InfiniBand

1. openfabrics.org Primary source for OFED. Somewhat convoluted, thus a bit of work to get them compiled on your platform. Compilation for MPI, MVAPICH, etc would be needed for HPC env. They include ULP (upper layer protocol) such as IPoIB.
   
2. RHEL ofed - RHEL backport fixes, and keep a more usable OFED stack than the community version. use this before using community version to avoid headache (eg luster compatibility).
   
3. mellanox.com - compile binaries for major distro, used to be about 3 months lag behind source, but probably best route for supported cards.
   If updateing OS, need to get new version of driver, and recompile for the OS. may take some 15 min to compile and install. May also disable opensmd systemd , just need to re-enable it
4. It seems that the current trend is to use the ofed drivers. some compilation needed. Compilation for MPI, MVAPICH, etc would be needed for HPC env. They include ULP (upper layer protocol) such as IPoIB.
   

InfiniBand 101

InfiniBand Generations

SDR - Single Data Rate              4x2.5 = 10 Gbps  (8 Gbps accounting for bit encoding overhead)
DDR - Double Data Rate              4x 5  = 20 Gbps
QDR - Quad Data Rate.       QSFP,   4x10 =  40 Gbps.  CX-2 ??
FDR - Fith? Data Rate.      QSFP,   4x14 =  56 Gbps.  CX-3
EDR - Enhanced Data Rate.   QSFP28, 4x25 = 100 Gbps.  CX-4 ?  CX-5?  NRZ modulation.    Backward compatible with FDR.
HDR - High?/Hyper? DR.      QSFP56, 4x50 = 200 Gbps.  CX-6 can do HDR200 or Y-split 2x HDR100.  PAM4 modulation.  2018.  Backward compatible at 10/25/40/50/100/200G, but matching cables would be needed.
NDR -  QSFP-DD?  8x50?
XDR -  OSFP ?    8x50? 

modulation (think encoding):
NRZ:   Non-Return-to-Zero: traditional binary, on/off 1 bit per cycle.  Is this same as PAM-2? 
PAM-4: Pulse Amplitude Modulation-4: 2 bits per PAM4 symbol, encoded via 4 mean voltages: VA, VB,VC, VD.

Arista 7280R2 - 25 Gbps SERDES?  QSFP28 ports, Max out at 100Gbps?  only do NRZ?
Arista 7280R3 - 50 Gbps SERDES?  QSFP56 ports, Capable of 200Gbps?  Capable of PAM4 ?

ConnectX-6 can connect to slower equipment, but note that HDR cable will negotiage PAM4.  EDR cable would be needed to have it negotiate to NRZ.
In practice, used ConnectX-6 HBA, HDR100 cable, connected to EDR switch, and it worked.  ibstat reports Rate: 100, so presumably did negotiate NRZ cuz of the EDR switch and use 4x25.
Still don't understand why HDR cables have IB version vs Ethernet version (they are not interchangeable).

1. wikipedia EDR vs HDR table
   
2. Review of ConnectX-5 by STH
   
3. Mellanox CX-6 spec sheet

MCX556A-EDAT -- decyphering this model number:
First 5 = CX-5
The 6   = dual port
The D   = PCIe 4.0

Note: Model with Lx designation run at lower speed than the max allowed by the protocol.  eg 50 Gbps instead of 100. (cuz not using all the lanes?)

Generation cross compatibility

InfiniBand 103

opensm 	# open fabric subnet manager.  run on one of the host.  
	# additional standy manager on other hosts can be setup.  
	# the daemon process will only bind to ONE IB port.  
	# HA fabrics should be joined into a single one.
  	# Fix opensm.conf to bind to desired guid (port).   
	# not specifiying one will get prompt to choose if multiple port is avail.
	# If have two-port card and need TWO subnet manager, then need to 
	# update /etc/init.d/opensm and start TWO instance of opensmd
	# use -g GUID to specify which port each instance use.  this would overwrite config in /etc/opensm/opensm.conf. eg 
 	# /usr/sbin/opensm --daemon 
 	# /usr/sbin/opensm --daemon -g 0x506b4b03007c00c2  --logfile /var/log/opensm_c2.log

