Linux Networking : What is of bond interface & when it will be useful

Thanks for ur reply  ...

Am DBA, we got the new server in our dept.
when I see.. I can see whole list of interfaces .. no sure why there are so many.. I am confused ..

please explain

==
 ifcfg-bond0
 ifcfg-bond
 ifcfg-bond:
 ifcfg-bond2
 ifcfg-bond3
 ifcfg-bond4
 ifcfg-eth286_2.4
 ifcfg-eth646_.4
 ifcfg-ib0
 ifcfg-ib0.8004
 
 ifcfg-ib
 ifcfg-ib.8004

ifcfg-lo
===

look at ifconfig -a
this will dispaly the active ones.
lspci to see your hardware
do you have FC HBA, 10G Network HBAs?
the bond interfaces might represent different IPs on different segments.

Manuel, addressed the utility of the bond interface.
from device failure handling, to expanding the available bandwidth/load balancing traffic.....

If you need further help, please let us know.  ..MFlores..

@mac_g, the list you show is  the list of config files which may or may not be active.
VLANs seem to be involved as well so please provide the output of the following commands:

ifconfig -a
-- This should produce the current active config

cat /proc/net/bonding/*
-- This should produce info on bonding used.

cat /proc/net/vlan/config
-- This should produce the VLAN configuration

@noci, for your info .. please refer below

---------
bonding: bond1: Adding slave eth1286_2.41.
bonding: bond1: Warning: enslaved VLAN challenged slave eth1286_2.41. Adding VLANs will be blocked as long as eth1286_2.41 is part of bond bond1
[mlx4_vnic] eth1286_2.41 is opened
bonding: bond1: enslaving eth1286_2.41 as a backup interface with an up link.
bonding: bond1: Adding slave eth646_1.41.
bonding: bond1: Warning: enslaved VLAN challenged slave eth646_1.41. Adding VLANs will be blocked as long as eth646_1.41 is part of bond bond1
[mlx4_vnic] eth646_1.41 is opened
bonding: bond1: enslaving eth646_1.41 as a backup interface with an up link.
bonding: bond1: link status definitely up for interface eth1286_2.41, 10000 Mbps full duplex.
bonding: bond1: making interface eth1286_2.41 the new active one.
bonding: bond1: link status definitely up for interface eth646_1.41, 10000 Mbps full duplex.
[root@RUHMPV-ADF01 ~]# dmesg | grep ib
Calibrating delay loop (skipped), value calculated using timer frequency.. 5786.21 BogoMIPS (lpj=2893108)
vgaarb: bridge control possible 0000:00:02.0
libata version 3.00 loaded.
Refined TSC clocksource calibration: 2893.029 MHz.
mlx4_ib: Mellanox ConnectX InfiniBand driver v1.0-ofed1.5.5 (April 4, 2008)
mlx4_core 0000:00:04.0: mlx4_ib: multi-function enabled
mlx4_core 0000:00:04.0: mlx4_ib: using node_guid:0x0021280001fc66e6
mlx4_core 0000:00:04.0: mlx4_ib: operating in qp1 tunnel mode
bonding: bond2: Adding slave ib0.
bonding: bond2: Warning: enslaved VLAN challenged slave ib0. Adding VLANs will be blocked as long as ib0 is part of bond bond2
bonding: bond2: enslaving ib0 as a backup interface with a down link.
ib0: enabling connected mode will cause multicast packet drops
ib0: mtu > 2044 will cause multicast packet drops.
ib0: mtu > 2044 will cause multicast packet drops.
bonding: bond4: Adding slave ib0.8004.
bonding: bond4: Warning: enslaved VLAN challenged slave ib0.8004. Adding VLANs will be blocked as long as ib0.8004 is part of bond bond4
bonding: bond4: enslaving ib0.8004 as a backup interface with a down link.
ib0.8004: enabling connected mode will cause multicast packet drops
ib0.8004: mtu > 2044 will cause multicast packet drops.
ib0.8004: mtu > 2044 will cause multicast packet drops.
bonding: bond3: Adding slave ib0.8006.
bonding: bond3: Warning: enslaved VLAN challenged slave ib0.8006. Adding VLANs will be blocked as long as ib0.8006 is part of bond bond3
bonding: bond3: enslaving ib0.8006 as a backup interface wi

you keep providing the information you included in your original question. These are log events, please provide the information asked if you would like assistance, answers and resolution of your issue.

