Networking

Unix and Linux network configuration. Multiple network interfaces. Bridged NICs. High-availability network configurations.

Applications

Reviews of latest Unix and Linux software. Helpful tips for application support admins. Automating application support.

Data

Disk partitioning, filesystems, directories, and files. Volume management, logical volumes, HA filesystems. Backups and disaster recovery.

Monitoring

Distributed server monitoring. Server performance and capacity planning. Monitoring applications, network status and user activity.

Commands & Shells

Cool Unix shell commands and options. Command-line tools and application. Things every Unix sysadmin needs to know.

Home » Networking

Troubleshooting InfiniBand Connections

Submitted by on May 4, 2011 – 10:36 am 6 Comments

The following is a brief troubleshooting guide for an InfiniBand network found in common HPC Linux clusters. Running these commands requires OFED 1.5.2 or later package installed on your systems. Additionally, the “pdsh” (parallel shell) command is part of the HP CMU cluster management software (version 4.2.1 used in our example) installed on the head node. If you don’t have CMU installed, below you will find a simple scripted alternative for running commands on multiple cluster nodes. In our cluster compute nodes are named “node1” through “node32”.

Identify hardware module used for the IB interface:

ls /sys/class/infiniband

Sample output:

mlx4_0

Check the state of the IB port:

cat /sys/class/infiniband/mlx4_0/ports/1/state

Sample output:

4: ACTIVE

Check the status of the subnet manager and start it if necessary:

if [ `/etc/init.d/opensmd status | grep -c "not running"` -gt 0 ]
then
     /etc/init.d/opensmd start
fi

Sample output:

Starting opensm:              done

Check the state of the IB port on the compute nodes:

pdsh –w node[1-32] cat /sys/class/infiniband/mlx4_0/ports/1/state

Sample output:

node1: 4: ACTIVE
node5: 4: ACTIVE
node2: 4: ACTIVE
node6: 4: ACTIVE
node7: 4: ACTIVE
...

Here’s a way of running the command above without pdsh:

i=1
while [ $i -le 32 ]
do
     echo -n "node$i: "
     ssh node$i "cat /sys/class/infiniband/mlx4_0/ports/1/state"
     (( i = i + 1 ))
done

The next step is to check the speed of IB ports on the head node:

cat /sys/class/infiniband/mlx4_0/ports/1/rate

Sample output:

40 Gb/sec (4X QDR)

… and on the compute nodes:

pdsh –w node[1-32] cat /sys/class/infiniband/mlx4_0/ports/1/rate

And here’s how you do this without pdsh:

i=1
while [ $i -le 32 ]
do
     echo -n "node$i: "
     ssh node$i "cat /sys/class/infiniband/mlx4_0/ports/1/rate"
     (( i = i + 1 ))
done

More detailed analysis of the IB connection can be performed with the ibdiagnet command:

ibdiagnet -pc -c 1000

Sample output:

Loading IBDIAGNET from: /usr/lib64/ibdiagnet1.5.4
-W- Topology file is not specified.
    Reports regarding cluster links will use direct routes.
Loading IBDM from: /usr/lib64/ibdm1.5.4
-I- Using port 1 as the local port.
-I- Discovering ... 39 nodes (3 Switches & 36 CA-s) discovered.


-I---------------------------------------------------
-I- Bad Guids/LIDs Info
-I---------------------------------------------------
-I- No bad Guids were found

-I---------------------------------------------------
-I- Links With Logical State = INIT
-I---------------------------------------------------
-I- No bad Links (with logical state = INIT) were found

-I---------------------------------------------------
-I- General Device Info
-I---------------------------------------------------

-I---------------------------------------------------
-I- PM Counters Info
-I---------------------------------------------------
-W- lid=0x0001 guid=0x0008f10500200898 dev=23130 Port=4
      Performance Monitor counter     : Value
      symbol_error_counter            : 0x4 (Increase by 4 during ibdiagnet
      scan.)