OFED has ib driver and opensm
Mellanox compiled their own version and provide support for it, MLNX_OFED, aka MOFED.  has driver and opensm.  No more support for ConnectX-2 cards in 2018.
SL7 version of OFED is said to have less trouble with LUSTER than general public OFED.


opensm --create-config /tmp/opensm.conf 
cp -i /tmp/opensm.conf /etc/opensm/
vi /etc/opensm/opensm.conf # set GUID for the port where openSM should bind to

Troubleshooting commands

ibstat
  - all ports should be active.  It would be active if there is running fabric manager to register them when the port comes alive.
  - If port is connected, but no running Subnet Manager, port status would be Initiating
  - If port status is Down, then there is problem with physical link (even when no SM is running)  Reseating after SM is running may help.
  - node guid
  - port guid

ibv_devinfo - similar info to ibstat

rds-ping
rds-info

ompi_info --all
ompi-top
ompi-ps



iblinkinfo 
iblinkinfo --node-name-map fabrics.node-name-map
	# iblinkinfo provides link info for whole fabric, very quick 
	# node-name-map converts GUID to human name
	# 0xe41d2d03004f1b40 "ib000.cf0 (Mellanox SX6025, S54-21)"

ibnetdiscover 
ibnetdiscover --node-name-map fabrics.node-name-map > ibnet.topo
	# generate a topology file 
	# switch LID would be listed above all the port, see comment field 


cat fabrics.conf | grep -v ^$ | tr '[]' '#' | awk -F\= '{print $1 " \"" $2 "\""}' > fabrics.node-name-map

MOFED (MLNX_OFED) for RH7
/etc/init.d/openibd status
/etc/init.d/opensmd status


ibcheck

ibacm...
        # ib adaptive control manager
        # not really needed by OFED, but included by yong


tool to update IB (HCA?) firmware:
flint ?
mfp ?

# select output of ibnetdiscover

switchguid=0x2c90200410d02(2c90200410d02)
Switch  24 "S-0002c90200410d02"   # "MT47396 Infiniscale-III Mellanox Technologies" base port 0 lid 2 lmc 0
#                                                                               ib switch LID      ^^^

[16]    "H-00188b9097fe8e41"[1](188b9097fe8e42)     # "rac1001 HCA-1" lid 4 4xDDR
[15]    "H-00188b9097fe906d"[1](188b9097fe906e)     # "rac1002 HCA-1" lid 9 4xDDR
^^^^ ib port where host is connected            comment area and host's LID ^^^

ibclearerrors		# reboot will induce errors, this is normal.
ibclearcounters		


hca_self_test.ofed  	# sanity check. ofed-scripts rpm.  /usr/bin
pdsh -g etna0 TERM=vt52 /usr/bin/hca_self_test.ofed | grep -i fail
# the script use tput and eed to force TERM when used with pdsh


perfquery		# simple to read ib counters.  pay special attention to:
			# SymbolErrorCounter
			# LinkErrorRecover
			# infiniband-diags by Mellanox, install to /sbin

ibqueryerrors -c -s XmWait


ibcheckerrors -b	# very noisy, from infiniband-diags rpm by Mellanox, /sbin



ibdiagnet		# scan fabric, takes a long time

wwibcheck by Yong Qin:
ibcheck -f fabrics.conf -C etna -E  -dd -a -b -O 20 > now.out
ibcheck -f fabrics.conf -C etna -E  -dd -a -b -O 20 > 2hrsLater.out
but can't simple diff to see errors, need to awk out the RcvPkgs columns...

per-node errors listed in bottom section.
This is the header of avail counters:

   NodeDesc Port               Guid ExcBufOverrunErrors LinkDowned LinkIntegrityErrors LinkRecovers RcvConstraintErrors    RcvData RcvErrors    RcvPkts RcvRemotePhysErrors RcvSwRelayErrors SymbolErrors VL15Dropped XmtConstraintErrors    XmtData XmtDiscards    XmtPkts

