Re: [hwloc-users] NetLoc subnets Problem

[Brice Goglin]

Cyril would know better but he's currently on vacation. I don't think
the required scotch release is already officially publicly available online.

I am not sure what problem you're trying to solve here. This feature
isn't really useful if you only have two nodes. It was rather designed
for process placement on large clusters (by the way, you'll need
something to build a communication pattern of your MPI application as a
communication matrix).

Regarding configure, once the right scotch release is installed, there's
no configure options if I remember correctly, you just need to point
your compiler to scotch include and lib directories. Something like this
may help:
export C_INCLUDE_PATH=/opt/scotch-6/include
export LD_LIBRARY_PATH=/opt/scotch-6/lib
export LIBRARY_PATH=/opt/scotch-6/lib

Brice



Le 22/02/2017 13:38, Михаил Халилов a écrit :
> I tried to configure hwloc with scotch, but I still haven't success
> with that case. I read Chapter 18 in doxygen and chapters about Netloc
> and installation, but not found about anything about Scotch configure.
> So, Scotch installed in /opt/scotch-6/ folder, and I want to install
> hwloc with netloc in /opt/hwloc/ . Which options should I use to give
> ./configure script information about Scotch?
>
> Best regards,
> Mikhail
>
> 2017-02-20 11:50 GMT+03:00 Brice Goglin  >:
>
> Inside the tarball that you downloaded, there's a
> doc/doxygen-doc/hwloc-a4.pdf with chapter 18 about Netloc with Scotch.
> Beware that this code is still under development.
>
> Brice
>
>
>
>
> Le 19/02/2017 20:20, Михаил Халилов a écrit :
>> Okay, but what configure options for Scotch should I use? I
>> didn't found any information about it in docs and readme
>>
>>
>>
>> 2017-02-19 20:52 GMT+03:00 Brice Goglin > >:
>>
>> The only publicly-installed netloc API is currently specific
>> to the scotch partitioner for process placement. It takes a
>> network topology and a communication pattern between a set of
>> process and it generates a topology-aware placement for these
>> processes.
>> This API only gets installed if you have scotch installed
>> (and tell configure where it is). That's why you don't get
>> any netloc API installed for now.
>>
>> We initially exposed the entire graph that netloc uses
>> internally (it's still true in v0.5 but not anymore in hwloc
>> 2.0) but there wasn't a clear list of what users want to do
>> with it. We didn't want to expose a random API without much
>> user feedback first. There are many ways to expose a graph
>> API, it was too risky. So it's not publicly installed anymore.
>>
>> You can use internal headers such as private/netloc.h for now
>> (you'll see edges, nodes, etc) and we'll make things public
>> once we know what you and others would like to do.
>>
>> Brice
>>
>>
>>
>>
>> Le 19/02/2017 17:29, Михаил Халилов a écrit :
>>> Hi again!
>>>
>>> Can I ask you, how can I use netloc API for my C programs?
>>> I configured hwloc only with --prefix=/opt/hwloc option. So,
>>> there are no netloc header files in /opt/hwloc/include
>>> directory. Also, I didn't understand how to use
>>> netloc_draw.html, because I found it only in extracted
>>> tarball. May be i should configure netloc with some other
>>> options?
>>>
>>> Best regards,
>>> Mikhail Khalilov
>>>
>>>
>>>
>>> ___
>>> hwloc-users mailing list
>>> hwloc-users@lists.open-mpi.org
>>> 
>>> https://rfd.newmexicoconsortium.org/mailman/listinfo/hwloc-users
>>> 
>> ___ hwloc-users
>> mailing list hwloc-users@lists.open-mpi.org
>> 
>> https://rfd.newmexicoconsortium.org/mailman/listinfo/hwloc-users
>> 
>>
>>
>> ___
>> hwloc-users mailing list
>> hwloc-users@lists.open-mpi.org
>> 
>> https://rfd.newmexicoconsortium.org/mailman/listinfo/hwloc-users
>> 
> ___ hwloc-users
> mailing list hwloc-users@lists.open-mpi.org
> 
> https://rfd.newmexicoconsortium.org/mailman/listinfo/hwloc-users
>  
>
> 

Re: [hwloc-users] NetLoc subnets Problem

[Михаил Халилов]

I tried to configure hwloc with scotch, but I still haven't success with
that case. I read Chapter 18 in doxygen and chapters about Netloc and
installation, but not found about anything about Scotch configure.
So, Scotch installed in /opt/scotch-6/ folder, and I want to install hwloc
with netloc in /opt/hwloc/ . Which options should I use to give ./configure
script information about Scotch?

Best regards,
Mikhail

2017-02-20 11:50 GMT+03:00 Brice Goglin :

> Inside the tarball that you downloaded, there's a
> doc/doxygen-doc/hwloc-a4.pdf with chapter 18 about Netloc with Scotch.
> Beware that this code is still under development.
>
> Brice
>
>
>
>
> Le 19/02/2017 20:20, Михаил Халилов a écrit :
>
> Okay, but what configure options for Scotch should I use? I didn't found
> any information about it in docs and readme
>
>
>
> 2017-02-19 20:52 GMT+03:00 Brice Goglin :
>
>> The only publicly-installed netloc API is currently specific to the
>> scotch partitioner for process placement. It takes a network topology and a
>> communication pattern between a set of process and it generates a
>> topology-aware placement for these processes.
>> This API only gets installed if you have scotch installed (and tell
>> configure where it is). That's why you don't get any netloc API installed
>> for now.
>>
>> We initially exposed the entire graph that netloc uses internally (it's
>> still true in v0.5 but not anymore in hwloc 2.0) but there wasn't a clear
>> list of what users want to do with it. We didn't want to expose a random
>> API without much user feedback first. There are many ways to expose a graph
>> API, it was too risky. So it's not publicly installed anymore.
>>
>> You can use internal headers such as private/netloc.h for now (you'll see
>> edges, nodes, etc) and we'll make things public once we know what you and
>> others would like to do.
>>
>> Brice
>>
>>
>>
>>
>> Le 19/02/2017 17:29, Михаил Халилов a écrit :
>>
>> Hi again!
>>
>> Can I ask you, how can I use netloc API for my C programs?
>> I configured hwloc only with --prefix=/opt/hwloc option. So, there are no
>> netloc header files in /opt/hwloc/include directory. Also, I didn't
>> understand how to use netloc_draw.html, because I found it only in
>> extracted tarball. May be i should configure netloc with some other
>> options?
>>
>> Best regards,
>> Mikhail Khalilov
>>
>>
>>
>> ___
>> hwloc-users mailing 
>> listhwloc-us...@lists.open-mpi.orghttps://rfd.newmexicoconsortium.org/mailman/listinfo/hwloc-users
>>
>> ___ hwloc-users mailing list
>> hwloc-users@lists.open-mpi.org https://rfd.newmexicoconsortiu
>> m.org/mailman/listinfo/hwloc-users
>
> ___
> hwloc-users mailing 
> listhwloc-us...@lists.open-mpi.orghttps://rfd.newmexicoconsortium.org/mailman/listinfo/hwloc-users
>
>
> ___
> hwloc-users mailing list
> hwloc-users@lists.open-mpi.org
> https://rfd.newmexicoconsortium.org/mailman/listinfo/hwloc-users
>
___
hwloc-users mailing list
hwloc-users@lists.open-mpi.org
https://rfd.newmexicoconsortium.org/mailman/listinfo/hwloc-users

