Stacking with Cisco Blade Switches

[sgeorgiev]

Hi,

We have a HP p-class Blade System with two Cisco Gigabit Ethernet Switch Modules (Cisco IOS Release 12.2(25)SE) for every enclosure. My question is how I can stack the two blade switches (in the same enclosure) so that every HP blade server to be connected with the two switches in the active/standby state? In other words, can and how I make a port-channel which includes ports from the two switches?

Unfortunately I couldn't found any documentation for the purpose.

[Jon Marshall]

Stanislav

Each server within the blade enclosure usually has 2 connections to the fabric and you can team these together to form an active/standby

As to how to connect these switches to your existing infrastructure - have a look at these white papers from Cisco which cover the recommended ways of deploying them in your network -

http://www.cisco.com/en/US/products/ps6748/prod_white_papers_list.html

If i have missed the point of your question please come back.

Jon

[francisco_1]

Jon is right in saying that you can team the internal NIC's on the blade switches connected to the blade servers so the servers can transit on one NIC and receive on the other and if one NIC is down, the server can transmit using the active NIC.

You can connected the blade switches together in the enclosure using the internal cross-connect ports i believe pors [17-20] but i do not recommend that at all to make your spanning-tree topology clean.

The way have done it before is to disable the cross-connect between the blade switches (No layer 2 connected between the blade swithes) and connect the external ports (port 21-24) as layer 2 uplinks to your core or distribution switches. you can even combine the ports and have a single port-channel for spanning-tree.

Also a useful feature to use is Link State Tracking in Blade Deployments.

see http://blog.scottlowe.org/2007/06/22/link-state-tracking-in-blade-deployments/

Francisco.

[bhedlund]

What model blade switch do you have? Do you have the 3020, or the 3120?

If you have the 3120 you can use the stacking cable to connect the two blade switches in a virtual switch stack, making the two separate switches joined together as one logical switch. Once that happens the blade servers will be able to establish an 802.3ad LACP Etherchannel for true Active/Active inbound and outbound forwarding, doubling the bandwidth of the server to 2Gbps.

Cheers,

Brad

[sgeorgiev]

Hi again,

The HP blade system is p-class with two blade CEGSM switches for every enclosure and four HP BL20p G3 blade servers. Now every server is dual-homed to the CGESMs for high availability. Every blade server is running some virtual host servers. The problem occurs when the active blade switch go down. In this situation virtual servers don't understand about the fail and all services go down.

So my question is it possible to make one ether-channel including interfaces from the two blade switches for example interface Gi0/1 from Switch-A and interface Gi0/1 from Switch-B. The idea is that the blade servers use NIC teaming configuration (802.3ad) but against to the two blade switches.

I hope this time my explanation was more clear.

The switches are:

Switch Ports Model SW Version SW Image

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

* 1 24 371098-001 12.2(25)SE1 CGESM-I6L2-M

[bhedlund · Accepted solution]

Hi,

With the CGESM switches you have you cannot use 802.3ad teaming against the two blade switches. This is possible with the Cisco 3120 blade switches for HP c-class.

Now, in your case, you can use standard Active/Standby NIC teaming, where the active NIC of each server connects to one switch and the standby NIC connects to the other switch. Should a blade switch completely fail, all of the servers with their active NIC connected to the failed switch should see the link go down and switch over to the standby NIC on the second switch. This should work very well.

One problem that can happen however is that the blade switch itself does not completely fail. Instead maybe the blade switch uplink fails, or the upstream switch it is connected to fails. In this situation the server NIC does not experience a failure and continues to send traffic to the blade switch that does not have anywhere to send the traffic, a black hole. This is perhaps the problem you are seeing?

The fortunate thing is that Cisco has a solution to this problem called Trunk Failover with Link State Tracking. What this allows you to do is to tell the blade switch that if it experiences a failure on its uplinks that it should also bring down the links to the servers as well. This will cause the servers to see a failure on their active NIC and perform a switchover to the second blade switch.

I looked up your switch and found that your 12.2(25)SE1 code does not support this feature. You will need to upgrade to 12.2(25)SEE4 and you will be able to configure this on your switches.

Here is a link to see how this configuration works:

http://www.cisco.com/en/US/docs/switches/blades/3020/software/release/12.2_25_sef1/configuration/guide/swethchl.html#wp1346176

Hope this helps.

If so please rate my posts.

Thanks,

Brad

[sgeorgiev]

Hi Brad and thanks for your quick and exact replay.

You rightly understood my question, I want to use 802.3ad teaming with the two blade switches, because my colleagues (server administrators) are notice that when the Active/Standby NIC teaming (which we use now) change its state, the Virtual machines that running over the real blade servers lost its connectivity for 60 seconds approximately.

In fact for 2 years in the real life this situation is not happened, but this blade system is part of real public WEB farm my colleagues want to be sure about the full WEB farm working.

Thanks again!

[kcoley]

While the CGESM does not support 802.3ad, it is not needed for Active/Standby NIC teaming. Set your server software to NFT mode and it will work just find.