IB port on switch side

ibportstate - manage specific port on the IB switch, eg turn it on/off, etc.

ibportstate [swlid] [swPortNum] [command]	
ibportstate  2        11         disable	# turn off IB switch port a specific host is connected to
ibportstate  6        11         disable
     # can also use enable, status.  Other form allow for changing speed, etc.

ibcheck

ibcheck needs root to run many cmd (ibcheck use fabrics.conf for GUID to hostname resolution, etna below is name of a cluster)
ibcheck -f ../fabrics.conf -C etna -E  -dd -a -b -O 20 > ibcheck-E-b-dd-a.etna.out.1226

ibcheck -h		# help (also pydoc ./ibcheck)

ibcheck -E      # errors and perf counter (top-like UI, h for help)

ibcheck -d -f fabrics.conf -C lrc-fabric
	-f      # specify a fabric definition file, so name displayed is looked up from fabrics.conf
	-d      # detail level 1, just a list of devices
	-dd     # detail level 2, switch and what is connected to each ports [most adequate output]
		# -dd   scan all ports with a link attached to it, ie list switch and what is connected to it.
	-ddd    # detail level 3, add vendor id, device id, guid etc to each switch or Ca.  empty ports explicity listed as such
		# -ddd  scan all ports - empty port will be explicity listed as such
	-C      # filter by Cluster - ie display only for entries matching specified cluster
	-i      # change scan interval, in seconds.
	-b      # batch mode, avoid the top-like interface for -E
	-O      # timeout, default to 3 sec, may need it to be longer

ibcheck -f fabrics.conf -dd -C cf0 -E -i3
		# monitor Error counter for specific cluster
		# top-like interface.  use space to toggle between error-only list vs all-node list
		# -f fabrics.conf is file maintained manually.  Use Node GUID : name key-value pair
		# note that nodes are defined in their cluster (eg [lr4]) and super cluster (ie [lrc-fabric])

ibcheck -l 32,33 -ddd 					# -l filter for specific LID (comma list)
ibcheck -f fabrics.conf -g 0x7cfe900300b515e0 -ddd 	# -g GUID, can be comma list
ibcheck -f fabrics.conf -t ibnetdiscover.out	 	# -t scan for changes against a saved ibnetdiscovery file
ibcheck -f fabrics.conf -dd -M 				# -M also display how many subnet manager(s) are running and where.

ibcheck -E -b         -O 10 -ddd

Ref

1. Yong Qin's IB Diag & Troubleshooting
2. ulowa basic ib troubleshooting
3. http://tinyurl.com/infiniband - my cache of assorted IB docs.

Y-split cable switch port config

# using switch lid worked, but need subnet manager active to assign lid-id to the switch

ibswitches 	# scan fabric for switches and display their lid
         	# lid 0 seems like default unconfigured lid that doesn't work

# q = query current setting
mlxconfig -d lid-912 q
mlxconfig -d lid-912 q SPLIT_MODE
mlxconfig -d lid-$ID q SPLIT_PORT
mlxconfig -d lid-999 q SPLIT_PORT[1..12]	# see config for ports 1 to 12, inclusive

# enable split mode, all ports will be split
# split port will go from x/y to have subports x/y/1 and x/y/2
mlxconfig -d lid-912 set SPLIT_MODE=1     # lid-912 as example , mlxconfig will prompt for confirmation
    flint -d lid-$ID swreset              # reboot switch, no password required! but only root can open dev path