-I---------------------------------------------------
-I- Fabric Partitions Report (see ibdiagnet.pkey for a full hosts list)
-I---------------------------------------------------
-I-    PKey:0x7fff Hosts:36 full:36 limited:0

-I---------------------------------------------------
-I- IPoIB Subnets Check
-I---------------------------------------------------
-I- Subnet: IPv4 PKey:0x7fff QKey:0x00000b1b MTU:2048Byte rate:10Gbps SL:0x00
-W- Suboptimal rate for group. Lowest member rate:20Gbps > group-rate:10Gbps

-I---------------------------------------------------
-I- Bad Links Info
-I- No bad link were found
-I---------------------------------------------------
----------------------------------------------------------------
-I- Stages Status Report:
    STAGE                                    Errors Warnings
    Bad GUIDs/LIDs Check                     0      0
    Link State Active Check                  0      0
    General Devices Info Report              0      0
    Performance Counters Report              0      1
    Partitions Check                         0      0
    IPoIB Subnets Check                      0      1

Please see /tmp/ibdiagnet.log for complete log
----------------------------------------------------------------

-I- Done. Run time was 11 seconds.

cat /sys/class/infiniband/mlx4_0/ports/1/lid
pdsh –w node[1-32] /sys/class/infiniband/mlx4_0/ports/1/lid
ibdiagnet –pc –c 1000

The final step is to check the error state of each port:

ibcheckerrors

Sample output:

#warn: counter SymbolErrors = 35        (threshold 10) lid 1 port 255
#warn: counter RcvSwRelayErrors = 512   (threshold 100) lid 1 port 255
Error check on lid 1 (Voltaire 4036 - 36 QDR ports switch) port all:  FAILED
#warn: counter SymbolErrors = 35        (threshold 10) lid 1 port 4
Error check on lid 1 (Voltaire 4036 - 36 QDR ports switch) port 4:  FAILED

## Summary: 39 nodes checked, 0 bad nodes found
##          134 ports checked, 1 ports have errors beyond threshold

In our case, the “ibcheckerrors” command revealed a problem with the IB switch. This turned out to be a hardware problem and the switch needed to be replaced.

Print Friendly, PDF & Email

6 Comments »

  • kass9191 says:

    I’m taking a couple Pre-IB (International Baccalaureate) classes right now, and I’d like to move down from those classes next year, which will then become IB, so that I can have time to actually practice music at home. For a Music Major my primary focus should be my knowledge of music theory and the quality of my performance, but I’m afraid that moving down from two pre-IB classes one level might hurt my chances.

  • Dr Hank says:

    i’m trying out for 2 ib schools in my area
    i’m really nervous because in 1 of the schools.. it states that only 100 students are accepted
    how should i prepare myself?
    how difficult are the test?
    thanks
    I’m trying out for 2 ib highschools in my area…
    its stated that in 1 of the schools only 100 are accepted
    how difficult are the test?
    how should I prepare for it?
    thanks!

  • Kevin says:

    I want to go to Cambridge in England when I am older.
    Would IB classes be a good thing to take?

    I’m in middle school right now going on to High School soon.
    By the way I’m just finishing 7th grade right now and I just finished Algebra 1 Honors and going onto Geometry 1 Honors in 8th grade.

  • cardskid22 says:

    The IB program is the international baccalaureate program and running start is taking classes at the community college nearby. Which would a college like yale prefer on an application??

    Pleease help!! Thanks!!!

  • Smashing Pumpkins says:

    I am planning on going to an american university when i finish the IB programme.
    Now… wtf is the difference between graduate and undergraduate degrees and which one will i have to apply to?

    Also do IB students get any extra credit or financial aid in american universities? I am also an international student (i study in Malaysia)

    please enlighten me i am very confused :(
    thank you guys :)
    please answer the question.

  • zigg3ns says:

    I just got a notice from the school saying that i need to pay for my IB Exams.

    I tought only those who are getting the IB Diploma need to take the exams or is it for both?
    Please answer.

1 Pingbacks »

Leave a Reply

%d bloggers like this: