ONTAP Discussions

Fabric Metrocluster. Controller-to-Switch Dual-Paths?

parrizas
6,195 Views

Hello,

This is Alfonso Párrias from PS Spain.

In a Fabric MetroCluster configuration is it a requirement to have two paths (cables) from each controller to each of the the two local brocade switches?.

It seems to be,  according to the cabling diagrams included in TR-3548 Metrocluster Design and Implementation Guide.

Is havig a single path (cable) from each controller to each local switch a supported configuration?.

As fas as I know, Data Ontap can only manage two paths to each disk, and with a single path each controller would already could see each disk using two different paths.

And, if this is true, what is the point in having controller-to-switch dual-paths?

Thanks!

1 ACCEPTED SOLUTION

srichard
6,195 Views

HI Alfonso,

yes, a Fabric Metrocluster is designed without *ANY* single point of failure, and also, almost all scenarios with two independent failures are survivable... Therefore, the design calls for the controllers to be connected via two paths to each of the switches.

You are also correct, that the filer (currently) can not manage more than 2 paths (one active, one pre-calculated standby). However, unlike in the V-Series, as the FMC is a closely controlled environment, the filer will use all 4 links (2 per fabric) for load-balancing, while retaining only 2 paths per Disk (storage show disk -p).

For constrained environments (limited requirements for HA, performance, or temporary workarounds), it is possible to have only a single link from the controller to each switch. However, this is discouraged, as any single issue needs to be resolved faster - and with potentially higher impact (maintenance window instead of interrupt-free maintenance).

So, unless there are real pressing issues, go with the depicted 2-links-per-fabric, no-single-point-of-failure

Regards

View solution in original post

3 REPLIES 3

srichard
6,196 Views

HI Alfonso,

yes, a Fabric Metrocluster is designed without *ANY* single point of failure, and also, almost all scenarios with two independent failures are survivable... Therefore, the design calls for the controllers to be connected via two paths to each of the switches.

You are also correct, that the filer (currently) can not manage more than 2 paths (one active, one pre-calculated standby). However, unlike in the V-Series, as the FMC is a closely controlled environment, the filer will use all 4 links (2 per fabric) for load-balancing, while retaining only 2 paths per Disk (storage show disk -p).

For constrained environments (limited requirements for HA, performance, or temporary workarounds), it is possible to have only a single link from the controller to each switch. However, this is discouraged, as any single issue needs to be resolved faster - and with potentially higher impact (maintenance window instead of interrupt-free maintenance).

So, unless there are real pressing issues, go with the depicted 2-links-per-fabric, no-single-point-of-failure

Regards

parrizas
6,195 Views

Thanks Richard.

flavio_onofre
6,195 Views

Hi,

I'm a little bit confusing about Fabric MetroCluster configuration. I would like some help from you, please.

My questions are referenced to "Brocade® 300 and Brocade 5100 Switch Configuration Guide"

1) I know that dual connection between each controller to each FC switch are just to have no single point of failure. Then why when configuring a zone for a 16-Port Cabling MetroCluster the second connection between the switch and controller is not set ?

Command: "zonecreate "STOR", "1,1; 1,5; 1,6; 1,9; 1,10; 3,1; 3,5; 3,6; 3,9; 3,10""

For exemple, should port 1,2 and 3,2 be set on the first fabric in this same zone ? If not, will this connection takeover the first one in a failure situation?

2) Do I must to set switch zoning and TI zoning to have Metrocluster configured ?

Thanks.

Public