# set specific port as non split
mlxconfig -d lid-912 set SPLIT_PORT[40]=0    # port 40 as non-split

# set specific port as split (they get the extra x/y/z treatment)
mlxconfig -d lid-912 set SPLIT_PORT[1..32]=1 # port 1 to 32 as split mode

other things to try:

sudo mst start
# which create /dev/mst device path

# set (switch?) to suport SPLIT_MODE:
# but my dev is for the local CX6 .... this didn't work:

# query for current settings:
sudo mlxconfig -d /dev/mst/mt4123_pciconf0 q | grep -i SPLIT       # q=query, but split_mode is on the swithc, not HBA
sudo mlxconfig -d /dev/mst/SW_MT54000_EVB-SX-1_L00_lid-0x0001 q    # doc says switch device path could exist under /dev/mst (not for me)

mlxconfig -f ./tlv_file.conf -d /dev/mst/mt4115_pciconf1 set_raw

Split port mode for Y-cable support in managed switch

serial into switch
admin/admin is def
	will prompt for pass change, as well as set pass for "monitor"
enable
config term
system profile ib split-ready
(reboot)

interface ib 1/1-1/36
shutdown
module-type qsfp-split-2
(ports 1 to 36 will be in split mode and ready for use, existing 2nd port of Y-split would be online right away

check status:

show interface ib status


to set back to non split mode
module-type qsfp

save config
? copy running-config startup-config 
? write mem

How to split or unsplit ports on a QM8700 using MLNX-OS CLI - 
This article introduces the procedure used to split the ports in two on a QM8700 switch. QM Series switches support split mode, which allows for the 200 Gb/s quad-lane ports to be divided into two 100 Gb/s dual-lane ports.


need the mallanox cable for serial connectivity.

IBoIP

# bring up interface after driver install, w/o reboot 
rmmod    ib_ipoib
modprobe ib_ipoib

/etc/sysconfig/network-scripts/ifcfg-ib0 ::
ConnectedMode
no  = use datagram (default?)
yes = like tcp?

docker login abcd 1234

config
interface ib 1/4
shutdown   ! shutdown the specific switch port

module-type qsfp-split-2   ! set to split mode for use with Y-split cable
module-type qsfp           ! revert, back to normal 1:1 cable


show interfaces ib status

# Note, QM8700 HDR switch has 40 ports.  

$ cat /etc/sysconfig/network-scripts/ifcfg-bond0
DEVICE=bond0
BOOTPROTO=none
IPADDR=10.2.2.98
NETMASK=255.255.255.0
NETWORK=10.2.2.0
BROADCAST=10.2.2.255
ONBOOT=yes
USERCTL=no
TYPE=Bonding

$ cat /etc/sysconfig/network-scripts/ifcfg-ib0  
DEVICE=ib0
BOOTPROTO=none
ONBOOT=yes
USERCTL=no
MASTER=bond0
SLAVE=yes
TYPE=InfiniBand

$ cat /etc/sysconfig/network-scripts/ifcfg-ib1
DEVICE=ib1
BOOTPROTO=none
ONBOOT=yes
USERCTL=no
MASTER=bond0
SLAVE=yes
TYPE=InfiniBand

# relevant entries in  /etc/modprobe.conf ::
## added for Oracle 10/11 (per Cambridge WI)
## make sure that the hangcheck-timer kernel module is set to load when the system boots
options hangcheck-timer hangcheck_tick=1 hangcheck_margin=10 hangcheck_reboot=1
alias ib0 ib_ipoib
alias ib1 ib_ipoib
alias net-pf-27 ib_sdp
##  For IB bonding
alias bond0 bonding
options bond0 miimon=100 mode=1 max_bonds=1

$ cat /etc/sysconfig/network-scripts/ifcfg-ib0
DEVICE=ib0
BOOTPROTO=none
STARTMODE=onboot
ONBOOT=yes
USERCTL=no
BROADCAST=192.168.2.255
NETMASK=255.255.255.0
#
IPADDR=0.0.0.0
#SLAVE=yes
#MASTER=bond1
TYPE=InfiniBand


$ cat /etc/sysconfig/network-scripts/ifcfg-ib1
DEVICE=ib1
BOOTPROTO=none
STARTMODE=onboot
ONBOOT=yes
USERCTL=no
BROADCAST=192.168.2.255
NETMASK=255.255.255.0
#
IPADDR=0.0.0.0
#SLAVE=yes
#MASTER=bond1
TYPE=InfiniBand


