FEX-Uplink failover, loadbalancing

[dakota2828]

I wish to understand more the technology within a vblock
I will progressively have a few questions

So here I go. pls be patient and thanks in advance...

VBlock 720 with the following generic config

2x Cisco 5000 Series Switch
2x FI 6248
6x Chassis 5108
FEX 2204
VIC 1240 (no Mezzanine Card)
8x M3 Blade, 2 CPU

4 x 10 G uplinks from each FEX to FI (port channel)
8 x 10 G uplink from each FI to Ciso 5000 series swith (vpc)

Assumptions:

All Chassis are fully populate

Each blade has consistient bandwidth thoroughput but of 6 GB

1. When the blades connect to FEX A and FEX B, are the connections active active?
   
   
2. For exampleIs blades 1 and 5 pinned to the same FEX port(s) on the same FEX?
   
   
3. What is the method that load balances blade pinging?
   
   
4. Assume Uplink 1 fails between FEX A and FI A, is Uplink 1 on FEX B overutilized.
   
   
5. How does the VM decide which CNA port to send traffic?
   
   

That it for now...

[Walter Dey · Accepted solution]

see inline

When the blades connect to FEX A and FEX B, are the connections active active?

I assume you mean FEX=IOM in the chassis;

Yes both fabrics are active; all VIC are dual homed to fabric a and b

For exampleIs blades 1 and 5 pinned to the same FEX port(s) on the same FEX?

You have the choice of static pinning (done by UCS Manager) blade slot to a IOM Uplink, or define a port channel between IOM and fabric Interconnect; port channel is the preferred mode.

Static pinning depends on IOM model, 2204 / 2208 and the number of uplinks, e.g.

2204 and 4 uplinks

chassis slot 1 and 5 -> uplink 1

chassis slot 2 and 6 -> uplink 2

......

2208 and 8 uplinks

chassis slot 1 -> uplink 1

chassis slot 2 -> uplink 2

......

With static pinning, if a uplinks fails, the corresponding slots fabric a or b, are down; there is no repinning to the remaining uplinks; teaming and multipathing will take appropriate action.

If you have a port channel, and you loose a link, no problem, as long as there is at least one link up; your bandwith will decrease, but th vnic/vhba are still up

What is the method that load balances blade pinging?

Don't understand; please be more specific

Assume Uplink 1 fails between FEX A and FI A, is Uplink 1 on FEX B overutilized.

Yes, in principle ! but you have to design oversubscription end to end anyway !

How does the VM decide which CNA port to send traffic?

This depends on the hypervisor; take e.g. VMware ESXi and vswitch: load balancing can be configured per switch and / or port-group; important caveat: don't use IP hash, because fabric interconnect doesn't support LACP; by default its originating virtual¨port id.

[dakota2828]

Thanks Wdey,

I am very appreciate of your quick reply and answers.. bare with me.. Im a growing fan of the Vblock, just also need ot know how it works

and where it is not a good candidate to host workload.

Interms of Load Balancing the Pinning of Blades.

Your answered by explaining that both links to FEX A and FEX B are active  (FEX=IOM)

and portchannel provide redunancy.

"How does the VM decide which CNA port to send traffic?

This depends on the hypervisor; take e.g. VMware ESXi and vswitch: load balancing can be configured per switch and / or port-group; important caveat: don't use IP hash, because fabric interconnect doesn't support LACP; by default its originating virtual¨port id."

Does this mean, when using Orignating Port ID,

     the blade is restricted to the bandwidth of the uplink?

     and VMs on that blade must rely on the hypervisor's vDSwitch for load balancing thier traffic?

     Does this mean if VCenter is unavailable loadbalancing is at risk?

    How does traffic affect traffic ingress to the blade and the VM (assume no Cost of Service is configured)

  Thanks everyone