23

Is there any reason to have 2 NICs on a server BESIDES the following cases?

  1. You need to connect to 2 different physical networks
  2. Redudancy (1 NIC fails, so you use the other)

Are there any other reasons?

10 Answers 10

29

(2a). Load balancing.

(3). Separation of traffic (i.e. you could have a combo web/database server, same network, put all web traffic on one NIC, db traffic on the other, makes it easier to calculate loads for traffic types). This also makes it easier to split the two later on, nobody has to change connection strings.

17
  1. A separate network for backups.

  2. A separate network for console access (these are typically now integrated on the system board for mainstream servers).

  3. Increased throughput (multiple GB networks) bonded together.

2
  • Bonded NICs is the one we use most. 4 x 1Gb NICs is Champion.. Apr 7, 2010 at 22:17
  • 1
    1x 10 gig is better! ;)
    – James
    Apr 7, 2010 at 22:51
6

There's only one other major reason that I can think of, which is similar to redundancy, and that's Trunking, so you could configure traffic to flow over 2x1Gbit ports, giving you 2Gbit of total throughput. Some refer to this as LaCP, linux knows this as mode=4 as described below.

Linux interface bonding modes give a few other options:

mode=0 (balance-rr) Round-robin policy

mode=1 (active-backup) Active-backup policy

mode=2 (balance-xor) XOR policy: Transmit based on [(source MAC address XOR'd with destination MAC address) modulo slave count].

mode=3 (broadcast) Broadcast policy: transmits everything on all slave interfaces. This mode provides fault tolerance.

mode=4 (802.3ad) IEEE 802.3ad Dynamic link aggregation. Creates aggregation groups that share the same speed and duplex settings.

mode=5 (balance-tlb) Adaptive transmit load balancing: channel bonding that does not require any special switch support.

mode=6 (balance-alb) Adaptive load balancing: includes balance-tlb plus receive load balancing (rlb) for IPV4 traffic, and does not require any special switch support.

6
  • 2
    802.3ad == LACP == Mode 4, they are all the same terminology for the same thing. LACP being the most common term used (which is the control protocol specified in 802.3ad)
    – Zypher
    Apr 6, 2010 at 23:32
  • I think I mentioned that somewhere in my answer.. Apr 6, 2010 at 23:38
  • 1
    Is the more common term for what you are describing "bonding"? I have not seen this called "Trunking".
    – iPaulo
    Apr 7, 2010 at 0:00
  • 1
    I've heard both Trunking and Bonding; with trunking being the more popular. The only problem with that, vLans have trunking as well, which can get confusing at times (it's bit me before). Different vendors have different names.
    – Chris S
    Apr 7, 2010 at 1:51
  • 1
    @iPaulo: Windows drivers tend to call it 'teaming' , another word to add to the mix :-) Apr 7, 2010 at 12:02
4

Microsoft clusters require an additional NIC port for heartbeat purposes between nodes.

1
  • Not only Microsoft clusters require this... lots of cluster setups require or at least recommend an additional NIC.
    – MikeyB
    Apr 9, 2010 at 14:36
1

Similar to #2, multiple NICs is also a requisite for high avaliability. Multiple network interfaces (on the same network, but should really be seperate) will be used to monitor which servers are avaliable (a heartbeat). Also, in the case of VMWare, a 3rd NIC can be used for migrating the active state of a virtual machine without interferring with the amount of bandwidth avaliable to the public interface, or the heartbeat.

0

Probably 2 NICs is the minimum required for a server, like other said at least for HA. If you start including console, heartbeat, possibly iSCSI storage... you can easily hit half dozen NICs for a server.

0

This might come under "# You need to connect to 2 different physical networks", but if the server was also a router, it would need two NICs.

0

I always use 2 NICs in a server for redundancy. More than once over the years I've unplugged the wrong cable (usually caused by rushing too much). It's really nice not to have the server disconnect as a result. I've also suffered failed NICs a couple of times.

As far as I'm concerned, if at all possible a server should have redundant everything. NICs are no exception. On the rare occasion I've also needed traffic isolation I add a second pair for that.

1
  • Add failing cables, connectors, and switches to the mix. In the area of virtual voice switching down time is not acceptible. Not during production, not during maintenance, not ever. Where we can live with a mail server or web server being down for a short time (seconds, minutes) this would kill the service on any voice carying device.
    – theking2
    Sep 6, 2021 at 18:31
0

I have never understood the NIC redundancy argument. Without moving parts they rarely fail ... you don't have 2 motherboards in one server, do you? Best be redundant on the whole server level.

However, I like the "human error" redundancy above when unplugging in error.

1
  • 2
    No moving parts? Cables are moving parts, especially when someone's working near them :). Joking aside, it can be a fairly complex path between a NIC and switch (cable -> port replicator -> horizontal cabling -> port replicator -> switch). Any one of those components just might be delicate and fail. Plus, while NIC ports themselves may rarely fail, switches will more often fail. You may need to upgrade the OS on a switch or reload it for some other reason - it's necessary to do so without taking an outage on a whole whack of servers.
    – MikeyB
    Apr 9, 2010 at 14:39
0

Having one interface dedicated to running snort or some other honeypot or IDS, and the other in your internal network is another good reason to have 2 nic cards. Agreed on the load balancing and trunking ideas as well.

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .