IP Multipathing support for FreeBSD

A

AndyUKG

Hi,

does anything exist that provides redundant network connections in FreeBSD? Ie to avoid single points of failure on the NIC, cabling, and Network switch? For example something like Solaris IPMP?
If not is anyone developing this? Presumably it would be possible and fairly easy to port IPMP from opensolaris? As an idea :p

thanks Andy.
 
Hi,

yes these can both be useful, but don´t provide the same functionality as, for example, Solaris IPMP. If we just compare link aggregation for example (speaking just of redundancy with a single host) typically the network hardware requires that you connect all links to the same physical switch, which means you loose out on the redundancy network side (ie what happens if the network switch breaks),

thanks Andy.
 
AndyUKG said:
If we just compare link aggregation for example (speaking just of redundancy with a single host) typically the network hardware requires that you connect all links to the same physical switch,
Click to expand...
As far as I know it doesn't.
 
It doesnt? Are you referring to "network hardware requires that you connect all links to the same physical switch"?
Having working with serveral different Cisco network guys in several different companies they always tell me that, with Cisco kit, port aggregation must be to the same switch. I dont know for non-cisco kit but I imagine the same limitation applies to most vendors.
 
AndyUKG said:
It doesnt? Are you referring to "network hardware requires that you connect all links to the same physical switch"?
Click to expand...
Yes because you can use lagg to switch between wired and wireless. See example 31-3 in the handbook.

Having working with serveral different Cisco network guys in several different companies they always tell me that, with Cisco kit, port aggregation must be to the same switch. I dont know for non-cisco kit but I imagine the same limitation applies to most vendors.
Click to expand...
You are referring to FEC and/or LACP. In those cases lagg is used to increase available bandwidth. Not as a fail-over.
 
Ah, ok. So the failover functionality of lagg this is completely hardware independent and is basing "up status" on link status provided by each NIC? This should be a satisfactory solution, thanks :)

If you arent aware, and in case you are interested. I believe Solaris IPMP provides a superior soltion in a couple of ways.
1) Link status is not only checked via physical link status of the NIC, but also whether some defined IPs (usually the default gateway) are reachable. This can detect some network problems that would go unnoticed with just checking physical link status.
2) Outbound IP traffic is load balanced across all available links (inbound traffic is not load balanced).
3) Physical links may be LACP port aggregated links.

Downside to IPMP is you need an extra IP address for each physical link over and above those that you wish to configure for services etc.

thanks for the information.
 
AndyUKG said:
If you arent aware, and in case you are interested.
Click to expand...
I am aware, I admin about 75 Solaris boxes ;)

I'm not sure what the license is on that bit of Solaris code. I do know Sun donated quite a lot to FreeBSD. Not sure what will happen now Oracle took over though :\
 
AndyUKG said:
yes these can both be useful, but don´t provide the same functionality as, for example, Solaris IPMP. If we just compare link aggregation for example (speaking just of redundancy with a single host) typically the network hardware requires that you connect all links to the same physical switch, which means you loose out on the redundancy network side (ie what happens if the network switch breaks)
Click to expand...

Nope. Link Aggregation Control Protocol (LACP) works across switches. You just need to have switches that support LACP and Spanning Tree.
 
AndyUKG said:
It doesnt? Are you referring to "network hardware requires that you connect all links to the same physical switch"?
Having working with serveral different Cisco network guys in several different companies they always tell me that, with Cisco kit, port aggregation must be to the same switch. I dont know for non-cisco kit but I imagine the same limitation applies to most vendors.
Click to expand...

Then they either don't know what they are talking about, or they want you to buy more expensive Cisco crap that supports it.

This has been available on low-end HP ProCurve switches for years.
 
phoenix said:
You just need to have switches that support LACP and Spanning Tree.
Click to expand...
STP might be the reason why the netadmins told the OP to use the same switch. If not configured properly a recalculation of the tree can take down the entire network for several minutes. Not what you want if only one switch dies.
 
AndyUKG said:
Ah, ok. So the failover functionality of lagg this is completely hardware independent and is basing "up status" on link status provided by each NIC? This should be a satisfactory solution, thanks :)

If you arent aware, and in case you are interested. I believe Solaris IPMP provides a superior soltion in a couple of ways.
1) Link status is not only checked via physical link status of the NIC, but also whether some defined IPs (usually the default gateway) are reachable. This can detect some network problems that would go unnoticed with just checking physical link status.
2) Outbound IP traffic is load balanced across all available links (inbound traffic is not load balanced).
3) Physical links may be LACP port aggregated links.

Downside to IPMP is you need an extra IP address for each physical link over and above those that you wish to configure for services etc.

thanks for the information.
Click to expand...

Sounds like carp(4), but internal to a single system.

On FreeBSD, lagg(4) is used to create redundant links on a single system. And carp(4) is used to create redundant links between identical systems (like clustering).
 
AndyUKG said:
Ah, ok. So the failover functionality of lagg this is completely hardware independent and is basing "up status" on link status provided by each NIC?
Click to expand...

Depends on how you configure the lagg(4) device. If you use one of the "dumb" settings like fail-over or round-robin, then yes, it just looks at the "up" status of the link. If you select one of the "smart" settings like LACP, then it actually sends control packets out to the switch to make sure everything is still working and to actively load-balance and fail-over correctly. The member devices of an LACP connection all work together to make sure things are working.
 
Hrm, weird, we've been doing redundant links to multiple switches for over a year now at one school. Maybe it's just via STP, then? Whatever it is, it works. And we're an all-ProCurve site.
 
You must log in or register to reply here.
Back
Top