# relevant entries in  /etc/modprobe.conf ::
options bond0 mode=balance-rr miimon=100    max_bonds=2

# ethernet bonds are configured as in stock RHEL 5.3 config.

#!/bin/sh

# init script to start ib-bond at boot time:
#
# run "manual" ib-bond config
# could not do this in /etc/sysconfig/network-scripts as bond1, ib0, ib1 scripts
# as somehow presence of eth bond would make ib-bond fail over not to work.
# maybe a bug in how the network-scripts are parsed...
#
##  nn = startLevel  Sxx Kxx
##  eg start at rc3 and rc5,   start as S56, kill is omitted so no Kxx script
##  maybe as S28 if need to be before oracleasm
##
# chkconfig: 35 28 -
# description: ib-bond config
#

# source function library
. /etc/rc.d/init.d/functions


RETVAL=0
prog="ib-bond-config"

start() {
        ifconfig ib0 up 0.0.0.0/24
        ifconfig ib1 up 0.0.0.0/24
        ib-bond --bond-name bond1 --bond-ip 192.168.2.101/24 --slaves ib0,ib1 --miimon 100
}

stop() {
        ib-bond --bond-name bond1 --stop
}

status() {
        ib-bond --status-all
}

case "$1" in
        start)
                echo -n $"Starting $prog: "
                start
                RETVAL=$?
                [ "$RETVAL" = 0 ] && logger local7.info "ib-bond-config start ok" || logger local7.err "ib-bond-config start failed"
                echo
                ;;
        stop)
                echo -n $"Stopping $prog: "
                #echo 'not implemented yet'
                stop
                [ "$RETVAL" = 0 ] && logger local7.info "ib-bond-config stop ok" || logger local7.err "ib-bond-config stop failed"
                echo
                ;;
        status)
                status
                echo
                ;;
        *)
                echo $"Usage: $0 {start|stop|status}"
                RETVAL=1
esac



exit $RETVAL

## setup aid:
##  sudo cp -p /nfshome/sa/config_backup/lx/conf.test-4000/ib-bond-config /etc/init.d/
##  sudo ln -s /etc/init.d/ib-bond-config /etc/rc.d/rc3.d/S56ib-bond-config
##  sudo ln -s /etc/init.d/ib-bond-config /etc/rc.d/rc5.d/S56ib-bond-config

/usr/bin/mst start
mst status

mlxvpd    -d /dev/mst/mt4119_pciconf0 	# show nic model part number, serial
mlxuptime -d /dev/mst/mt4119_pciconf0	# show dev cpu freq and uptime



ibv_devinfo	# find firmware version

/sbin/lspci -d 15b3:		# list only mellanox devices, get device id for use with mstflint

mstflint -hh # detail help on burning firmware on mellanox device

sudo /usr/bin/mstflint --device af:00.0 query 
sudo /usr/bin/mstflint -d       af:00.0 q     full  # full query, but still no OPN info


## burn .bin firmware (not .zip).  -ir burn does some image reactivation prior burning, which was what the example shows.
sudo /usr/bin/mstflint --device af:00.0 --image fw-ConnectX5-rel-16_27_2008-MCX555A-ECA_Ax-UEFI-14.20.22-FlexBoot-3.5.901.bin -ir burn 
## but that --ir thing wasn't accepted, so skiped, and end up doing:
sudo /usr/bin/mstflint --device af:00.0 --image fw-ConnectX5-rel-16_28_2006-MCX555A-ECA_Ax_Bx-UEFI-14.21.17-FlexBoot-3.6.102.bin  burn 


# can't verify again after firmware upgrade, unless do a reload of the module first.  easier to just reboot.
sudo /usr/bin/mstflint --device af:00.0 verify


# to check for cable problem, connectivity with switch:
lspci | grep Mellano # to get pci device number
mlxlink -d af:00.0 -m -e -c --show_fec
mlxlink -d af:00.0 --speeds 10G,25G,50G,100G

mlxlink -d af:00.0 -a TG # Toggle link up and down (to reset speed)