Re: [hwloc-users] NetLoc subnets Problem

[Brice Goglin]

Inside the tarball that you downloaded, there's a
doc/doxygen-doc/hwloc-a4.pdf with chapter 18 about Netloc with Scotch.
Beware that this code is still under development.

Brice



Le 19/02/2017 20:20, Михаил Халилов a écrit :
> Okay, but what configure options for Scotch should I use? I didn't
> found any information about it in docs and readme
>
>
>
> 2017-02-19 20:52 GMT+03:00 Brice Goglin  >:
>
> The only publicly-installed netloc API is currently specific to
> the scotch partitioner for process placement. It takes a network
> topology and a communication pattern between a set of process and
> it generates a topology-aware placement for these processes.
> This API only gets installed if you have scotch installed (and
> tell configure where it is). That's why you don't get any netloc
> API installed for now.
>
> We initially exposed the entire graph that netloc uses internally
> (it's still true in v0.5 but not anymore in hwloc 2.0) but there
> wasn't a clear list of what users want to do with it. We didn't
> want to expose a random API without much user feedback first.
> There are many ways to expose a graph API, it was too risky. So
> it's not publicly installed anymore.
>
> You can use internal headers such as private/netloc.h for now
> (you'll see edges, nodes, etc) and we'll make things public once
> we know what you and others would like to do.
>
> Brice
>
>
>
>
> Le 19/02/2017 17:29, Михаил Халилов a écrit :
>> Hi again!
>>
>> Can I ask you, how can I use netloc API for my C programs?
>> I configured hwloc only with --prefix=/opt/hwloc option. So,
>> there are no netloc header files in /opt/hwloc/include directory.
>> Also, I didn't understand how to use netloc_draw.html, because I
>> found it only in extracted tarball. May be i should configure
>> netloc with some other options?
>>
>> Best regards,
>> Mikhail Khalilov
>>
>>
>>
>> ___
>> hwloc-users mailing list
>> hwloc-users@lists.open-mpi.org
>> 
>> https://rfd.newmexicoconsortium.org/mailman/listinfo/hwloc-users
>> 
> ___ hwloc-users
> mailing list hwloc-users@lists.open-mpi.org
> 
> https://rfd.newmexicoconsortium.org/mailman/listinfo/hwloc-users
>  
>
> ___
> hwloc-users mailing list
> hwloc-users@lists.open-mpi.org
> https://rfd.newmexicoconsortium.org/mailman/listinfo/hwloc-users
___
hwloc-users mailing list
hwloc-users@lists.open-mpi.org
https://rfd.newmexicoconsortium.org/mailman/listinfo/hwloc-users

Re: [hwloc-users] NetLoc subnets Problem

[Михаил Халилов]

Okay, but what configure options for Scotch should I use? I didn't found
any information about it in docs and readme



2017-02-19 20:52 GMT+03:00 Brice Goglin :

> The only publicly-installed netloc API is currently specific to the scotch
> partitioner for process placement. It takes a network topology and a
> communication pattern between a set of process and it generates a
> topology-aware placement for these processes.
> This API only gets installed if you have scotch installed (and tell
> configure where it is). That's why you don't get any netloc API installed
> for now.
>
> We initially exposed the entire graph that netloc uses internally (it's
> still true in v0.5 but not anymore in hwloc 2.0) but there wasn't a clear
> list of what users want to do with it. We didn't want to expose a random
> API without much user feedback first. There are many ways to expose a graph
> API, it was too risky. So it's not publicly installed anymore.
>
> You can use internal headers such as private/netloc.h for now (you'll see
> edges, nodes, etc) and we'll make things public once we know what you and
> others would like to do.
>
> Brice
>
>
>
>
> Le 19/02/2017 17:29, Михаил Халилов a écrit :
>
> Hi again!
>
> Can I ask you, how can I use netloc API for my C programs?
> I configured hwloc only with --prefix=/opt/hwloc option. So, there are no
> netloc header files in /opt/hwloc/include directory. Also, I didn't
> understand how to use netloc_draw.html, because I found it only in
> extracted tarball. May be i should configure netloc with some other
> options?
>
> Best regards,
> Mikhail Khalilov
>
>
>
> ___
> hwloc-users mailing 
> listhwloc-us...@lists.open-mpi.orghttps://rfd.newmexicoconsortium.org/mailman/listinfo/hwloc-users
>
>
>
> ___
> hwloc-users mailing list
> hwloc-users@lists.open-mpi.org
> https://rfd.newmexicoconsortium.org/mailman/listinfo/hwloc-users
>
___
hwloc-users mailing list
hwloc-users@lists.open-mpi.org
https://rfd.newmexicoconsortium.org/mailman/listinfo/hwloc-users

Re: [hwloc-users] NetLoc subnets Problem

[Brice Goglin]

The only publicly-installed netloc API is currently specific to the
scotch partitioner for process placement. It takes a network topology
and a communication pattern between a set of process and it generates a
topology-aware placement for these processes.
This API only gets installed if you have scotch installed (and tell
configure where it is). That's why you don't get any netloc API
installed for now.

We initially exposed the entire graph that netloc uses internally (it's
still true in v0.5 but not anymore in hwloc 2.0) but there wasn't a
clear list of what users want to do with it. We didn't want to expose a
random API without much user feedback first. There are many ways to
expose a graph API, it was too risky. So it's not publicly installed
anymore.

