This post is more than 5 years old
2 Intern
•
12 Posts
0
1894
August 10th, 2018 11:00
Microsoft NLB on a 3 unit S4048T-ON stack with L3 routing
I recently upgraded a core switch stack from some older PC 6248's to 3 S4048T-ON units running OS version 14.0. Everything is working fine except some MS NLB clusters running in unicast mode. What appears to be happening is the switch stack is learning the the NLB cluster MAC and VIP and pinning it to a specific port in the ARP table.
If I run the command (where 11:22:33:44:55:66 is the NLB cluster MAC):
show mac-address-table address 11:22:33:44:55:66
I don't see the MAC address listed, as I would expect the switch stack to not learn it.
However, when I run the command:
show arp macaddress 11:22:33:44:55:66
The following results are displayed:
Protocol Address Age(min) Hardware Address Interface VLAN CPU --------------------------------------------------------------------------------------------- Internet 10.10.0.10 0 11:22:33:44:55:66 Te 2/23 Vl 30 CP Internet 10.10.0.11 0 11:22:33:44:55:66 Te 2/23 Vl 30 CP Internet 10.10.0.12 126 11:22:33:44:55:66 Te 1/14 Vl 30 CP
10.10.0.10 is the cluster VIP, .11 is node 1 and .12 is node 2. VLAN 30 is route-able with an IP of 10.10.0.1. in this situation, node 1 is responding to requests and things work, however if I drain stop node 1, node 2 fails to respond.
Running a packet capture on both nodes shows that when node 1 is drained, node 2 fails to ever get the packets that should be port flooded to all nodes on the VLAN. This leads me to believe that the stack is not respecting the port flooding setting and is instead routing based on the arp table.
I have read all the documentation I can find on this, and searched far and wide. I really don't see anyone else having this issue, so I feel like maybe it's my configuration.
According to the information in https://www.dell.com/support/article/ae/en/aebsdt1/how10789/dell-networking-force10-switches-and-microsoft-network-load-balancing?lang=en and https://downloads.dell.com/manuals/all-products/esuprt_ser_stor_net/esuprt_networking/esuprt_net_fxd_prt_swtchs/networking-s4048t-on_user%27s-guide19_en-us.pdf (page 569).
What am I missing here? Thanks for any help.
Edictsystems
2 Intern
•
12 Posts
1
August 16th, 2018 07:00
I opened a support case and got my answer. Just responding in case someone else runs into this.
Our issue was related to the CAM ACL table. It was not sufficiently configured to support enough NLB addresses, even though the number of NLB clusters we had configured was below 8.
This can be identified by looking at the switch logs and looking for errors similar to the following:
This shows each stack members error of insufficient space in the CAM for NLB addresses. The nlbclusteracl setting for the CAM ACL needed to be increased to either 1 or 2 from it's default value of 0.
In our case, we set it to 2 using the command provided by support:
Hopefully the documentation gets updated to include a section under the MS NLB configuration to point to this problem.