christopher Neale wrote:
>> >> What I have is only my own personal unpublished data. But add to that
>> the number of years since REST2 came out and the low number of papers in
>> which anybody did something useful with it and I'd say that suggests many
>> others have tried and run into the same issues that I did (or maybe I'm
>> just not reading the right papers, so please correct me if I am wrong).
>> When my desired use failed I fell back to trp-cage folding in water and
>> found the same things, so it's not like REST2 only fails for some exotic
>> systems. Running 3 entirely separate repeats of regular T-REMD and 3 of
>> REST2, I found that overall T-REMD is of similar or perhaps better
>> efficiency (even though it has more replicas) compared to REST2. It's an
>> on/off area of research for me, attempting to fix REST, but I still don't
>> have a good answer for it.
>> >>
>> >>
>> >> ________________________________________
>> >> From: gromacs.org_gmx-users-bounces at maillist.sys.kth.se <
>> gromacs.org_gmx-users-bounces at maillist.sys.kth.se> on behalf of Szil?rd
>> P?ll <pall.szilard at gmail.com>
>> >> Sent: 14 April 2017 12:26:41
>> >> To: Discussion list for GROMACS users
>> >> Subject: Re: [gmx-users] Replica exchange simulations more than number
>> of       processor
>> >>
>> >> On Fri, Apr 14, 2017 at 5:51 PM, Christopher Neale <
>> >> chris.neale at alum.utoronto.ca> wrote:
>> >>
>> >>> Just my 2 cents: REST2 doesn't work. The random walk is much better
>> than
>> >>> REST but the sampling efficiency at low lambda is now poor.
>> >> Interesting. Is there any (published) data you could point to that
>> supports
>> >> this?
>> >>
>> >>
>> >>> Your mileage may vary, but be aware that REST2 is not necessarily
>> going to
>> >>> be useful for all systems.
>> >>> ________________________________________
>> >>> From: gromacs.org_gmx-users-bounces at maillist.sys.kth.se <
>> >>> gromacs.org_gmx-users-bounces at maillist.sys.kth.se> on behalf of Thomas
>> >>> Piggot <t.piggot at soton.ac.uk>
>> >>> Sent: 14 April 2017 09:23:35
>> >>> To: gmx-users at gromacs.org
>> >>> Subject: Re: [gmx-users] Replica exchange simulations more than number
>> of
>> >>> processor
>> >>>
>> >>> In addition, using another technique like REST2 may substantially
>> reduce
>> >>> the numbers of replica's you need. Still, more compute is definitely
>> >>> recommended if you can get it.
>> >>>
>> >>> Cheers
>> >>>
>> >>> Tom
>> >>>
>> >>> On 14/04/17 12:48, Smith, Micholas D. wrote:
>> >>>> You can try over assignment of cores, but that will be horribly slow.
>> Or
>> >>> use more than 1 node (if you can get it).
>> >>>> ===================
>> >>>> Micholas Dean Smith, PhD.
>> >>>> Post-doctoral Research Associate
>> >>>> University of Tennessee/Oak Ridge National Laboratory
>> >>>> Center for Molecular Biophysics
>> >>>>
>> >>>> ________________________________________
>> >>>> From: gromacs.org_gmx-users-bounces at maillist.sys.kth.se <
>> >>> gromacs.org_gmx-users-bounces at maillist.sys.kth.se> on behalf of ?????
>> >>> ????????? <b.mijiddorj at gmail.com>
>> >>>> Sent: Friday, April 14, 2017 4:16 AM
>> >>>> To: gromacs.org_gmx-users at maillist.sys.kth.se
>> >>>> Subject: [gmx-users] Replica exchange simulations more than number of
>> >>>    processor
>> >>>> Dear gmx users,
>> >>>>
>> >>>> I would like to simulate folding of a peptides. I have only 12 core
>> >>>> processor, and I assumed the number of replica using temperature
>> >>> generator
>> >>>> as following link
>> >>>> http://folding.bmc.uu.se/remd/.
>> >>>>
>> >>>> The number of replica is about 60. How can I solve this problem? Can
>> you
>> >>>> advice me, please?
>> >>>>
>> >>>> Best regards,
>> >>>> Mijee
>> >>>> --
>> >>>> Gromacs Users mailing list
>> >>>>
>> >>>> * Please search the archive at http://www.gromacs.org/Support
>> >>> /Mailing_Lists/GMX-Users_List before posting!