You can use internal headers such as private/netloc.h for now (you'll
see edges, nodes, etc) and we'll make things public once we know what
you and others would like to do.

Brice



Le 19/02/2017 17:29, Михаил Халилов a écrit :
> Hi again!
>
> Can I ask you, how can I use netloc API for my C programs?
> I configured hwloc only with --prefix=/opt/hwloc option. So, there are
> no netloc header files in /opt/hwloc/include directory. Also, I didn't
> understand how to use netloc_draw.html, because I found it only in
> extracted tarball. May be i should configure netloc with some other
> options?
>
> Best regards,
> Mikhail Khalilov
>
>
>
> ___
> hwloc-users mailing list
> hwloc-users@lists.open-mpi.org
> https://rfd.newmexicoconsortium.org/mailman/listinfo/hwloc-users

___
hwloc-users mailing list
hwloc-users@lists.open-mpi.org
https://rfd.newmexicoconsortium.org/mailman/listinfo/hwloc-users

Re: [hwloc-users] NetLoc subnets Problem

[Михаил Халилов]

Hi again!

Can I ask you, how can I use netloc API for my C programs?
I configured hwloc only with --prefix=/opt/hwloc option. So, there are no
netloc header files in /opt/hwloc/include directory. Also, I didn't
understand how to use netloc_draw.html, because I found it only in
extracted tarball. May be i should configure netloc with some other
options?

Best regards,
Mikhail Khalilov
___
hwloc-users mailing list
hwloc-users@lists.open-mpi.org
https://rfd.newmexicoconsortium.org/mailman/listinfo/hwloc-users

Re: [hwloc-users] NetLoc subnets Problem

[Brice Goglin]

Please run "hwloc-gather-topology --io foo" on the head node 'Use the
gather-topology script that comes with the hwloc nightly snapshot).

The output tarball will likely be large, so feel free to send it to me
directly.

My guess is that your very old kernel may miss some files that we need
in the hwloc development snapshot (the I/O discovery changed
significantly in hwloc 2.0).

Brice




Le 17/02/2017 10:26, Михаил Халилов a écrit :
> I ran ibstat on head node it gives information in attach.
>
> 2017-02-17 12:16 GMT+03:00 Brice Goglin  >:
>
> For some reason, lstopo didn't find any InfiniBand information on
> the head node. I guess running lstopo won't show any "mlx4_0" or
> "ib0" object. Is the InfiniBand service really running on that
> machine?
>
> Brice
>
>
>
>
>
> Le 17/02/2017 10:04, Михаил Халилов a écrit :
>> All files in attach. I run netloc_ib_gather_raw with this
>> parameters netloc_ib_gather_raw /home/halilov/mycluster-data/
>> --hwloc-dir=/home/halilov/mycluster-data/hwloc/ --verbose --sudo
>>
>> 2017-02-17 11:55 GMT+03:00 Brice Goglin > >:
>>
>> Please copy-paste the exact command line of your
>> "netloc_ib_gather_raw" and all the messages it printed. And
>> also send the output of the hwloc directory it created (it
>> will contain the lstopo XML output of the node where you ran
>> the command).
>>
>> Brice
>>
>>
>>
>> Le 17/02/2017 09:51, Михаил Халилов a écrit :
>>> I installed nightly tarball, but it still isn't working. In
>>> attach info of ibnetdiscover and ibroute. May be it wlii help...
>>> What could be the problem?
>>>
>>> Best regards,
>>> Mikhail Khalilov
>>>
>>> 2017-02-17 9:53 GMT+03:00 Brice Goglin
>>> >:
>>>
>>> Hello
>>>
>>> As identicated on the netloc webpages, the netloc
>>> development now occurs
>>> inside the hwloc git tree. netloc v0.5 is obsolete even
>>> if hwloc 2.0
>>> isn't released yet.
>>>
>>> If you want to use a development snapshot, take hwloc
>>> nightly tarballs
>>> from https://ci.inria.fr/hwloc/job/master-0-tarball/
>>>  or
>>> https://www.open-mpi.org/software/hwloc/nightly/master/
>>> 
>>>
>>> Regards
>>> Brice
>>>
>>>
>>>
>>>
>>>
>>> Le 16/02/2017 19:15, miharuli...@gmail.com
>>>  a écrit :
>>> > I downloaded gunzip from openmpi site here:
>>> https://www.open-mpi.org/software/netloc/v0.5/
>>> >
>>> > There are three identical machines in my cluster, but
>>> now third node is broken, and i tried on two machines.
>>> They all connected by InfiniBand switch, and when I try
>>> to use ibnetdiscovery or ibroute, it works perfectly...
>>> >
>>> >
>>> >
>>> > Отправлено с iPad
>>> >> 16 февр. 2017 г., в 18:40, Cyril Bordage
>>> >
>>> написал(а):
>>> >>
>>> >> Hi,
>>> >>
>>> >> What version did you use?
>>> >>
>>> >> I pushed some commits on master on ompi repository.
>>> With this version it
>>> >> seems to work.
>>> >> You have two machines because you tried netloc on
>>> these two?
>>> >>
>>> >>
>>> >> Cyril.
>>> >>
>>> >>> Le 15/02/2017 à 22:44, miharulidze a écrit :
>>> >>> Hi!
>>> >>>
>>> >>> I'm trying to use NetLoc tool for detecting my
>>> cluster topology.
>>> >>>
>>> >>> I have 2 node cluster with AMD Processors, connected
>>> by InfiniBand. Also
>>> >>> I installed latest versions of hwloc and netloc tools.
>>> >>>
>>> >>> I followed the instruction of netloc and when I
>>> tried to use
>>> >>> netloc_ib_gather_raw as root, i recieved this message
>>> >>> root:$ netloc_ib_gather_raw
>>> >>> --out-dir=/home/halilov/mycluster-data/result/
>>> >>> --hwloc-dir=/home/halilov/mycluster-data/hwloc/ --sudo
>>> >>>
>>> >>> Found 0 subnets in hwloc directory:
>>> >>>
>>> >>>
>>> >>> There are two files in
>>> /home/halilov/mycluster-data/hwloc/ generated by
>>> >>> 

Re: [hwloc-users] NetLoc subnets Problem

[Михаил Халилов]

I ran ibstat on head node it gives information in attach.