# find cable type via Part Number:
mlxlink -d af:00.0 -m  | grep Part
	Vendor Part Number  : MCP1600-C002	# EDR QSFP28 DAC ## n0002sav3 no FDR, only SDR (CX-5 HBA, FDR switch)
	Vendor Part Number  : MCP1600-E002E30	# EDR QSFP+  DAC ## n0004sav3 works as FDR (CX5 HBA, FDR switch)
	Vendor Part Number  : MC2207130-002	# FDR QSFP+  DAC VPI 56 Gbps IB/40 Gbps Eth ## n0150sav3 

the C0002 version is an ethernet cable, in IB it will negotiate to 10Gb in FDR config, and not work at all for EDR config.  
the E002E30 version is IB and can negotiate down to 56Gb and run at FDR rate!
part matter!!

ref: 
https://www.mellanox.com/related-docs/prod_cables/PB_MCP1600-Cxxx_100GbE_QSFP28_DAC.pdf     - Cxxx = Ethernet
https://www.mellanox.com/related-docs/prod_cables/PB_MCP1600-E0xxEyy_100Gbps_QSFP28_DAC.pdf - E0xxEyy = IB
https://store.mellanox.com/products/nvidia-mc2207130-002-passive-copper-cable-vpi-up-to-56gb-s-qsfp-2m.html - FDR VPI

HDR100 vs 100GbE cables (Y-split version):
both cables have the same Model number:  MCP7H50
but a single character differ in the part number:
MCP7H50-V002R26 - 100GbE
MCP7H50-H002R26 - HDR100  (presumably works in IB mode "only")
(The 100GbE marking was on the cable connector, use of cable negotiated to 5 Gbps when used as IB.  ibnetdiscover reported it as SDR).

mstflint doc from mellanox.

IB to Ethenert

mlxconfigfrom mellanox doc.

Check if Mellanox ConnectX-n card is in IB or Eth/VPI mode

mst start
mlxconfig query | grep LINK_TY

# 1 = IB
# 2 = Eth/VPI

change IB card to Eth mode

mlxconfig --dev /dev/mst/mt4119_pciconf0 set LINK_TYPE_P1=2                  # single port NIC
mlxconfig --dev /dev/mst/mt4119_pciconf0 set LINK_TYPE_P1=2 LINK_TYPE_P2=2   # dual   port NIC


for Mellanox ConnectX-4 

may need to run 
mst start
mst status
to create /dev/mst device 
allegedly once in eth mode, no need for any of these things.
mst in turn req OFED driver installed.
(have not found way to run these out of docker container, even when Mellanox seems to have something for this, maybe for some product they have which i have not found doc about yet).

also need a Mellanox MAM1Q00A-QSA for each port you want to be on ethernet as well as SR optics.  



https://community.mellanox.com/s/article/getting-started-with-connectx-5-100gb-s-adapters-for-linux



Note that the LINK_TYPE_P1 and LINK_TYPE_P2 equal 1 (InfiniBand) by default.

 mlxconfig -d /dev/mst/mt4119_pciconf0 query
 (or simply mlxdonfig query)
 --dev name depends on what "mst status" reports

eg2 (connectX-3 MT27500):
mlxconfig --dev /dev/mst/mt4099_pciconf0 query | grep LINK_TY
         LINK_TYPE_P1                        VPI(3)          
		 LINK_TYPE_P2                        VPI(3)          

eg3 (ConnectX-5 MT27800):
         LINK_TYPE_P1                        IB(1)           


before eth mode on CX-5 card, ifconfig reports:
	2: enp1s0f0:   [NO-CARRIER,BROADCAST,MULTICAST,UP] mtu 1500 qdisc mq state DOWN group default qlen 1000
	3: enp1s0f1:   [NO-CARRIER,BROADCAST,MULTICAST,UP] mtu 1500 qdisc mq state DOWN group default qlen 1000
	4: enp134s0f0: [BROADCAST,MULTICAST,UP,LOWER_UP] mtu 1500 qdisc mq state UP group default qlen 1000
	5: enp134s0f1: [NO-CARRIER,BROADCAST,MULTICAST,UP] mtu 1500 qdisc mq state DOWN group default qlen 1000
	6: docker0:    [NO-CARRIER,BROADCAST,MULTICAST,UP] mtu 1500 qdisc noqueue state DOWN group default 
	25: ib0:       [NO-CARRIER,BROADCAST,MULTICAST,UP] mtu 4092 qdisc mq state DOWN group default qlen 256

ibdev2netdev  # from mlnx-ofa_kernel-4.7-OFED.4.7.1.0.0.1.g1c4bf42.rhel7u7.x86_64.rpm
	mlx5_0 port 1 ... ib0 (Down)



