-
From: Brice Goglin
Sent: 29 January 2019 15:39
To: Biddiscombe, John A. ; Hardware locality user list
Subject: Re: [hwloc-users] unusual memory binding results
Only the one in brackets is set, others are unset alternatives.
If you write "madvise" in that file, it'll become &qu
On the 8 numa node machine
$cat /sys/kernel/mm/transparent_hugepage/enabled
[always] madvise never
is set already, so I'm not really sure what should go in there to disable it.
JB
-Original Message-
From: Brice Goglin
Sent: 29 January 2019 15:29
To: Biddiscombe, John A. ; Hardware
Message-
From: Brice Goglin
Sent: 29 January 2019 10:35
To: Biddiscombe, John A. ; Hardware locality user list
Subject: Re: [hwloc-users] unusual memory binding results
Crazy idea: 512 pages could be replaced with a single 2MB huge page.
You're not requesting huge pages in your allocation
2019 10:35
To: Biddiscombe, John A. ; Hardware locality user list
Subject: Re: [hwloc-users] unusual memory binding results
Crazy idea: 512 pages could be replaced with a single 2MB huge page.
You're not requesting huge pages in your allocation but some systems have
transparent huge pages enabled
0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0
On the 8 numa node machine it sometimes gives the right answer even with 512
pages.
Still baffled
JB
-Original Message-
From: hwloc-users On Behalf Of
Biddiscombe, John A.
Sent: 28 January 2019 16:14
To: Brice
Brice
>Can you print the pattern before and after thread 1 touched its pages, or even
>in the middle ?
>It looks like somebody is touching too many pages here.
Experimenting with different threads touching one or more pages, I get
unpredicatable results
here on the 8 numa node device, the
either the syscall is wrong, or the kernel is doing something else.
I welcome any suggestions on what might be wrong.
Thanks for trying to help.
JB
-Original Message-
From: Brice Goglin
Sent: 26 January 2019 10:19
To: Biddiscombe, John A.
Cc: Hardware locality user list
Subject: Re: [hwloc-u
Dear List/Brice
I experimented with disabling the memory touch on threads except for N=1,2,3,4
etc and found a problem in hwloc, which is that the function
hwloc_get_area_memlocation was returning '0' when the status of the memory null
move operation was -14 (#define EFAULT 14 /* Bad address
Brice
Apologies, I didn't explain it very well, I do make sure that if the tile size
256*8 < 4096 (pagesize), then I double the number of tiles per page, I just
wanted to keep the explanation simple.
here are some code snippets to give you the flavour of it
initializing the helper sruct
ll
things magically work if I upgrade to hwloc 2.0 etc etc
Thanks
JB
-Original Message-
From: hwloc-users [mailto:hwloc-users-boun...@lists.open-mpi.org] On Behalf Of
Biddiscombe, John A.
Sent: 13 November 2017 15:37
To: Hardware locality user list <hwloc-users@lists.open-mpi.org
the
original alloc call?
thanks
JB
From: hwloc-users [hwloc-users-boun...@lists.open-mpi.org] on behalf of
Biddiscombe, John A. [biddi...@cscs.ch]
Sent: 13 November 2017 14:59
To: Hardware locality user list
Subject: Re: [hwloc-u
locality user list
Subject: Re: [hwloc-users] question about hwloc_set_area_membind_nodeset
Use get_area_memlocation()
membind() returns where the pages are *allowed* to go (anywhere)
memlocation() returns where the pages are actually allocated.
Brice
Le 13/11/2017 06:52, Biddiscombe, John
Thank you to you both.
I modified the allocator to allocate one large block using hwloc_alloc and then
use one thread per numa domain to touch each page according to the tiling
pattern - unfortunately, I hadn't appreciated that now
hwloc_get_area_membind_nodeset
always returns the full machine
, Biddiscombe, John A. a écrit :
> Dear list
>
> According to the release notes Add HWLOC_MEMBIND_BYNODESET flag was
> added in 1.11.3 - if I protect some code with
>
> #if HWLOC_API_VERSION >= 0x00010b00
> then versions 1.11.0, 1.11.1, 1.11.2 still cause build failures.
>
&
Dear list
According to the release notes Add HWLOC_MEMBIND_BYNODESET flag was added in
1.11.3 - if I protect some code with
#if HWLOC_API_VERSION >= 0x00010b00
then versions 1.11.0, 1.11.1, 1.11.2 still cause build failures.
is there some VERSION flag that distinguishes between the patch
> -Original Message-
> From: hwloc-users [mailto:hwloc-users-boun...@open-mpi.org] On Behalf
> Of Chris Samuel
> Sent: 26 March 2014 13:42
> To: Hardware locality user list
> Subject: Re: [hwloc-users] BGQ question.
>
> On Wed, 26 Mar 2014 11:56:08 AM Biddiscombe, John A. wr
Chris
> Out of interest, why on an I/O node?
I'm targeting the BGQ BGAS nodes with flash cards installed. We've done tests
with GPFS mounted on the flash and are trying to get comparable results with an
in-house driver.
JB
Goglin [mailto:brice.gog...@inria.fr]
Sent: 25 March 2014 09:28
To: Biddiscombe, John A.; Hardware locality user list
Subject: Re: [hwloc-users] BGQ question.
Can you run hwloc-gather-topology foo and send the resulting foo.tar.bz2 ?
If the tarball is too bug, feel free to send it to me in a private
for some special projects where we are trying to
customise the IO.
JB
From: Brice Goglin [mailto:brice.gog...@inria.fr]
Sent: 25 March 2014 08:43
To: Hardware locality user list; Biddiscombe, John A.
Subject: Re: [hwloc-users] BGQ question.
Wait, I missed the "io node" part of your first mai
with something set?
Thanks
JB
From: hwloc-users [mailto:hwloc-users-boun...@open-mpi.org] On Behalf Of Brice
Goglin
Sent: 25 March 2014 08:04
To: Hardware locality user list
Subject: Re: [hwloc-users] BGQ question.
Le 25/03/2014 07:51, Biddiscombe, John A. a écrit :
I'm compiling hwloc using
I'm compiling hwloc using clang (bgclang++11 from ANL) to run on IO nodes af a
BGQ. It seems to have compiled ok, and when I run lstopo, I get an output like
this (below), which looks reasonable, but there are 15 sockets instead of 16.
I'm a little worried because the first time I compiled, I
21 matches
Mail list logo