what to do with new new uplinks on our hosts

[DITGUY2012]

I have a distributed switch across four hosts that's using 2 uplinks per host. Each of the two vmnics are in an etherchannel on our cisco switches creating a 2 gig aggregate link. the distributed switch has 2 port groups in it. one for dmz traffic and one for server traffic. they both use the same active-active etherchannel but they use separate VLANs.

we now need to migrate some virtual desktops that are on a different VLAN to the cluster. I connected the last two 1gig ports on each host to the cisco switch. What we do with them is what I'm asking. Our thought was already to use a separate VLAN which we did. However the concern was - what if something saturates the line on the desktop VMs. We don't want that to affect server traffic. the way I see it we have a couple of options:

1. Add the last two links to the same distributed switch but create a new port group and only use those two links in that port group not on the others. I don't quite know what the benefit is here by adding them to the same distributed switch if they're not going to be usable by the other port groups.

2. create a new distributed switch with a single port group and add them to that network. not sure on the benefits of this option either.

3. similar to option 1 but create a single 4gig aggregate. create a new port group for that vlan. then because we're using 5.1 enterprise plus edition create network IO control rules that prevent that portgroup from ever using more than say 1 or 2gig (25-50%). this allows more bandwidth for servers if they need it while preventing desktop bandwidth sprawl from affecting servers negatively. the only downside I can see here is that it complicates things.

Any advice anyone?

[rcporto]

1. Add the last two links to the same distributed switch but create a new port group and only use those two links in that port group not on the others. I don't quite know what the benefit is here by adding them to the same distributed switch if they're not going to be usable by the other port groups.

The benefit here is that you will need to manage only one vDS.

2. create a new distributed switch with a single port group and add them to that network. not sure on the benefits of this option either.

The benefit is that you can limit the impact of erroneous configuration, but you will get more administrative effort to manage multiple vDS.

3. similar to option 1 but create a single 4gig aggregate. create a new port group for that vlan. then because we're using 5.1 enterprise plus edition create network IO control rules that prevent that portgroup from ever using more than say 1 or 2gig (25-50%). this allows more bandwidth for servers if they need it while preventing desktop bandwidth sprawl from affecting servers negatively. the only downside I can see here is that it complicates things.

Instead of use LAG, why not use make the things easy and use LBT: » Etherchannel and IP Hash or Load Based Teaming? Long White Virtual Clouds

[DITGUY2012]

Our cisco switches are already trunked and doing LACP for our other hosts and VDSs. The setup wasn't too complicated. But thanks for the article. Was a good read.

By the way, my network engineer was asking about how it does the seudo "qos". My understanding is that the desktop VLAN can use more than it's allocated so long as there isn't contention for that bandwidth. If in total there was contention then the percentages come into play to make sure a SLA is met.

His question is how VMware handles congestion. Does it end up dropping packets? Does it use session based control where it tightens the pipe of the session so that it can't talk as fast? Hopefully I'm not butchering this.