1

I have a cluster of servers that are all running either with BCM5716C NetXtreme II cards or BCM5709C NetXtreme II cards. All servers have their cards teamed by the BACS3 software as:

  • Smart Load Balancing and Failover
  • No standby members
  • No live link
  • No VLAN

All servers then connect to a pair of Dell 6224 switches. The problem I've been observing is that if one of these switches is off (or on but stuck in a diagnostic loop) then the servers loose all connectivity to anything. I was under the impression that with the above configuration they would just drop their speed and continue with the remaining switch until the one with issues is restored.

Why does all connectivity stop and how do I fix it? All servers are Windows 2008 R2 Standard or Datacenter. Some are used for a Hyper-V cluster.

Greg Askew
  • 35,880
  • 5
  • 54
  • 82
Gup3rSuR4c
  • 661
  • 3
  • 14
  • 29
  • Have you verified that you have connectivity with the good link when its not setup in a nic team? – floyd May 19 '12 at 23:28
  • are the 6224 switches considered stackable? – tony roth May 20 '12 at 16:30
  • @floyd, yes there is connectivity on the links to an active switch when they're not teamed. – Gup3rSuR4c May 21 '12 at 21:33
  • @tonyroth, the switches are not stacked. – Gup3rSuR4c May 21 '12 at 21:33
  • I would verify the arp tables on the server and clients after one of the switches is down. Verify all nics have similar settings/updated drivers etc. You may also need to move one NIC to be a standby member if you dont have 3+ NICs in a loadbalance/failover team mode. – floyd May 27 '12 at 21:41

1 Answers1

-2

Just I tried NIC teaming methods of SLB and LACP with BACS4 on our R730 server.

At first let me share SLB option:

In netxtreme document itself suggesting its feasible for small business servers, so there is no recommendation for datacenter servers.

Also this method is more mean to unmanageable networking switches.

Internally the GARP is not all the way of DC architecture, bcaz the App n DB can't reside same broadcast in DC design.

Now its about LACP: Its looks fine without fallback. The MAC address of 1st NIC added with Teaming binded with TCP/IP stack, during failover it discards all incoming frames on another NIC where suppose it has to accept for. so the fallback won't happen.

So we called up the DELL support, i fell terrifically hard make the engineer to understand the concepts, finally he asked me a LAB for testing and failed to make him a point that we are procuring the server for production networks. So fallback won't happening with LACP. Need to configure like unmanageable era. bcaz its just advanced than that of the era.

Finally the point is It is all not upto the networking standards.

Thanks all arul

arulmozhiarsu
  • 1