Hi,
What could be the reason that open(/dev/infiniband/umad0, O_RDWR|O_NONBLOCK)
blocks and does not return.
I did not find any errors in dmesg.
Thanks,
Eli
___
general mailing list
general@lists.openfabrics.org
Eli Dorfman wrote:
What could be the reason that open(/dev/infiniband/umad0,
O_RDWR|O_NONBLOCK)
blocks and does not return. I did not find any errors in dmesg.
Eli,
You can examine the kernel stack of all processes, including yours... using
sysrq ($ echo 1 /proc/sysrq-trigger and then $
Hi,
OFED-1.4.1-rc5 was removed from OFA downloads.
OFED-1.4.1-rc6 will be released as soon as dependence issue between nfs and
ib_core will be resolved and tested.
Regards,
Vladimir
___
general mailing list
general@lists.openfabrics.org
Dotan Barak wrote:
You can't find such samples in the verbs library; It can be found in
the rdma cma library, you should search for rping or ucmatose.
Dotan
2009/5/12 zhouyong...@ict.ac.cn:
Hi all,
I'm using libibverbs to build a cluster memory pool, and using TCP/IP
handshake to
This email was generated automatically, please do not reply
git_url: git://git.openfabrics.org/ofed_1_4/linux-2.6.git
git_branch: ofed_kernel
Common build parameters:
Passed:
Passed on i686 with linux-2.6.16
Passed on i686 with linux-2.6.19
Passed on i686 with linux-2.6.17
Passed on i686 with
Subject: [PATCH] perftest: send_lat/bw: Attach to multicast group when QP is in
INIT
If multicast is enabled, the current code of send_lat/bw attaches the QP
to a multicast group while it's still in RESET state.
Since the IB spec does not strictly specify the QP state for this operation
and
Brian M. Rzycki wrote:
Greetings,
I have the following SLES 11 SP0 machine:
It looks like the OFED installer isn't building ib_iser.ko even when I
choose 2,3.
This is the same bug reported on rc5. We removed rc5 and will publish
RC6 soon
Tziporet
Steve Wise wrote:
Signed-off-by: Steve Wise sw...@opengridcomputing.com
---
Applied
Tziporet
___
general mailing list
general@lists.openfabrics.org
http://lists.openfabrics.org/cgi-bin/mailman/listinfo/general
To unsubscribe, please visit
Hi,
Thanks for this patch. But I've to NACK because
1) Greg KH has already done a similar patch in his tree.
See http://lists.openfabrics.org/pipermail/general/2009-May/059442.html
2) Your patch is incomplete
Regards
Nam
Roel Kluin roel.kl...@gmail.com wrote on 11.05.2009 22:25:07:
From:
Steve Wise wrote:
Trond Myklebust wrote (earlier in this thread):
All I should need to know is that I can advertise either dma handles or
kernel VAs, and know that I can choose between two functions, say,
ib_send_wr_fastreg_dma_init() and ib_send_wr_fastreg_kva_init() to
initialise the
On 09:51 Wed 06 May , Ira Weiny wrote:
2 changes I noted in the output from ibqueryerrors.
Link Info: was not being printed when -r was used.
The header: Errors for 0xguid node name
Should only be printed when errors are found.
The following patch cleans those up.
Ira
On 09:53 Wed 06 May , Ira Weiny wrote:
Another corner case: If there are no down links on a switch and -d is
selected then the header for that switch should not be printed.
Ira
From: Ira Weiny wei...@llnl.gov
Date: Thu, 30 Apr 2009 13:41:38 -0700
Subject: [PATCH] Clean up
On 06:46 Sun 10 May , Hal Rosenstock wrote:
Sasha has been adamant that any device supplied data errors use
something other than ERROR log level.
But I think that VERBOSE is more appropriate than for such cases than
just DEBUG. Another way is to add another level for subnet warnings.
On 10:33 Thu 07 May , Hal Rosenstock wrote:
Signed-off-by: Hal Rosenstock hal.rosenst...@gmail.com
Applied. Thanks.
Sasha
___
general mailing list
general@lists.openfabrics.org
http://lists.openfabrics.org/cgi-bin/mailman/listinfo/general
To
On 14:30 Sun 10 May , Doron Shoham wrote:
set SAQUERY_CMD_CLASS_PORT_INFO instead of CLASS_PORT_INFO
Signed-off-by: Doron Shoham dor...@voltaire.com
Applied. Thanks.
Sasha
___
general mailing list
general@lists.openfabrics.org
On Tue, May 12, 2009 at 1:55 PM, Sasha Khapyorsky sas...@voltaire.com wrote:
On 06:46 Sun 10 May , Hal Rosenstock wrote:
Sasha has been adamant that any device supplied data errors use
something other than ERROR log level.
But I think that VERBOSE is more appropriate than for such cases
On 14:29 Sat 09 May , Eli Dorfman (Voltaire) wrote:
lid persistent range wrong check
used lids were not properly chekced which
caused duplicate lid assignment in some cases.
Signed-off-by: Eli Dorfman e...@voltaire.com
Applied. Thanks.
