HP Network Teaming - SLB & Cisco switch

[KHuysmans]

Hi,

One of our servers (Windows 2003) has 2 Gbit NIC's and we have set them up with HP Network Teaming and Switch-assisted Load Balancing. Both NIC's connect to a Gbit port on a Cisco Catalyst 3550XL switch.

I have read the interesting parts of the HP document on Network Teaming (ftp://ftp.compaq.com/pub/products/servers/networking/TeamingWP.pdf), it was of great value to help me understand how this works.

Now, with SLB configured on the server, I see that Incoming (RX) traffic is being load balanced. However, all documentation I have read on the subject specificly explains that for SLB to work there has to be bundling set up on the switch. On my Cisco switch, that would mean configuring etherchannel. This is currently NOT configured.

I am trying to figure out how it can be possible that SLB is functioning, id est that incoming traffic is being load balanced, since that has to be done on the switch and the 2 ports to which the 2 NIC's of my server are connected are configured as "simple" access ports.

You might wonder how I have tested the load balancing. This actually puzzles me even more. I test this by copying a file to the server over a network share. The only thing my switch is capable of load balancing (if it were to be configured, which it isn't!), is based on MAC address. Since all of this traffic I am sending is originating from the same host and thus the same MAC address, how come the ingress traffic on my server is being received on both NIC's, thus achieving load balancing.

When you configure HP teaming with SLB, the "lesser" options NFT and TLB are also supported. I thought for a while that SLB might not be actually functioning and that this type of behaviour could be explained because instead of working in SLB mode the HP Teaming was working as TLB. In TLB (Transmit Load Balance), load balancing is achieved on traffic outgoing from the server. At layer 2 this traffic gets the source MAC address of the actual egress port and I thought that my sending device would respond to this traffic by sending it back into the ethernet segment to the MAC address with which the server originally sent it (if that made any sense at all). Bottom line though, I manually switched the Team Type Selection to TLB and noticed it didn't work anymore.

So... My HP teaming type is currently SLB and it seems to be working correctly (both my NIC's on the teamed server are Gbit, the server sending traffic towards the teamed server over the network share is also on Gbit, when I am sending data, I can see on the teamed server that 1 port's RX is at about 300-400Mbps and the other is like 50Mbps. Granted, this isn't a perfect load balancing, but it sure beats the 0 bps I get on the second NIC with any other method).

Now, does anybody have any idea how this is being achieved? Because I am at a loss and would love to understand how this works.

Thanks for reading so far!

With kind regards,

Kevin Huysmans

[Matthijs Wijers_1]

Hi Kevin,

Did you enable LACP on your switchports?
If you look at the team properties information, what current mode is the teaming operating in?
Switch-assisted Load Balancing (SLB) or 802.3ad Dynamic?

NFT and TLB are almost always supported without any special switch-side configuration.

For receive load-balancing to work there's always some help from the switch needed, LACP or trunk/channel config.

Regards,
Matthijs