2017-02-17 12:16 GMT+03:00 Brice Goglin :

> For some reason, lstopo didn't find any InfiniBand information on the head
> node. I guess running lstopo won't show any "mlx4_0" or "ib0" object. Is
> the InfiniBand service really running on that machine?
>
> Brice
>
>
>
>
>
> Le 17/02/2017 10:04, Михаил Халилов a écrit :
>
> All files in attach. I run netloc_ib_gather_raw with this parameters
> netloc_ib_gather_raw /home/halilov/mycluster-data/
> --hwloc-dir=/home/halilov/mycluster-data/hwloc/ --verbose --sudo
>
> 2017-02-17 11:55 GMT+03:00 Brice Goglin :
>
>> Please copy-paste the exact command line of your "netloc_ib_gather_raw"
>> and all the messages it printed. And also send the output of the hwloc
>> directory it created (it will contain the lstopo XML output of the node
>> where you ran the command).
>>
>> Brice
>>
>>
>>
>> Le 17/02/2017 09:51, Михаил Халилов a écrit :
>>
>> I installed nightly tarball, but it still isn't working. In attach info
>> of ibnetdiscover and ibroute. May be it wlii help...
>> What could be the problem?
>>
>> Best regards,
>> Mikhail Khalilov
>>
>> 2017-02-17 9:53 GMT+03:00 Brice Goglin < 
>> brice.gog...@inria.fr>:
>>
>>> Hello
>>>
>>> As identicated on the netloc webpages, the netloc development now occurs
>>> inside the hwloc git tree. netloc v0.5 is obsolete even if hwloc 2.0
>>> isn't released yet.
>>>
>>> If you want to use a development snapshot, take hwloc nightly tarballs
>>> from https://ci.inria.fr/hwloc/job/master-0-tarball/ or
>>> https://www.open-mpi.org/software/hwloc/nightly/master/
>>>
>>> Regards
>>> Brice
>>>
>>>
>>>
>>>
>>>
>>> Le 16/02/2017 19:15, miharuli...@gmail.com a
>>> écrit :
>>> > I downloaded gunzip from openmpi site here: https://www.open-mpi.org/
>>> software/netloc/v0.5/
>>> >
>>> > There are three identical machines in my cluster, but now third node
>>> is broken, and i tried on two machines. They all connected by InfiniBand
>>> switch, and when I try to use ibnetdiscovery or ibroute, it works
>>> perfectly...
>>> >
>>> >
>>> >
>>> > Отправлено с iPad
>>> >> 16 февр. 2017 г., в 18:40, Cyril Bordage 
>>> написал(а):
>>> >>
>>> >> Hi,
>>> >>
>>> >> What version did you use?
>>> >>
>>> >> I pushed some commits on master on ompi repository. With this version
>>> it
>>> >> seems to work.
>>> >> You have two machines because you tried netloc on these two?
>>> >>
>>> >>
>>> >> Cyril.
>>> >>
>>> >>> Le 15/02/2017 à 22:44, miharulidze a écrit :
>>> >>> Hi!
>>> >>>
>>> >>> I'm trying to use NetLoc tool for detecting my cluster topology.
>>> >>>
>>> >>> I have 2 node cluster with AMD Processors, connected by InfiniBand.
>>> Also
>>> >>> I installed latest versions of hwloc and netloc tools.
>>> >>>
>>> >>> I followed the instruction of netloc and when I tried to use
>>> >>> netloc_ib_gather_raw as root, i recieved this message
>>> >>> root:$ netloc_ib_gather_raw
>>> >>> --out-dir=/home/halilov/mycluster-data/result/
>>> >>> --hwloc-dir=/home/halilov/mycluster-data/hwloc/ --sudo
>>> >>>
>>> >>> Found 0 subnets in hwloc directory:
>>> >>>
>>> >>>
>>> >>> There are two files in /home/halilov/mycluster-data/hwloc/
>>> generated by
>>> >>> hwloc: head.xml and node01.xml
>>> >>>
>>> >>> P.S. in attach archieve with .xml files
>>> >>>
>>> >>>
>>> >>> Best regards,
>>> >>> Mikhail Khalilov
>>> >>>
>>> >>>
>>> >>>
>>> >>> ___
>>> >>> hwloc-users mailing list
>>> >>> hwloc-users@lists.open-mpi.org
>>> >>> https://rfd.newmexicoconsortium.org/mailman/listinfo/hwloc-users
>>> >> ___
>>> >> hwloc-users mailing list
>>> >> hwloc-users@lists.open-mpi.org
>>> >> https://rfd.newmexicoconsortium.org/mailman/listinfo/hwloc-users
>>> > ___
>>> > hwloc-users mailing list
>>> > hwloc-users@lists.open-mpi.org
>>> > https://rfd.newmexicoconsortium.org/mailman/listinfo/hwloc-users
>>>
>>> ___
>>> hwloc-users mailing list
>>> hwloc-users@lists.open-mpi.org
>>> https://rfd.newmexicoconsortium.org/mailman/listinfo/hwloc-users
>>>
>>
>>
>>
>> ___
>> hwloc-users mailing 
>> listhwloc-us...@lists.open-mpi.orghttps://rfd.newmexicoconsortium.org/mailman/listinfo/hwloc-users
>>
>> ___ hwloc-users mailing list
>> hwloc-users@lists.open-mpi.org https://rfd.newmexicoconsortiu
>> m.org/mailman/listinfo/hwloc-users
>
> ___
> hwloc-users mailing 
> listhwloc-us...@lists.open-mpi.orghttps://rfd.newmexicoconsortium.org/mailman/listinfo/hwloc-users
>
>
> ___
> hwloc-users mailing list
> hwloc-users@lists.open-mpi.org
> 

Re: [hwloc-users] NetLoc subnets Problem

[Brice Goglin]

For some reason, lstopo didn't find any InfiniBand information on the
head node. I guess running lstopo won't show any "mlx4_0" or "ib0"
object. Is the InfiniBand service really running on that machine?