>> >>>> * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists
>> >>>>
>> >>>> * For (un)subscribe requests visit
>> >>>> https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or
>> >>> send a mail to gmx-users-request at gromacs.org.
>> >>> --
>> >>> Dr Thomas Piggot
>> >>> Visiting Fellow
>> >>> University of Southampton, UK.
>> >>>
>> >>> --
>> >>> Gromacs Users mailing list
>> >>>
>> >>> * Please search the archive at http://www.gromacs.org/Support
>> >>> /Mailing_Lists/GMX-Users_List before posting!
>> >>>
>> >>> * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists
>> >>>
>> >>> * For (un)subscribe requests visit
>> >>> https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or
>> >>> send a mail to gmx-users-request at gromacs.org.
>> >>> --
>> >>> Gromacs Users mailing list
>> >>>
>> >>> * Please search the archive at http://www.gromacs.org/Support
>> >>> /Mailing_Lists/GMX-Users_List before posting!
>> >>>
>> >>> * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists
>> >>>
>> >>> * For (un)subscribe requests visit
>> >>> https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or
>> >>> send a mail to gmx-users-request at gromacs.org.
>> >>>
>> >> --
>> >> Gromacs Users mailing list
>> >>
>> >> * Please search the archive at http://www.gromacs.org/
>> Support/Mailing_Lists/GMX-Users_List before posting!
>> >>
>> >> * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists
>> >>
>> >> * For (un)subscribe requests visit
>> >> https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or
>> send a mail to gmx-users-request at gromacs.org.
>> > --
>> > Dr Thomas Piggot
>> > Visiting Fellow
>> > University of Southampton, UK.
>> >
>> > --
>> > Gromacs Users mailing list
>> >
>> > * Please search the archive at http://www.gromacs.org/
>> Support/Mailing_Lists/GMX-Users_List before posting!
>> >
>> > * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists
>> >
>> > * For (un)subscribe requests visit
>> > https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or
>> send a mail to gmx-users-request at gromacs.org.
>>
>> --
>> Dr Thomas Piggot
>> Visiting Fellow
>> University of Southampton, UK.
>>
>>
>>
>> ------------------------------
>>
>> Message: 2
>> Date: Fri, 14 Apr 2017 19:23:06 +0000
>> From: Christopher Neale <chris.neale at alum.utoronto.ca>
>> To: "gmx-users at gromacs.org" <gmx-users at gromacs.org>
>> Subject: Re: [gmx-users] Replica exchange simulations more than number
>>         of processor
>> Message-ID:
>>         <SN2PR03MB2301C9C82BC8E629AC3F313CC5050 at SN2PR03MB2301.
>> namprd03.prod.outlook.com>
>>
>> Content-Type: text/plain; charset="gb2312"
>>
>> Dear Thomas:
>>
>> Thanks for the suggestion. I have a way foward, I'm just being selective
>> about what I share at this point. This thread got a bit side-tracked from
>> my reason for posting, which was simply to help a user with a single
>> 12-core node to realize that dedicating it to a 6-month REST2 simulation
>> might not be a good idea.
>> ________________________________________
>> From: gromacs.org_gmx-users-bounces at maillist.sys.kth.se <
>> gromacs.org_gmx-users-bounces at maillist.sys.kth.se> on behalf of Thomas
>> Piggot <t.piggot at soton.ac.uk>
>> Sent: 14 April 2017 15:06:29
>> To: gmx-users at gromacs.org
>> Subject: Re: [gmx-users] Replica exchange simulations more than number of
>> processor
>>
>> Have you thought about a more selective scaling of the interactions
>> (i.e. HREX with only scalings applied for certain bits of the protein,
>> not the complete thing as in REST2) so as to reduce the protein
>> hydrophobicity at the lower lambda's? I've seen this selective scaling
>> work well in other types of system to counteract broadly similar types
>> of sampling issues at lower lambda's.
>>
>> Just a thought, it might not work well for what you are doing,
>>
>> Cheers
>>
>> Tom
>>
>> On 14/04/17 19:27, Christopher Neale wrote:
>> > In REST2, the protein becomes way too hydrophobic as lambda is
>> decreased. Just run a standard simulation at 300K, another at 600K and then
>> another one at 300K with REST2 lambda=0.5 and you'll see what I mean.
