network load balancing vs HP Teaming

[bookouri]

I currently have two Oracle Application Servers set up with in a network load balancing cluster. Each machine currently has one NIC included in the cluster. Each machine has two more NICs that are not currently used. Users are compaining about performance so I want to start using the unconfigured NICs. I can go into NLB and add the additional nics, that's not a problem. But, I could also use the HP server's NIC teaming software to team the NICs. Does anybody have any suggestions about which method would be preferred? Do I add the unused NICs to the NLB cluster or do I "team" the network cards and use the "team" address in the NLB cluster?

any comments would be appreciated..

 

[Zelandakh]

The latter.

 

[jet042]

Zelandakh is right. The latter option will give you much greater increase in bandwidth.

Assuming these are all gigabit NICs, if you just add them to the load balancing you will end up with 6 1Gbps pipes that packets are sent to according to an algorithm to make sure no one pipe gets over-loaded. If you team three NICs first, then add the team to the load balancing, you will get two 3Gbps pipes in the same configuration. Basically, the first option gives you 1Gbps of throughput, the second give you 3.

 

[bookouri]

great, thanks for the input

ill team them up and see how it works...

 

[victorv]

hi,
all these things MS-NLB, Teaming, are triks:

the simple association:

1 NIC <-> 1 IP address <-> 1 MAC-address

is masked by these methods.

Sometime 1 trik goes, but 2 triks sometime go with
more difficult.

Are you sure low perfs come from NIC ?
(Task Manager, Network Tab)

If you don't need more network performance,
you have already NLB for raliability:
caution to complicate system.

However you have to try !

bye
vic

 

[58sniper]

Assuming these are all gigabit NICs, if you just add them to the load balancing you will end up with 6 1Gbps pipes that packets are sent to according to an algorithm to make sure no one pipe gets over-loaded. If you team three NICs first, then add the team to the load balancing, you will get two 3Gbps pipes in the same configuration. Basically, the first option gives you 1Gbps of throughput, the second give you 3.

If you load balance 2 1 Gig connections, they provide 2GB of bandwidth, since each can handle 1GB at a time, simultaneous to the other. If you team them, you get 2GB of bandwidth, since you have one connection over 2 1 GB connections. The results are the same. My testing has shown that all things being equal, you'll end up at the same result regardless of whether you're load balancing or teaming. The caveat being that the team is configured for speed and not failover.


You need the load balancing in order to spread the load over the two servers. Teaming (each servers NICs, then load balancing them) would make more sense because it would more likely ensure that connections were balanced between the servers than if you load balanced each of the NICs. This is because with teaming, you'd have one (virtual) connection for each server. So the traffic goes server a/server b/server a/server b. With load balancing all of the NICs, you could end up having traffic going like server a NIC1/server a NIC2/server b NIC1/server b NIC2. It would be more difficult to ensure the load was properly balanced.

Make sense?

Pat Richard MVP
Plan for performance, and capacity takes care of itself. Plan for capacity, and suffer poor performance.

http://www.ucblogs.net/blogs/exchange/

 

[NortonES2]

How is the switch configured that the server is patched to, Etherchannel?

-------------------------------

If it doesn't leak oil it must be empty!!

 

[bookouri]

No etherchannel..

"Teaming (each servers NICs, then load balancing them) would make more sense because it would more likely ensure that connections were balanced between the servers than if you load balanced each of the NICs. "

Thats what I plan to do. The big unknown is how well NLB and the teaming software solutions play together.

 

[NortonES2]

NLB and teaming works OK, have you looked at the NIC throughput to confirm that this is the bottleneck as I have never seen a Windows server hit the limit of a 1Gb NIC.

-------------------------------

If it doesn't leak oil it must be empty!!

 

[acl03]

We had to set up a port group on the cisco switch to get incoming teaming to work at 2Gb/sec. If you don't, you only get outgoing.

Turns out our disk was the bottleneck anyway, so going over 1Gb/sec was a waste of time.

Also consider that having HP/MS teaming is one more driver that can fail.

Thanks,
Andrew

Hard work often pays off over time, but procrastination pays off right now!

 

[bookouri]

yes, having one more point to fail is one of the reasons I started researching this.. right now there is only the NLB and if I add the teaming into that, I immediately have one more potential problem.

I've been thinking about the switches too. We have a vendor who maintains all our switches so I dont get involved in that end of things very often. I was going to take the quick and dirty route and throw some more bandwidth at it and see if we got any immediate improvement.