NIC's naming standard - Suggestions

[joe_91]

Looking for standards in naming NIC's

for a host with hostname myhost the DNS entry would typically be

myhost
myhost-mgt
myhost-nas
myhost-ebr

myhost: the primary access to the server, it can be a bonded interfaces with redundant connections or a single interface.

myhost-mgt: the remote management console: DRAC, - always a single 100/Full connection

myhost-nas: the newly required, bonded with primary/redundant 1000/full connection for NAS.

myhost-ebr: the generally single 1000/full connection for backup traffic, may require routing to master/media servers unless on same subnet.

For each of these groups, if there are multiple non-floating IPs add an instance number added to the end

ie if there is a need for multiple EBR connections:
1st: myhost-ebr <-primary ebr connection and what all server with start out with
2nd: myhost-ebr1
3rd: myhost-ebr2

Thanks

Joe.

[rick jones]

When I see "naming NICs" I usually think of deciding which NIC should be eth0, which eth1 etc. :)

As for naming conventions, for names to be mapped to IP addresses associated with services, I'm not familiar with any particular standard. It is really up to you to pick a convention and stick with it.

I should also point-out that by default, linux takes a very strong interpretation of following the "weak end system model" and considers IP addresses a property of the system rather more than of any one NIC. As such, any interface in the system is happy to receive traffic for any IP assigned to that system. And that extends to responding to ARP requests - any of the interfaces in the same broadcast domain will respond to an ARP request for any of the system's IP addresses.

You will probably want to address that - look into things like "arp_ignore" settings and the like.

[joe_91]

Rick

Thanks for the reply. Could you please expand on your statement? I am very familiar with HP-UX networking NIC's etc but rather new to RHEL.

Thanks

Joe.

[rick jones · Accepted solution]

I'd suggest some web searching on strong vs weak end system model. It has to do with whether an IP address is considered a property of the system (towards weak end system model) or if it is considered a property of the interface (towards strong end system model). HP-UX tends towards the stronger end of the end system model. If you have two physical interfaces connected to the same broadcast domain (eg what can be reached via layer 2) in HP-UX when the ARP request is received via both of them, only the one on which the requested IP address is assigned will have a reply go out.

In Linux, any interface receiving an ARP request for any IP assigned to the system can and will respond to the request. So, it becomes something of a crap shoot as to which MAC the remote system will think belongs to the desired IP.

So, if you have partitioned the work among several IP addresses, but the interfaces involved are connected to the same broadcast domain, by default, traffic will come and go through any of them for any of the IP addresses.

While you are searching, you might also lookup information on "udev" as you may need to deal with that - Linux's device name persistence heuristics are not quite the same as those of HP-UX.