Brice




Le 17/02/2017 10:04, Михаил Халилов a écrit :
> All files in attach. I run netloc_ib_gather_raw with this parameters
> netloc_ib_gather_raw /home/halilov/mycluster-data/
> --hwloc-dir=/home/halilov/mycluster-data/hwloc/ --verbose --sudo
>
> 2017-02-17 11:55 GMT+03:00 Brice Goglin  >:
>
> Please copy-paste the exact command line of your
> "netloc_ib_gather_raw" and all the messages it printed. And also
> send the output of the hwloc directory it created (it will contain
> the lstopo XML output of the node where you ran the command).
>
> Brice
>
>
>
> Le 17/02/2017 09:51, Михаил Халилов a écrit :
>> I installed nightly tarball, but it still isn't working. In
>> attach info of ibnetdiscover and ibroute. May be it wlii help...
>> What could be the problem?
>>
>> Best regards,
>> Mikhail Khalilov
>>
>> 2017-02-17 9:53 GMT+03:00 Brice Goglin > >:
>>
>> Hello
>>
>> As identicated on the netloc webpages, the netloc development
>> now occurs
>> inside the hwloc git tree. netloc v0.5 is obsolete even if
>> hwloc 2.0
>> isn't released yet.
>>
>> If you want to use a development snapshot, take hwloc nightly
>> tarballs
>> from https://ci.inria.fr/hwloc/job/master-0-tarball/
>>  or
>> https://www.open-mpi.org/software/hwloc/nightly/master/
>> 
>>
>> Regards
>> Brice
>>
>>
>>
>>
>>
>> Le 16/02/2017 19:15, miharuli...@gmail.com
>>  a écrit :
>> > I downloaded gunzip from openmpi site here:
>> https://www.open-mpi.org/software/netloc/v0.5/
>> >
>> > There are three identical machines in my cluster, but now
>> third node is broken, and i tried on two machines. They all
>> connected by InfiniBand switch, and when I try to use
>> ibnetdiscovery or ibroute, it works perfectly...
>> >
>> >
>> >
>> > Отправлено с iPad
>> >> 16 февр. 2017 г., в 18:40, Cyril Bordage
>> >
>> написал(а):
>> >>
>> >> Hi,
>> >>
>> >> What version did you use?
>> >>
>> >> I pushed some commits on master on ompi repository. With
>> this version it
>> >> seems to work.
>> >> You have two machines because you tried netloc on these two?
>> >>
>> >>
>> >> Cyril.
>> >>
>> >>> Le 15/02/2017 à 22:44, miharulidze a écrit :
>> >>> Hi!
>> >>>
>> >>> I'm trying to use NetLoc tool for detecting my cluster
>> topology.
>> >>>
>> >>> I have 2 node cluster with AMD Processors, connected by
>> InfiniBand. Also
>> >>> I installed latest versions of hwloc and netloc tools.
>> >>>
>> >>> I followed the instruction of netloc and when I tried to use
>> >>> netloc_ib_gather_raw as root, i recieved this message
>> >>> root:$ netloc_ib_gather_raw
>> >>> --out-dir=/home/halilov/mycluster-data/result/
>> >>> --hwloc-dir=/home/halilov/mycluster-data/hwloc/ --sudo
>> >>>
>> >>> Found 0 subnets in hwloc directory:
>> >>>
>> >>>
>> >>> There are two files in
>> /home/halilov/mycluster-data/hwloc/ generated by
>> >>> hwloc: head.xml and node01.xml
>> >>>
>> >>> P.S. in attach archieve with .xml files
>> >>>
>> >>>
>> >>> Best regards,
>> >>> Mikhail Khalilov
>> >>>
>> >>>
>> >>>
>> >>> ___
>> >>> hwloc-users mailing list
>> >>> hwloc-users@lists.open-mpi.org
>> 
>> >>>
>> https://rfd.newmexicoconsortium.org/mailman/listinfo/hwloc-users
>> 
>> >> ___
>> >> hwloc-users mailing list
>> >> hwloc-users@lists.open-mpi.org
>> 
>> >>
>> https://rfd.newmexicoconsortium.org/mailman/listinfo/hwloc-users
>> 
>> > ___
>> > hwloc-users mailing list
>> > hwloc-users@lists.open-mpi.org
>>   

Re: [hwloc-users] NetLoc subnets Problem

[Михаил Халилов]

All files in attach. I run netloc_ib_gather_raw with this parameters
netloc_ib_gather_raw /home/halilov/mycluster-data/
--hwloc-dir=/home/halilov/mycluster-data/hwloc/ --verbose --sudo

2017-02-17 11:55 GMT+03:00 Brice Goglin :