Sasha
On 07:21 Tue 12 May , Hal Rosenstock wrote:
to what's needed (based on NodeDescription length)
Signed-off-by: Hal Rosenstock hal.rosenst...@gmail.com
Applied. Thanks.
Sasha
___
general mailing list
general@lists.openfabrics.org
On 14:06 Tue 12 May , Hal Rosenstock wrote:
Yes, VERBOSE level is more consistent than DEBUG level with what is
done elsewhere in OpenSM.
Ok, I'm changing to VERBOSE.
Sasha
___
general mailing list
general@lists.openfabrics.org
Signed-off-by: Hal Rosenstock hal.rosenst...@gmail.com
---
diff --git a/opensm/opensm/osm_port.c b/opensm/opensm/osm_port.c
index 17bac73..cb8b153 100644
--- a/opensm/opensm/osm_port.c
+++ b/opensm/opensm/osm_port.c
@@ -381,7 +381,7 @@ uint8_t osm_physp_calc_link_op_vls(IN osm_log_t * p_log,
On 14:32 Tue 12 May , Hal Rosenstock wrote:
Signed-off-by: Hal Rosenstock hal.rosenst...@gmail.com
Oops, I committed this already in the local branch :) Will use your
version instead.
Applied. Thanks.
Sasha
___
general mailing list
On 18:55 Thu 07 May , Doron Shoham wrote:
do not enforce PortInfo update if max_op_vls = 0
Signed-off-by: Doron Shoham dor...@voltaire.com
---
opensm/opensm/osm_port.c |2 +-
opensm/opensm/osm_subnet.c |8
2 files changed, 9 insertions(+), 1 deletions(-)
diff
On 09:49 Sun 10 May , Eli Dorfman (Voltaire) wrote:
Doron Shoham wrote:
check if op_vls = 0 before max_op_vls comparison
Signed-off-by: Doron Shoham dor...@voltaire.com
---
opensm/opensm/osm_port.c |9 +
1 files changed, 5 insertions(+), 4 deletions(-)
diff
On 11:17 Sun 10 May , Doron Shoham wrote:
check if op_vls = 0 before max_op_vls comparison
Signed-off-by: Doron Shoham dor...@voltaire.com
Applied. Thanks. See comments below.
---
opensm/opensm/osm_port.c | 11 ++-
1 files changed, 6 insertions(+), 5 deletions(-)
diff
On 18:01 Wed 06 May , Ira Weiny wrote:
The number reported as max hops from ibnetdiscover can change depending on
the algorithm used to discover the fabric. As Hal says in the message below
using this number is therefore dangerous.
If no one is currently using this number I propose the
Hi all,
I'm using libibverbs to build a cluster memory pool, and
using TCP/IP
handshake to exchange memory information and establish the connection
before the IB communication. While I found this process costed a lot
of time, 100ms in 1GEth LAN, so I want to use the rdma_cm or ib_ucm to
Davis, Arlin R arlin.r.da...@intel.com wrote:
For a connection (socket connect, exchanging QP info, private data, qp modify)
using uDAPL socket cm versus rdma_cm I get:
socket_cm on 1Ge == ~900us
socket_cm on IPoIB (mlx4 ddr) == ~400us
rdma_cm on IB (mlx4 ddr) == ~2200us
As you can see, the
Just to make sure we're on the same page: both IPoIB and the RDMA-CM
use SA path queries
But ipoib caches its path records...
- Sean
___
general mailing list
general@lists.openfabrics.org
http://lists.openfabrics.org/cgi-bin/mailman/listinfo/general
Davis, Arlin R arlin.r.da...@intel.com wrote:
For a connection (socket connect, exchanging QP info,
private data, qp modify)
using uDAPL socket cm versus rdma_cm I get:
socket_cm on 1Ge == ~900us
socket_cm on IPoIB (mlx4 ddr) == ~400us
rdma_cm on IB (mlx4 ddr) == ~2200us
As you can see,
Just to make sure we're on the same page: both IPoIB and the RDMA-CM
use SA path queries
But ipoib caches its path records...
Yes, of-course. But, to start with, lets analyze the case of each node
running --one-- rank and then take it from there to the case where
each node runs C ranks.
Or.
Davis, Arlin R arlin.r.da...@intel.com wrote:
Just to make sure we're on the same page: both IPoIB and the RDMA-CM
use SA path queries (ipoib for the unicast arp reply, and rdma-cm for
rdma_resolve_route), going into details, things look like:
I am running IPoIB connected so I assume there is no
Yes, of-course. But, to start with, lets analyze the case of each node
running --one-- rank and then take it from there to the case where
each node runs C ranks.
The caching is independent of running MPI though. To get a fair comparison,
you'd probably have to reboot the entire cluster before
The QP contains references to the protection domain (PD), memory
regions (MR), address handles, completion queues (CQ), address
handles (AH), etc.
The QP should be destroyed before any other objects are destroyed
so that the referenced object is not busy.
Signed-off-by: Ralph Campbell
33 matches
Mail list logo
