3750X Stack to EMC SAN Design Question

[bbuadministrator]

Hi Folks,

I'm looking to upgrade my single 3560X iSCSI switch to a new pair of 3750-X Stacked. The switch connects my 2 VMware servers to my EMC VNXe3300. This is a very common setup. The EMC VNXe3300 is using all Gig ports on each controller. Here's more information:

VMware Server A/B----------->connect to single switch 3560x (configured 9000 Jumbo Frames globally)------------------->EMC SAN Controller A = 4x1 Gig Ports/Controller B = 4x1 Gig Ports

In a new 3750-x Stacked environment, what would be the best design and configuration to replace a setup like this?

I'm trying to accomplish redundancy, efficiency and added performance. Your guidance is greatly appreciated.

Thanks!

[Liam Kenneally]

Hi bbuadministrator,

With a pair of 3750Xs you would multihome them to get the best redundancy & thru-put... see below

Server A Nic Port (1 - 2) ----->  3750X(A) G0/47 - 8 

Server A Nic Port (3 - 4) -----> 3750X(B) G0/47 - 8

Server B Nic Port (1 - 2) -----> 3750X(A) G0/45 - 6

Server A Nic Port (3 - 4) -----> 3750X(B) G0/45 - 6

You will then create  2 LACP  etherchannels one for server A one for server b

LACP etherchannel 1 will have G0/47 - 8(SWA) & G0/47 - 8(SWB)

LACP etherchannel 2 would have G0/45 - 6(SWA) & G0/45 - 6(SWB)

This will give a 4gb thru-put unless a switch in the stack fails at which point the thru-put will drop to 2Gb.

Ofcourse the VMware infrastructure will have to support LACP link aggregation.

3750X(A) G0/1 - 4 -----> Controller A Ports 1 - 4

3750X(B) G0/1 - 4 -----> Controller B Ports 1 -4

Now I don't know much about EMC SANs but if both controllers are in a  single chassis then you can bundle all 8x 1gb ports into a 3rd LACP  etherchannel and have and 8gb pipe unless one controller/switch fails at  which point the thru-put drops to 4gb. Again this is assuming EMC supports LACP link aggregation.

IF the two controllers work indervidually, you would take the same steps that I have done with the server. And have 4 LACP Etherchannels.

Here is a document of cross-stack Etherchannels.

http://www.cisco.com/en/US/products/hw/switches/ps5023/products_configuration_example09186a00806cb982.shtml

HTH

Kind Regards,

Liam

[bbuadministrator]

Liam,

I appreciate your information! I believe there is a small error in your details. Correct me if I'm wrong but shouldn't it be like this?

Server A Nic Port (1 - 2) -----> 3750X(A) G0/47 - 8

Server A Nic Port (3 - 4) -----> 3750X(B) G0/47 - 8

Server B Nic Port (1 - 2) -----> 3750X(B) G0/45 - 6 *I put a correction of "B: here*

Server B Nic Port (3 - 4) -----> 3750X(A) G0/45 - 6 *I put a correction of "B" and "A" here*

The EMC SAN has two independant controllers so they can't be grouped together. Are connections to SANs usually LACP etherchannels? How would you setup in this place?

[Liam Kenneally]

Hi bbuadministrator,

Yes you are correct that is what happens when you copy and paste

As stated above if their independant you would multihome them similar to what I did with the servers... I am by no means a storage expert but am pretty sure EMC will support LACP...

My configuration would be something like this ...

interface GigabitEthernetx/x/x(interface(s) for 3750XA)

description EMC_CONTROLLER_A

switchport trunk encapsulation dot1q

switchport trunk native vlan X

switchport mode trunk

speed 1000

duplex full

channel-protocol lacp

channel-group 1 mode active

!

interface GigabitEthernetx/x/x(interface(s) for 3750XB)

description EMC_CONTROLLER_A

switchport trunk encapsulation dot1q

switchport trunk native vlan X

switchport mode trunk

speed 1000

duplex full

channel-protocol lacp

channel-group 1 mode active

HTH

Kind Regards,

Liam

[bbuadministrator]

Liam,

Let me clarify what you are saying to do on the SAN:

LACP #3

3750X(A) - G0/43 - 4 -------> Controller A Gi0/1 - 2

3750X(B) - G0/43 - 4 -------> Controller A Gi0/3 - 4

LACP#4

3750X(A) - G0/41 - 2 -------> Controller B Gi0/1 - 2

3750X(B) - G0/41 - 2 -------> Controller B Gi0/3 - 4

Does this look correct? If so, if switch #A fails, that means I have 2 GB throughput correct?

[Liam Kenneally]

Hi,

Yea you got it :) ...

Both switches up abd running you will have 4Gb thruput... In the event switch A or B fail you will have 2Gb thruput.

The great thing about LACP is also it has error loggin. So if you have a syslog server LACP will log that it has lost Switch A or Bs interfaces from the portchannel.

Kind Regards,
Liam


Sent from Cisco Technical Support Android App

[bbuadministrator]

Liam,

Thanks for your response. Why wouldn't I plug in all SPA links to Switch A and all SPB links to Switch B?

What' the downside to that if any?

Please let me know!

Thanks!

[Liam Kenneally]

Hi bbuadministrator,

What is SPA and SPB?

Kind Regards,

Liam

[bbuadministrator]

SPA is Storage Processor A or Controller A (regarding SAN) and vice versa for B.

-Thanks

[Liam Kenneally]

Hi bbuadministrator,

Oh got you   ...

To be honest the reason I configured 2ports of each controller to each switch was as followers...

I was not sure if controller A & B were redundant controllers for a single Chassis... Or if Controller A was to one SAN and Controller B was to another SAN...

As stated limited knowledge of EMC ...

You are correct in saying SPA ---> SwA  |  SPB ----> SWB.... this is assuming SPA & SPB are redundant fabric for a single chassis does this make sense? This is in my opinion the most straightforward way to do things

But if SPA & SPB are two controllers for a single chassis, I don't see why you cant bundle all 8interfaces? Or am I missing something here?

Kind Regards,

Liam

[bbuadministrator]

Yes Controller A and Controller B are on the same SAN and they are configured to be redundant in case one fails the other automatically picks up.

I'll look into bundling all 8 interfaces. But for now, it is safe to say that the prefferred way of setting up the links between 3750-X stacked and the VNXe3300 SAN is to point all iSCSI SPA Ports to SwA and all iSCSI SPB Ports to SwB with no LACP or with LACP?

[Liam Kenneally]

Hi bbuadministrator,

Oh ok so the controllers are Active/Standby one kicks in when one fails...

Now this makes sence! I was under the impression both Controller A & B were Active(simultaneously passing traffic) Hence bundling all eight interfaces...

You will want to setup as show in first topology with LACP!

Kind Regards,

Liam