> Please copy-paste the exact command line of your "netloc_ib_gather_raw"
> and all the messages it printed. And also send the output of the hwloc
> directory it created (it will contain the lstopo XML output of the node
> where you ran the command).
>
> Brice
>
>
>
> Le 17/02/2017 09:51, Михаил Халилов a écrit :
>
> I installed nightly tarball, but it still isn't working. In attach info of
> ibnetdiscover and ibroute. May be it wlii help...
> What could be the problem?
>
> Best regards,
> Mikhail Khalilov
>
> 2017-02-17 9:53 GMT+03:00 Brice Goglin :
>
>> Hello
>>
>> As identicated on the netloc webpages, the netloc development now occurs
>> inside the hwloc git tree. netloc v0.5 is obsolete even if hwloc 2.0
>> isn't released yet.
>>
>> If you want to use a development snapshot, take hwloc nightly tarballs
>> from https://ci.inria.fr/hwloc/job/master-0-tarball/ or
>> https://www.open-mpi.org/software/hwloc/nightly/master/
>>
>> Regards
>> Brice
>>
>>
>>
>>
>>
>> Le 16/02/2017 19:15, miharuli...@gmail.com a écrit :
>> > I downloaded gunzip from openmpi site here:
>> https://www.open-mpi.org/
>> software/netloc/v0.5/
>> >
>> > There are three identical machines in my cluster, but now third node is
>> broken, and i tried on two machines. They all connected by InfiniBand
>> switch, and when I try to use ibnetdiscovery or ibroute, it works
>> perfectly...
>> >
>> >
>> >
>> > Отправлено с iPad
>> >> 16 февр. 2017 г., в 18:40, Cyril Bordage < 
>> cyril.bord...@inria.fr> написал(а):
>> >>
>> >> Hi,
>> >>
>> >> What version did you use?
>> >>
>> >> I pushed some commits on master on ompi repository. With this version
>> it
>> >> seems to work.
>> >> You have two machines because you tried netloc on these two?
>> >>
>> >>
>> >> Cyril.
>> >>
>> >>> Le 15/02/2017 à 22:44, miharulidze a écrit :
>> >>> Hi!
>> >>>
>> >>> I'm trying to use NetLoc tool for detecting my cluster topology.
>> >>>
>> >>> I have 2 node cluster with AMD Processors, connected by InfiniBand.
>> Also
>> >>> I installed latest versions of hwloc and netloc tools.
>> >>>
>> >>> I followed the instruction of netloc and when I tried to use
>> >>> netloc_ib_gather_raw as root, i recieved this message
>> >>> root:$ netloc_ib_gather_raw
>> >>> --out-dir=/home/halilov/mycluster-data/result/
>> >>> --hwloc-dir=/home/halilov/mycluster-data/hwloc/ --sudo
>> >>>
>> >>> Found 0 subnets in hwloc directory:
>> >>>
>> >>>
>> >>> There are two files in /home/halilov/mycluster-data/hwloc/ generated
>> by
>> >>> hwloc: head.xml and node01.xml
>> >>>
>> >>> P.S. in attach archieve with .xml files
>> >>>
>> >>>
>> >>> Best regards,
>> >>> Mikhail Khalilov
>> >>>
>> >>>
>> >>>
>> >>> ___
>> >>> hwloc-users mailing list
>> >>> hwloc-users@lists.open-mpi.org
>> >>> https://rfd.newmexicoconsortium.org/mailman/listinfo/hwloc-users
>> >> ___
>> >> hwloc-users mailing list
>> >> hwloc-users@lists.open-mpi.org
>> >> https://rfd.newmexicoconsortium.org/mailman/listinfo/hwloc-users
>> > ___
>> > hwloc-users mailing list
>> > hwloc-users@lists.open-mpi.org
>> > https://rfd.newmexicoconsortium.org/mailman/listinfo/hwloc-users
>>
>> ___
>> hwloc-users mailing list
>> hwloc-users@lists.open-mpi.org
>> https://rfd.newmexicoconsortium.org/mailman/listinfo/hwloc-users
>>
>
>
>
> ___
> hwloc-users mailing 
> listhwloc-us...@lists.open-mpi.orghttps://rfd.newmexicoconsortium.org/mailman/listinfo/hwloc-users
>
>
>
> ___
> hwloc-users mailing list
> hwloc-users@lists.open-mpi.org
> https://rfd.newmexicoconsortium.org/mailman/listinfo/hwloc-users
>



  











  
  
  
  
  
  



  
  


  
  


  

  

  
  


  
  


  

  

  
  


  
  


  

  

  
  


  
  


  

  

  
  

Re: [hwloc-users] NetLoc subnets Problem

[Brice Goglin]

Please copy-paste the exact command line of your "netloc_ib_gather_raw"
and all the messages it printed. And also send the output of the hwloc
directory it created (it will contain the lstopo XML output of the node
where you ran the command).

Brice


Le 17/02/2017 09:51, Михаил Халилов a écrit :
> I installed nightly tarball, but it still isn't working. In attach
> info of ibnetdiscover and ibroute. May be it wlii help...
> What could be the problem?
>
> Best regards,
> Mikhail Khalilov
>
> 2017-02-17 9:53 GMT+03:00 Brice Goglin  >:
>
> Hello
>
> As identicated on the netloc webpages, the netloc development now
> occurs
> inside the hwloc git tree. netloc v0.5 is obsolete even if hwloc 2.0
> isn't released yet.
>
> If you want to use a development snapshot, take hwloc nightly tarballs
> from https://ci.inria.fr/hwloc/job/master-0-tarball/
>  or
> https://www.open-mpi.org/software/hwloc/nightly/master/
> 
>
> Regards
> Brice
>
>
>
>
>
> Le 16/02/2017 19:15, miharuli...@gmail.com
>  a écrit :
> > I downloaded gunzip from openmpi site here:
> https://www.open-mpi.org/software/netloc/v0.5/
> 
> >
> > There are three identical machines in my cluster, but now third
> node is broken, and i tried on two machines. They all connected by
> InfiniBand switch, and when I try to use ibnetdiscovery or
> ibroute, it works perfectly...
> >
> >
> >
> > Отправлено с iPad
> >> 16 февр. 2017 г., в 18:40, Cyril Bordage
> > написал(а):
> >>
> >> Hi,
> >>
> >> What version did you use?
> >>
> >> I pushed some commits on master on ompi repository. With this
> version it
> >> seems to work.
> >> You have two machines because you tried netloc on these two?
> >>
> >>
> >> Cyril.
> >>
> >>> Le 15/02/2017 à 22:44, miharulidze a écrit :
> >>> Hi!
> >>>
> >>> I'm trying to use NetLoc tool for detecting my cluster topology.
> >>>
> >>> I have 2 node cluster with AMD Processors, connected by
> InfiniBand. Also
> >>> I installed latest versions of hwloc and netloc tools.
> >>>
> >>> I followed the instruction of netloc and when I tried to use
> >>> netloc_ib_gather_raw as root, i recieved this message
> >>> root:$ netloc_ib_gather_raw
> >>> --out-dir=/home/halilov/mycluster-data/result/
> >>> --hwloc-dir=/home/halilov/mycluster-data/hwloc/ --sudo
> >>>
> >>> Found 0 subnets in hwloc directory:
> >>>
> >>>
> >>> There are two files in /home/halilov/mycluster-data/hwloc/
> generated by
> >>> hwloc: head.xml and node01.xml
> >>>
> >>> P.S. in attach archieve with .xml files
> >>>
> >>>
> >>> Best regards,
> >>> Mikhail Khalilov
> >>>
> >>>
> >>>
> >>> ___
> >>> hwloc-users mailing list
> >>> hwloc-users@lists.open-mpi.org
> 
> >>>
> https://rfd.newmexicoconsortium.org/mailman/listinfo/hwloc-users
> 
> >> ___
> >> hwloc-users mailing list
> >> hwloc-users@lists.open-mpi.org
> 
> >>
> https://rfd.newmexicoconsortium.org/mailman/listinfo/hwloc-users
> 
> > ___
> > hwloc-users mailing list
> > hwloc-users@lists.open-mpi.org
> 
> > https://rfd.newmexicoconsortium.org/mailman/listinfo/hwloc-users
> 
>
> ___
> hwloc-users mailing list
> hwloc-users@lists.open-mpi.org 
> https://rfd.newmexicoconsortium.org/mailman/listinfo/hwloc-users
> 
>
>
>
>
> ___
> hwloc-users mailing list
> hwloc-users@lists.open-mpi.org
> https://rfd.newmexicoconsortium.org/mailman/listinfo/hwloc-users

___
hwloc-users mailing list
hwloc-users@lists.open-mpi.org
https://rfd.newmexicoconsortium.org/mailman/listinfo/hwloc-users

Re: [hwloc-users] NetLoc subnets Problem

[Михаил Халилов]

I installed nightly tarball, but it still isn't working. In attach info of
ibnetdiscover and ibroute. May be it wlii help...
What could be the problem?

