Teaming with LACP over multiple switch

[Rokhyui]

Hi all,

Kindly need your advise,

I have 2 Hp procurve 2510 and 1 Hp procurve 2910

Is it possible to do LACP over multiple switch. As i know LACP doesn't provide bound over multiple switch, or any idea to make switch redundancy between SAP server. What the suitable mode teaming need to used, currently used SLB. 

diagram link

[paulgear]

Hi Rokhyui,

The 2510 is a pretty basic managed switch; it doesn't support distributed trunking, and neither does the 2910, last time i checked. You need a 3800 series or better to do that.

I'm not sure what the options are on Windows, but there are plenty of options for NIC redundancy with Linux and VMware that don't require distributed trunking.

[Peter_Debruyne]

Paul is correct, no distributed LACP on these. The new 2920 does support full stacking, so that will work as well.

For your current setup, configure the teaming software with transmit loadbalancing (no lacp/802.3ad). That will tell the nic team to use a difference source MAC address for the packets sent via nic1 and nic2.

This will ensure the switches do not see a mac-flap (like in case of LACP, a single server MAC address would be used).

They call this transmit load balancing (only), since the receive side (upload to server) is not loadbalanced. The server will reply in ARP requests with a single MAC, so all routers/other server on the subnet will send data to that MAC only (using a single path).

That is the key in Receive load balancing teaming config (no config needed on switches, will work with your setup as well) : The server will send an arp reply for some hosts with mac1, for some other hosts with mac2. This results in different hosts sending data to the 2 macs (so the switches will not be confused). In case of nic/link failure, both macs will be active on the remaining nic.

Hope this helps for the current setup, so when you really want the lacp, you need new switches (2920/3800 would be best, 3500/5400 can be configured with distributed trunking, but (for me) it is not as clean as a real stack).

And, as a Comware fan, I can also recommend an IRF stack of 2x5120EI or 2x5500EI switches for this setup, but keep in mind that the cli management is quite different from the provision systems.

Best regards,Peter

[Rokhyui]

Thanks paul and peter for advise,

Thats means, no configuration needed in both switch 2510 and 1 switch 2910. Recently i have a excessive bandwidth in port 2910 that point to 2510. Also have broadcast storm when i try to shutting down 1 of 2510 switch in purpose for testing redunduncy on teaming mode SLB. Is it, cause by SLB mode in server NIC. 

Fyi,  Spanning tree was enable in global mode 2910. Any configuration should i assign. All Server was used HP proliant 350e gen 8 wilth mode SLB Teaming.

[Rokhyui]

Hi, peter and paul

Teaming in HP server only have

1) auto

2) 802.3ad

3) Switch load balancing

4) transmit load balancing

5)network fault tolerance

Which one need to choose

[Rokhyui]

I already used TLB due have 2 switch in between server. In transmit load balancing (TLB) also have a option. Which option i need to used. Please refer attachment

[paulgear]

True stacking on the 2920, eh? That makes them a pretty interesting option for ProCurve-only shops.

[paulgear]

Hi Rokhyui,

If you get a broadcast storm when shutting down a running switch, there's something wrong with your configuration. Make sure that you have the correct switch priorities set and that your root bridge is the device you expect.

[Rokhyui]

Hi Paul,

HP 2910 already set as root, the rest 2 2510 switch, no spanning tree.

Kindly please advise me, how to setup this network between 3 switch with teaming. So far i am try using teaming mode TLB without any configuration in both switch 2510. Result, have many packet lost.

[John Gelten]

The broadcast-storm you mention might be caused by using SLB on the server, that is not a valid choice in your setup.

Also, do not configure any interface-trunking (like LACP) on any of these switches.

The maximum teaming you can use in the server is TLB; this gives you 1x1Gbps bandwidth towards the server (receive), and 2x1Gbps towards the network (transmit).

When using TLB, all traffic from other hosts TOWARDS your server are sent over one of the two links. So if there is actually a lot of traffic in that direction, I can imagine that one of the two uplinks gets 'high bandwidth' warnings. This is by design.

By the way; in many occasions, these warnings are fired when there is just a little spike of traffic, so don't get alarmed only because you see such an event.

[Rokhyui]

Thank John

I will used team type TLB, but which transmit method should be used

1) auto

2) tcp connection

3) destination ip address

4) destination mac address

anyway thanks again for your explanation

[John Gelten]

In 95% of cases, auto works just fine. Only change that if there is a good reason to do so (ie if there are just a few other servers communicating with this server, you may be able to optimize the load-balancing).

[Rokhyui]

Thanks,

I already assign TLB with configure flow control in all switches. Need monitoring the traffic, next week will inform the result.