>> >
>> > REST has a hidden barrier corresponding to the point at which water
>> becomes a theta solvent and the exchange is very poor across that boundary,
>> with globular replicas staying at higher lambda and extended replicas
>> staying at lower lambda. However, with REST2, there is no such hidden
>> barrier because water never gets anywhere near being a theta solvent. So in
>> REST2 you get good exchange for all the wrong reasons.
>> >
>> > Work continues....
>> >
>> >
>> > ________________________________________
>> > From: gromacs.org_gmx-users-bounces at maillist.sys.kth.se <
>> gromacs.org_gmx-users-bounces at maillist.sys.kth.se> on behalf of Thomas
>> Piggot <t.piggot at soton.ac.uk>
>> > Sent: 14 April 2017 14:16:26
>> > To: gmx-users at gromacs.org
>> > Subject: Re: [gmx-users] Replica exchange simulations more than number
>> of processor
>> >
>> > Hi Chris,
>> >
>> > This is interesting. Do you have an idea why the sampling at low lambda
>> > values isn't very good?
>> >
>> > Cheers
>> >
>> > Tom
>> >
>> > On 14/04/17 17:40, Christopher Neale wrote:
>> >> What I have is only my own personal unpublished data. But add to that
>> the number of years since REST2 came out and the low number of papers in
>> which anybody did something useful with it and I'd say that suggests many
>> others have tried and run into the same issues that I did (or maybe I'm
>> just not reading the right papers, so please correct me if I am wrong).
>> When my desired use failed I fell back to trp-cage folding in water and
>> found the same things, so it's not like REST2 only fails for some exotic
>> systems. Running 3 entirely separate repeats of regular T-REMD and 3 of
>> REST2, I found that overall T-REMD is of similar or perhaps better
>> efficiency (even though it has more replicas) compared to REST2. It's an
>> on/off area of research for me, attempting to fix REST, but I still don't
>> have a good answer for it.
>> >>
>> >>
>> >> ________________________________________
>> >> From: gromacs.org_gmx-users-bounces at maillist.sys.kth.se <
>> gromacs.org_gmx-users-bounces at maillist.sys.kth.se> on behalf of Szil?rd
>> P?ll <pall.szilard at gmail.com>
>> >> Sent: 14 April 2017 12:26:41
>> >> To: Discussion list for GROMACS users
>> >> Subject: Re: [gmx-users] Replica exchange simulations more than number
>> of       processor
>> >>
>> >> On Fri, Apr 14, 2017 at 5:51 PM, Christopher Neale <
>> >> chris.neale at alum.utoronto.ca> wrote:
>> >>
>> >>> Just my 2 cents: REST2 doesn't work. The random walk is much better
>> than
>> >>> REST but the sampling efficiency at low lambda is now poor.
>> >> Interesting. Is there any (published) data you could point to that
>> supports
>> >> this?
>> >>
>> >>
>> >>> Your mileage may vary, but be aware that REST2 is not necessarily
>> going to
>> >>> be useful for all systems.
>> >>> ________________________________________
>> >>> From: gromacs.org_gmx-users-bounces at maillist.sys.kth.se <
>> >>> gromacs.org_gmx-users-bounces at maillist.sys.kth.se> on behalf of Thomas
>> >>> Piggot <t.piggot at soton.ac.uk>
>> >>> Sent: 14 April 2017 09:23:35
>> >>> To: gmx-users at gromacs.org
>> >>> Subject: Re: [gmx-users] Replica exchange simulations more than number
>> of
>> >>> processor
>> >>>
>> >>> In addition, using another technique like REST2 may substantially
>> reduce
>> >>> the numbers of replica's you need. Still, more compute is definitely
>> >>> recommended if you can get it.
>> >>>
>> >>> Cheers
>> >>>
>> >>> Tom
>> >>>
>> >>> On 14/04/17 12:48, Smith, Micholas D. wrote:
>> >>>> You can try over assignment of cores, but that will be horribly slow.
>> Or
>> >>> use more than 1 node (if you can get it).
>> >>>> ===================
>> >>>> Micholas Dean Smith, PhD.
>> >>>> Post-doctoral Research Associate
>> >>>> University of Tennessee/Oak Ridge National Laboratory
>> >>>> Center for Molecular Biophysics
>> >>>>
>> >>>> ________________________________________
>> >>>> From: gromacs.org_gmx-users-bounces at maillist.sys.kth.se <
>> >>> gromacs.org_gmx-users-bounces at maillist.sys.kth.se> on behalf of ?????
