Sunday, April 14, 2013

Resilient backbone with stacked 3Com 4050 and D-Link DGS-3324SR

As part of our network resiliency project, we decided to deploy a redundant trunk connection between the main and secondary server rooms. The main room houses our core 3Com 4050 cluster, and the secondary has a pair of D-Link DGS-3324SR boxes. Both of them support multi-chassis trunking, which is very nice of them.

3Com 4050 clustering

The 40x0 series supports the 3Com XRN stack technology, which can be implemented for long-distance on a GBIC, or short-distance with an XRN fabric module (which we use). Here comes the XRN fabric FAQ:

  • The fabric bandwidth is 8Gbps;
  • Link aggregation (trunking) uses IP-based hashing, or falls back to MAC-based hashing. It's not configurable;
  • Distributed link aggregation has "Intelligent Local Forwarding": it tries to avoid using the fabric link if possible. If a frame was received on switch1, it'll use a link aggregation member link on switch1, not on switch2. This can lead to some asymmetric traffic flows.
When building the stack, you're best off with the new member reset to factory defaults. I actually had to reset both members, as I had some issues when entering the Bridge/Vlan menu, and it crashed both CLI and GUI. I had the same issue with some 3Com 4400 stacks.

D-Link DGS-3324SR cluster

This platform supports clustering more than 2 switches, using HDMI-like stacking cables. You can use linear or loop topology. Stack bandwidth is 10Gbps.
Multi-chassis aggregation is supported, I just had some issues with the 'traffic segmentation' feature. The GUI only reported "error", the CLI was more helpful. Traffic distribution can be configured system-wide for a combination of MAC/IP, source/destination/source+destination. No such thing as the 'ILF' on the 3Com, traffic will leave the ports based on the hash.

Plugging in the wires

Both switches support LACP and static, I went with static. I had some RSTP issues earlier between them, so vlan1 untagged on both sides is a must, even if you use something else as the management vlan. I even managed to achieve extra redundancy by not using the same 8 core fiber for both connections, but routed one of them through another building.

Monitoring

The 3Com provides an SNMP object for the XRN fabric interfaces, so you can monitor the bandwidth of the stack link (no such thing on the DGS), but it won't provide an SNMP object for the aggregated AL1-13 interfaces. MRTG can be configured to do basic arithmetic, so you can measure the bandwidth of multiple aggregated links together.

No comments:

Post a Comment