The output of the next 3 commands wil be more informative than what has been provided until now:
(though they might invoke requests for more information)

ifconfig -a
cat /proc/net/bonding/*
cat /proc/net/vlan/config

And yes you may obfusciate addresses be make it consistent....
Next addresses need no obfusciation they are private.:
192.168.... , 172.16-31..., 10....   and used in multiple organisations throughout the world in duplicates.

if you change the prefix to xxx.yyy.zzz for all addresses then it will produce a meaningfull listing for f.e.  
a host that has aan address ending .1  ( 1.1.1.1 will change tp xxx.yyy.zzz.1)  that is, until duplicates appear.  (1.1.2.1 should be changed to xxx.yyy.aaa.1 to show this).

@noci
all the output is attached for your reference. please refer it
1.txt

Right....

Bond0-Bond4 are useful all configured as Active-Backup (meaning only one is active and if the link fails it will switch over the the other link).

bond5 upto bond10 are unused. (not usefull)

Bond0 is based on infiniband adapter0/channel 8008 (ib0.8008) and inifband adapter1/channel 8008 (ib1.8008)    (hardly used)

Bond1 = ethernet adapters with eth1286_2.41 as active.  The 41 might be vlan41 but that is unclear as vlan info in not included in the 1.txt,   (10 Gbps adapters)

bond2 = ib0 & ib1 base adapters (channel 0)     (light use)

Bond3 = ib0.8006 & ib1.8006  (most used)

bond4 = ib0.8004 & ib1.8004 (light use)

In All bonds the first adapter is leading by default, where the other is backup (ie. only one active).
If it fails for 5 seconds it will switch over the the backup link. the link state is checked 4 times per second.

You may get better performance by switching a few of the infiniband bonds so ib1 is active all but the .8006 channel one.

To actualy answer your question: if in a bonded set of interfaces one fails it will be replaced by the surviving ones. (active-backup).
You will need more than just bonded interfaces here, you also need double switches so if one switch fails  (its link will fail) then other switch can take it's place. So the system can still communicate.

Bond may also be configured in other styles active/active: both interfaces are used and adds to the capacity of the total bandwidth. Failure means loosing part of the bandwidth.
A bond may contain more than 2 interfaces btw.

With Active/Active there are more options on how to select which packet leaves over with interface.
For that please check this page for current available modes: http://www.cloudibee.com/network-bonding-modes/

@noci

1) why some interfaces are configured with IP, some without IP ?

2) Can we put the interface name as we like ?   Or the naming should be based on its use ?

1) why has IF no IP:
Slaves have no IP, they are bound (enslaved) to the bond interface.
bond5..bond10... are unused, no slaves, no config. and can be deleted.

2) Interface naming as you wish
Yes and no....
Bond interface naming is free... but scripting might expect a name like bondX.
Ethernet Interfaces can be renamed, but systemd uses that to create this type of names
Vlan selection is reflected in the interface name. (.xx)

IMHO it is unwise to arbitrarily rename if names. OTOH is can be done, but be aware it might cause issues down the line with scripting for interface management.  (Ethernet interfaces are expected to start with eth...

@noci,

Thanks so much for the patience.
I am getting more clarity with your answers.


1> when one needs to configure Masters interface & Slave interface ?  
2> Can you give a simple example for such requirements to understand better ?

1) A bond by definition consists of one Master interface (usually bondX X=0,1,2...)
and the slave interfaces that are glued together. ("Physical" interfaces).

The Physical interfaces do the heavylifting of the transportation of packets. The Bond interface is used to manage those interfaces and dispatch packets according to the "Mode" of operation. (See previous provided link).

2) Bonds are mostly used for redundancy so there are no Single points of Failure (SPOF's) w.r.t. network access. It is about High Availability environments resillience against failures.
It involves more than just bonds, it also requires multiple powersupplies, multiple switches, multiple routers, multiple SAN's (with mirrored storage) possibly at separated locations (with 10, 20.. 500Km distance between sites, with 500Km distance one typicaly goes for Disaster Tollerance [ think earth quake, flooding, Hurricanes ] ).

Thanks for your reply.

I have attached one image where our server env is going to be built on this architecture.
In such case, for the servers from image : ERPDB1 & ERPDB2,

1) what would be best practice to configure the interface, for each server(master & slave)

The above 2-servers connected to 1) Public NW  && 2) connected privately to have communication between them.

Can you please advice
EE-01.png

thanks @noic ...