>> >>> ????????? <b.mijiddorj at gmail.com>
>> >>>> Sent: Friday, April 14, 2017 4:16 AM
>> >>>> To: gromacs.org_gmx-users at maillist.sys.kth.se
>> >>>> Subject: [gmx-users] Replica exchange simulations more than number of
>> >>>    processor
>> >>>> Dear gmx users,
>> >>>>
>> >>>> I would like to simulate folding of a peptides. I have only 12 core
>> >>>> processor, and I assumed the number of replica using temperature
>> >>> generator
>> >>>> as following link
>> >>>> http://folding.bmc.uu.se/remd/.
>> >>>>
>> >>>> The number of replica is about 60. How can I solve this problem? Can
>> you
>> >>>> advice me, please?
>> >>>>
>> >>>> Best regards,
>> >>>> Mijee
>> >>>> --
>> >>>> Gromacs Users mailing list
>> >>>>
>> >>>> * Please search the archive at http://www.gromacs.org/Support
>> >>> /Mailing_Lists/GMX-Users_List before posting!
>> >>>> * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists
>> >>>>
>> >>>> * For (un)subscribe requests visit
>> >>>> https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or
>> >>> send a mail to gmx-users-request at gromacs.org.
>> >>> --
>> >>> Dr Thomas Piggot
>> >>> Visiting Fellow
>> >>> University of Southampton, UK.
>> >>>
>> >>> --
>> >>> Gromacs Users mailing list
>> >>>
>> >>> * Please search the archive at http://www.gromacs.org/Support
>> >>> /Mailing_Lists/GMX-Users_List before posting!
>> >>>
>> >>> * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists
>> >>>
>> >>> * For (un)subscribe requests visit
>> >>> https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or
>> >>> send a mail to gmx-users-request at gromacs.org.
>> >>> --
>> >>> Gromacs Users mailing list
>> >>>
>> >>> * Please search the archive at http://www.gromacs.org/Support
>> >>> /Mailing_Lists/GMX-Users_List before posting!
>> >>>
>> >>> * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists
>> >>>
>> >>> * For (un)subscribe requests visit
>> >>> https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or
>> >>> send a mail to gmx-users-request at gromacs.org.
>> >>>
>> >> --
>> >> Gromacs Users mailing list
>> >>
>> >> * Please search the archive at http://www.gromacs.org/
>> Support/Mailing_Lists/GMX-Users_List before posting!
>> >>
>> >> * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists
>> >>
>> >> * For (un)subscribe requests visit
>> >> https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or
>> send a mail to gmx-users-request at gromacs.org.
>> > --
>> > Dr Thomas Piggot
>> > Visiting Fellow
>> > University of Southampton, UK.
>> >
>> > --
>> > Gromacs Users mailing list
>> >
>> > * Please search the archive at http://www.gromacs.org/
>> Support/Mailing_Lists/GMX-Users_List before posting!
>> >
>> > * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists
>> >
>> > * For (un)subscribe requests visit
>> > https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or
>> send a mail to gmx-users-request at gromacs.org.
>>
>> --
>> Dr Thomas Piggot
>> Visiting Fellow
>> University of Southampton, UK.
>>
>> --
>> Gromacs Users mailing list
>>
>> * Please search the archive at http://www.gromacs.org/
>> Support/Mailing_Lists/GMX-Users_List before posting!
>>
>> * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists
>>
>> * For (un)subscribe requests visit
>> https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or
>> send a mail to gmx-users-request at gromacs.org.
>>
>> ------------------------------
>>
>> --
>> Gromacs Users mailing list
>>
>> * Please search the archive at http://www.gromacs.org/
>> Support/Mailing_Lists/GMX-Users_List before posting!
>>
>> * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists
>>
>> * For (un)subscribe requests visit
>> https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or
>> send a mail to gmx-users-request at gromacs.org.
>>
>> End of gromacs.org_gmx-users Digest, Vol 156, Issue 73
>> ******************************************************
>>
> --
> Gromacs Users mailing list
>
> * Please search the archive at http://www.gromacs.org/Support/Mailing_Lists/GMX-Users_List before posting!
>
> * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists
>
> * For (un)subscribe requests visit
> https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or send a mail to gmx-users-request at gromacs.org.

More information about the gromacs.org_gmx-users mailing list