Best regards,
Mikhail Khalilov

2017-02-17 9:53 GMT+03:00 Brice Goglin :

> Hello
>
> As identicated on the netloc webpages, the netloc development now occurs
> inside the hwloc git tree. netloc v0.5 is obsolete even if hwloc 2.0
> isn't released yet.
>
> If you want to use a development snapshot, take hwloc nightly tarballs
> from https://ci.inria.fr/hwloc/job/master-0-tarball/ or
> https://www.open-mpi.org/software/hwloc/nightly/master/
>
> Regards
> Brice
>
>
>
>
>
> Le 16/02/2017 19:15, miharuli...@gmail.com a écrit :
> > I downloaded gunzip from openmpi site here: https://www.open-mpi.org/
> software/netloc/v0.5/
> >
> > There are three identical machines in my cluster, but now third node is
> broken, and i tried on two machines. They all connected by InfiniBand
> switch, and when I try to use ibnetdiscovery or ibroute, it works
> perfectly...
> >
> >
> >
> > Отправлено с iPad
> >> 16 февр. 2017 г., в 18:40, Cyril Bordage 
> написал(а):
> >>
> >> Hi,
> >>
> >> What version did you use?
> >>
> >> I pushed some commits on master on ompi repository. With this version it
> >> seems to work.
> >> You have two machines because you tried netloc on these two?
> >>
> >>
> >> Cyril.
> >>
> >>> Le 15/02/2017 à 22:44, miharulidze a écrit :
> >>> Hi!
> >>>
> >>> I'm trying to use NetLoc tool for detecting my cluster topology.
> >>>
> >>> I have 2 node cluster with AMD Processors, connected by InfiniBand.
> Also
> >>> I installed latest versions of hwloc and netloc tools.
> >>>
> >>> I followed the instruction of netloc and when I tried to use
> >>> netloc_ib_gather_raw as root, i recieved this message
> >>> root:$ netloc_ib_gather_raw
> >>> --out-dir=/home/halilov/mycluster-data/result/
> >>> --hwloc-dir=/home/halilov/mycluster-data/hwloc/ --sudo
> >>>
> >>> Found 0 subnets in hwloc directory:
> >>>
> >>>
> >>> There are two files in /home/halilov/mycluster-data/hwloc/ generated
> by
> >>> hwloc: head.xml and node01.xml
> >>>
> >>> P.S. in attach archieve with .xml files
> >>>
> >>>
> >>> Best regards,
> >>> Mikhail Khalilov
> >>>
> >>>
> >>>
> >>> ___
> >>> hwloc-users mailing list
> >>> hwloc-users@lists.open-mpi.org
> >>> https://rfd.newmexicoconsortium.org/mailman/listinfo/hwloc-users
> >> ___
> >> hwloc-users mailing list
> >> hwloc-users@lists.open-mpi.org
> >> https://rfd.newmexicoconsortium.org/mailman/listinfo/hwloc-users
> > ___
> > hwloc-users mailing list
> > hwloc-users@lists.open-mpi.org
> > https://rfd.newmexicoconsortium.org/mailman/listinfo/hwloc-users
>
> ___
> hwloc-users mailing list
> hwloc-users@lists.open-mpi.org
> https://rfd.newmexicoconsortium.org/mailman/listinfo/hwloc-users
>
#
# Topology file: generated on Fri Feb 17 12:47:35 2017
#
# Initiated from node 0002c903004a9972 port 0002c903004a9973

vendid=0x2c9
devid=0xbd36
sysimgguid=0x2c90200442bd3
switchguid=0x2c90200442bd0(2c90200442bd0)
Switch  8 "S-0002c90200442bd0"  # "Infiniscale-IV Mellanox 
Technologies" base port 0 lid 4 lmc 0
[3] "H-0002c903004a996e"[1](2c903004a996f)  # "node01 HCA-1" lid 2 
4xQDR
[7] "H-0002c903004a9972"[1](2c903004a9973)  # "head HCA-1" lid 1 
4xQDR

vendid=0x2c9
devid=0x673c
sysimgguid=0x2c903004a9971
caguid=0x2c903004a996e
Ca  1 "H-0002c903004a996e"  # "node01 HCA-1"
[1](2c903004a996f)  "S-0002c90200442bd0"[3] # lid 2 lmc 0 
"Infiniscale-IV Mellanox Technologies" lid 4 4xQDR

vendid=0x2c9
devid=0x673c
sysimgguid=0x2c903004a9975
caguid=0x2c903004a9972
Ca  1 "H-0002c903004a9972"  # "head HCA-1"
[1](2c903004a9973)  "S-0002c90200442bd0"[7] # lid 1 lmc 0 
"Infiniscale-IV Mellanox Technologies" lid 4 4xQDR
Unicast lids [0x0-0x4] of switch Lid 4 guid 0x0002c90200442bd0 (Infiniscale-IV 
Mellanox Technologies):
  Lid  Out   Destination
   Port Info 
0x0001 007 : (Channel Adapter portguid 0x0002c903004a9973: 'head HCA-1')
0x0002 003 : (Channel Adapter portguid 0x0002c903004a996f: 'node01 HCA-1')
0x0004 000 : (Switch portguid 0x0002c90200442bd0: 'Infiniscale-IV Mellanox 
Technologies')
3 valid lids dumped 
___
hwloc-users mailing list
hwloc-users@lists.open-mpi.org
https://rfd.newmexicoconsortium.org/mailman/listinfo/hwloc-users

Re: [hwloc-users] NetLoc subnets Problem

[Brice Goglin]

Hello

As identicated on the netloc webpages, the netloc development now occurs
inside the hwloc git tree. netloc v0.5 is obsolete even if hwloc 2.0
isn't released yet.

