STP causes temporary disconnects

[vMario156]

I having a problem with a network design and need some help / advice, because networking / cisco switch configuration isn´t my strongest skill.

In this szenario blade servers are used as ESXi hosts. The blade chassie contains 2x 10GbE Connection Blades (Switches). Each switch has an uplink to an Cisco (3750x) stack, connected to the TenGigabitEthernet Ports.

If I disconnect an uplink to the cisco switch, the connection blade shutdown all internal ports to the blade servers and the ESXi host "detect" vmnic0 (for example) as disconnected and uses vmnic1, which is in the nic team in most port groups (in some also as standby).

Now the problem is when I connect the uplink again. By fact vmnic0 is active within the nic team it comes directly online at the ESXi host, because the connection blade enable all blade ports directly when the uplink cable to the cisco switch is plugged in again.

The port at the cisco switch needs around 30 seconds to get into the forwarding mode. Durring this period I am getting of course problems with the ESXi hosts, which "think" that vmnic0 is up and running again.

I know that this problem is cause by STP and that you always need to set PortFast on the Cisco switches to the ports where your ESXi hosts are connected. But in my case the ESXi host isn´t connected directly to the Cisco switch. I talked to the network admin, who told me that he can´t enable PortFast on a port where a switch is connected (in this case the connection blades), because this would end up with network problems.

So my questions to the network expert here :smileyhappy: : What is the best to do in such a szenario? I think it´s not such an unusal setup.

Thanks for every advice.

Regards,

Mario

[beckham007fifa]

why can't you have STP disabled and port fast enabled, if both can't go parallelly.

[iw123]

Which mode of spanning tree is running on the network? using RSTP or RSTP+ should reduce the 30 second delay significantly. I guess your reasons for not setting 'no spanning-tree trunk' on the cisco up stream connection is due to there being non-esx servers in that enclosure?

Which switches are sat between the esx blades and the cisco switch?

[vMario156]

Thanks for the quick response. I will talk to the network admin and find out which STP mode the cisco switches are using.

The blade chassie have 2 of this connection blades: http://globalsp.ts.fujitsu.com/dmsp/Publications/public/ds-py-bx-cb-eth-switch-ibp-10gb-188.pdf

They are set to MSTP at the moment, but also support RSTP I saw.

[iw123]

Ah, you would have to be careful before changing anything because the same version of STP should be used in an stp domain. Do you have other servers in your blade chassis or are they all ESX?

[vMario156]

The cisco switches runs with PVST and the Fujitsu with MSTP.

Only ESXi Servers are used in the blade chassie.

[iw123]

can you enable the portfast equivalent on the ports connecting to your esx hosts?

[rickardnobel]

vMario156 wrote:

The cisco switches runs with PVST and the Fujitsu with MSTP.

Only ESXi Servers are used in the blade chassie.

MSTP is an upgraded RSTP, but is without real advantege if not properly configured (defining "instances" and other). An MSTP switch should however go down to RSTP behavior if talking to a non-MSTP switch.

PVST is a Cisco proprietary protocol which might not be totally compabile with your other switch.

Could you see if the Cisco switch has support to change into standard RSTP?

[chriswahl]

With a blade environment, the chassis switches are typically running in end-host mode (or transparent mode) and do not participate in STP. They are unable to form a loop and do not accept or send BPDUs. In that situation, it is safe (and recommended) to turn on portfast on the Ethernet connections to the upstream switches.

I've not worked with Fujitsu gear, but it holds true for HP BladeSystem and Cisco UCS.