Change to Ethernet mode for dual port card:

eg# mlxconfig -d /dev/mst/mt4119_pciconf0 set LINK_TYPE_P1=2 LINK_TYPE_P2=2

mlxconfig --dev /dev/mst/mt4119_pciconf0 set LINK_TYPE_P1=2

Device #1:
----------

Device type:    ConnectX5       
Name:           N/A             
Description:    N/A             
Device:         /dev/mst/mt4119_pciconf0

Configurations:                              Next Boot       New
         LINK_TYPE_P1                        IB(1)           ETH(2)          


reboot needed for change to be effected.


~~~~
https://www.servethehome.com/changing-mellanox-connectx-vpi-ports-ethernet-infiniband-linux/

VPI mode automagically do Eth and IB, no need to change its mode?
Some card can only do IB or only Eth, can't be changed?

**no, likely need to still flip to eth so that os assign a ifconfig nic name to it. **


# ip a | egrep UP\|inet
2: enp1s0f0: [BROADCAST,MULTICAST,UP,LOWER_UP] mtu 1500 qdisc mq state UP group default qlen 1000
    inet 128.3.7.58/24 brd 128.3.7.255 scope global noprefixroute enp1s0f0
3: enp1s0f1: [NO-CARRIER,BROADCAST,MULTICAST,UP] mtu 1500 qdisc mq state DOWN group default qlen 1000
4: ib0: [BROADCAST,MULTICAST,UP,LOWER_UP] mtu 2044 qdisc mq state UP group default qlen 256
    inet 192.168.207.58/24 brd 192.168.207.255 scope global noprefixroute ib0
    inet6 fe80::e61a:4fc3:cba3:72c9/64 scope link noprefixroute 

enp1s0f? are likely the onboard RJ45 ports.

MLNX-OS

MLNX-OS CLI change password https://docs.nvidia.com/networking/display/MLNXOSv381206/User+Management+and+Security+Commands

show whoami

show usernames		# show configured user.  default is admin monitor xmladmin xmluser  but most disabled, need to set password to enable
			# Acc Status will indicate if password is set

enable
config term

username bofh  capability admin # create user , not usable till set a password
username bofh  password 0       # 0 = use clear text password, will prompt for it
# or use
username bofh  password 7     	# 7 = encrypted text (which alg?)

username admin   password 0     # change password for user admin
username monitor password 0     # def pw is monitor for this acc

username admin nopassword   # erase password?

saving config
see: https://docs.nvidia.com/networking/display/MLNXOSv381206/Configuration+Management

enable
config term

show configuration files	# default one is called "initial"

configuration write    		# save to default? current? config file.  disable zero touch config ZTC thing on first write.  Think Cisco copy run start

configuration write to NewCfFilename		# create/save to a new config file and switch to it as active config file (used at boot?)
configuration write to NewCfFilename no-switch	# save to a new config file , but don't make it the active config (ie good for creating backup)

configuration switch-to NewCfFilename		# load a new config file

show running


reset factory keep-basic			# restore to factory default

Set management interface IP

enable
config terminal

show interfaces mgmt0

# default is dhcp, manually disable it 
no interface mgmt0 dhcp

# setting static ip would disable dhcp 
interface    mgmt0 ip address 10.6.1.10 255.255.0.0

---

[Doc URL: http://tin6150.github.io/psg/]
[Doc URL: http://tin6150.gitlab.io/psg/]
Last Updated: 2020-10-27
(cc) Tin Ho. See main page for copyright info.

---

nSarCoV2
hoti1
sn5050
psg101 sn50 tin6150