Help with redundant switches

[TheTone]

Can somone tell me if it is required to join two switches together when setting up redundant paths and is so why does this need to be done?

To give an example I have one ESXi 3.5 box with 2 Uplinks teamed together for software iSCSI storage. In this team I have one cable going to one switch and another cable going to another switch. My san also has the same setup for teaming.

Do the switches need to be connected togther and if so why?

Thanks

Tony

[Chuck8773]

I assume your SAN has two IPS? One on each switch and they are different IPs? What would happen if your host would connect to the ip on the other switch? Also down the road if you add another SAN array, if it can join the other one as EqualLogic can, those two need to be able to talk to each other always.

It may work without connecting the switches, but it is a best practice, and less prone to issue, to have the SAN arrays on the same fabric.

Charles Killmer, VCP

If you found this or other information useful, please consider awarding points for "Correct" or "Helpful".

[TheTone]

The SAN IP address has a single IP address. It is a bonded connection using balance-alb load balancing. Is it best to have the switches connected together and if so why does this need to be done?

Thanks

Tony

[Chuck8773]

Is it possible to set it up with an ip for each nic? If you do, then things like MPIO are available. If you have one IP, you will never get above 1 Gbps from the host to the SAN. With two IP's, you can use MPIO and get better performance, while maintaining reliability and availability.

In your setup, it sounds like you will be fine with the way it is. I would set it up with two ip's and connected switches.

What kind of SAN is this?

Charles Killmer, VCP

If you found this or other information useful, please consider awarding points for "Correct" or "Helpful".

[TheTone]

The is Open-E DSS. I dont understand what you mean with MPIO. We are using software ISCSI initiatior. I thought you could not use MPIO?

I was planning on creating a seperate bond on the SAN using the same balance-alb load balancing and using a different IP address. I would then allow certain IP addresses (initiator addresses) access to certain targets so that the pNIC's were balanced across the volumes. On ESX I was thinking about creating a single vSwtich with 2 VMkernal portgroups for ISCSI. Then giving each portgroup an IP address on the same network as the two SAN bonded IP addresses. I would then set each port group in the vSwitch to use a different preffered pNIC and have failback enabled. Would this method give me 2GB to the SAN and is what I propose correct?

The other method I have also thought about is keeping things simple and using LeftHand recommendation on how they setup their SAN. This would be to have a single vSwitch with 1 VMKernal portgroup with 2 pNICS for iSCSI Storage using the default load balancing with failback set to No.

Would using the LeftHand method be slower than my current recommendation?

I just think somtimes its best to keep things simple but if it definatly works and can speed up the performance I will do that.

Thanks

Tony

[Chuck8773]

Correct, ESX 3.5 cannot use MPIO. ESX 4 can. I would set things up in preparation for that.

Also, in ESX 3.5, I don't believe you can have multiple vmkernels in the same subnet. I know that you cannot use serparate vmkernel in the software iSCSI though.

Charles Killmer, VCP

If you found this or other information useful, please consider awarding points for "Correct" or "Helpful".

[TheTone]

So is it best to use the SAN's mode of bonding or use the vSphere MPIO to do the load balancing and failover? Or would it be best to do both. I could use the 4 pNICS on the SAN to create two ALB bonds. Both with different IP address on different subnets. Then when I upgrade to vSphere in the future use the 2 bonded IPs for MPIO.

Thanks

[Chuck8773]

That sounds like a good compromise between the two setups. I am not very familiar with that SAN so cannot give you a best answer. But that sounds good to me.

Charles Killmer, VCP

If you found this or other information useful, please consider awarding points for "Correct" or "Helpful".

[TheTone]

Ok thanks, I will try that. Now back to the original question. Is it required to trunk the two switches together and if so what does this do? I have noticied most diagrams have the two switches connected togthere aswell as the nodes connecting to each switch. What is the purpose of connecting the two switches together?

Thanks again

Tony

[Chuck8773 · Accepted solution]

The typical purpose of connecting the two switches together is to make them behave as one large switch and to increase redundancy.

Charles Killmer, VCP

If you found this or other information useful, please consider awarding points for "Correct" or "Helpful".

[TheTone]

Thanks