If you want to use a development snapshot, take hwloc nightly tarballs
from https://ci.inria.fr/hwloc/job/master-0-tarball/ or
https://www.open-mpi.org/software/hwloc/nightly/master/

Regards
Brice





Le 16/02/2017 19:15, miharuli...@gmail.com a écrit :
> I downloaded gunzip from openmpi site here: 
> https://www.open-mpi.org/software/netloc/v0.5/
>
> There are three identical machines in my cluster, but now third node is 
> broken, and i tried on two machines. They all connected by InfiniBand switch, 
> and when I try to use ibnetdiscovery or ibroute, it works perfectly...
>
>
>
> Отправлено с iPad
>> 16 февр. 2017 г., в 18:40, Cyril Bordage  написал(а):
>>
>> Hi,
>>
>> What version did you use?
>>
>> I pushed some commits on master on ompi repository. With this version it
>> seems to work.
>> You have two machines because you tried netloc on these two?
>>
>>
>> Cyril.
>>
>>> Le 15/02/2017 à 22:44, miharulidze a écrit :
>>> Hi!
>>>
>>> I'm trying to use NetLoc tool for detecting my cluster topology.
>>>
>>> I have 2 node cluster with AMD Processors, connected by InfiniBand. Also
>>> I installed latest versions of hwloc and netloc tools.
>>>
>>> I followed the instruction of netloc and when I tried to use
>>> netloc_ib_gather_raw as root, i recieved this message
>>> root:$ netloc_ib_gather_raw
>>> --out-dir=/home/halilov/mycluster-data/result/
>>> --hwloc-dir=/home/halilov/mycluster-data/hwloc/ --sudo
>>>
>>> Found 0 subnets in hwloc directory:
>>>
>>>
>>> There are two files in /home/halilov/mycluster-data/hwloc/ generated by
>>> hwloc: head.xml and node01.xml
>>>
>>> P.S. in attach archieve with .xml files
>>>
>>>
>>> Best regards,
>>> Mikhail Khalilov
>>>
>>>
>>>
>>> ___
>>> hwloc-users mailing list
>>> hwloc-users@lists.open-mpi.org
>>> https://rfd.newmexicoconsortium.org/mailman/listinfo/hwloc-users
>> ___
>> hwloc-users mailing list
>> hwloc-users@lists.open-mpi.org
>> https://rfd.newmexicoconsortium.org/mailman/listinfo/hwloc-users
> ___
> hwloc-users mailing list
> hwloc-users@lists.open-mpi.org
> https://rfd.newmexicoconsortium.org/mailman/listinfo/hwloc-users

___
hwloc-users mailing list
hwloc-users@lists.open-mpi.org
https://rfd.newmexicoconsortium.org/mailman/listinfo/hwloc-users

Re: [hwloc-users] NetLoc subnets Problem

[miharulidze]

I downloaded gunzip from openmpi site here: 
https://www.open-mpi.org/software/netloc/v0.5/

There are three identical machines in my cluster, but now third node is broken, 
and i tried on two machines. They all connected by InfiniBand switch, and when 
I try to use ibnetdiscovery or ibroute, it works perfectly...



Отправлено с iPad
> 16 февр. 2017 г., в 18:40, Cyril Bordage  написал(а):
> 
> Hi,
> 
> What version did you use?
> 
> I pushed some commits on master on ompi repository. With this version it
> seems to work.
> You have two machines because you tried netloc on these two?
> 
> 
> Cyril.
> 
>> Le 15/02/2017 à 22:44, miharulidze a écrit :
>> Hi!
>> 
>> I'm trying to use NetLoc tool for detecting my cluster topology.
>> 
>> I have 2 node cluster with AMD Processors, connected by InfiniBand. Also
>> I installed latest versions of hwloc and netloc tools.
>> 
>> I followed the instruction of netloc and when I tried to use
>> netloc_ib_gather_raw as root, i recieved this message
>> root:$ netloc_ib_gather_raw
>> --out-dir=/home/halilov/mycluster-data/result/
>> --hwloc-dir=/home/halilov/mycluster-data/hwloc/ --sudo
>> 
>> Found 0 subnets in hwloc directory:
>> 
>> 
>> There are two files in /home/halilov/mycluster-data/hwloc/ generated by
>> hwloc: head.xml and node01.xml
>> 
>> P.S. in attach archieve with .xml files
>> 
>> 
>> Best regards,
>> Mikhail Khalilov
>> 
>> 
>> 
>> ___
>> hwloc-users mailing list
>> hwloc-users@lists.open-mpi.org
>> https://rfd.newmexicoconsortium.org/mailman/listinfo/hwloc-users
> ___
> hwloc-users mailing list
> hwloc-users@lists.open-mpi.org
> https://rfd.newmexicoconsortium.org/mailman/listinfo/hwloc-users
___
hwloc-users mailing list
hwloc-users@lists.open-mpi.org
https://rfd.newmexicoconsortium.org/mailman/listinfo/hwloc-users

Re: [hwloc-users] NetLoc subnets Problem

[Cyril Bordage]

Hi,

What version did you use?

I pushed some commits on master on ompi repository. With this version it
seems to work.
You have two machines because you tried netloc on these two?


Cyril.

Le 15/02/2017 à 22:44, miharulidze a écrit :
> Hi!
> 
> I'm trying to use NetLoc tool for detecting my cluster topology.
> 
> I have 2 node cluster with AMD Processors, connected by InfiniBand. Also
> I installed latest versions of hwloc and netloc tools.
> 
> I followed the instruction of netloc and when I tried to use
> netloc_ib_gather_raw as root, i recieved this message
> root:$ netloc_ib_gather_raw
> --out-dir=/home/halilov/mycluster-data/result/
> --hwloc-dir=/home/halilov/mycluster-data/hwloc/ --sudo
> 
> Found 0 subnets in hwloc directory:
> 
> 
> There are two files in /home/halilov/mycluster-data/hwloc/ generated by
> hwloc: head.xml and node01.xml
> 
> P.S. in attach archieve with .xml files
> 
> 
> Best regards,
> Mikhail Khalilov
> 
> 
> 
> ___
> hwloc-users mailing list
> hwloc-users@lists.open-mpi.org
> https://rfd.newmexicoconsortium.org/mailman/listinfo/hwloc-users
> 
___
hwloc-users mailing list
hwloc-users@lists.open-mpi.org
https://rfd.newmexicoconsortium.org/mailman/listinfo/hwloc-users