DESIGN ANALYSIS
Today's stacking technology offers high capacity and highly resilient configurations similar to chassis switches, which are a good fit for smaller companies like TacDoh, our fictional purveyor of deep-fried snacks in retail stores. Chassis have more capacity in the backplane but are generally more expensive, while stackables are more flexible and usually less costly. 3Com's stacking technology, called Expandable Resilient Networking (XRN), can pile on eight switches that act like a single unit.
Where older stacking technology had a single path from the top of the stack to the bottom, modern iterations like XRN create a ring topology, where the switches at the top and bottom of the stack are also connected. Now there are two paths for traffic to pass through. If a switch in the middle of the stack fails, the rest of the devices are still interconnected through the remaining path.
The failed switch can be replaced with a new one, and once the configuration is restored to the new switch, the stack is fully operational.
In addition, 3Com has added redundant uplinks between the access switches and the core switch stack, made up of three 5500G model switches. For locations that called for more than two switches in a stack, 3Com specified four aggregated 1-Gbps uplinks for a combined 4-Gbps capacity.
The XRN stacking allows uplink ports from any switch in the stack to be aggregated, increasing overall capacity as well as providing redundancy. If one uplink fails, it's simply dropped out of the aggregated pool. Locations with 48 ports or fewer aggregated two 1-Gbps links, while the data center stack aggregated two 10-Gbps uplinks to the core.
The downside of stacking is that the maximum bandwidth is limited to the capacity of the stacking cable. According to 3Com, XRN stacking can support 96 Gbps of throughput, but that is broken down as 48 Gbps per cable in full-duplex mode, where 24 Gbps can be transmitted and 24 Gbps can be received. With two cables, one going "up" and the other going "down," the combined bandwidth totals 96 Gbps.
Now, with just two switches, we could, in theory, get full bandwidth to each port. But add another switch, and we could potentially overrun the available capacity.
To be clear, we don't consider the 96-Gbps capacity limit to be an issue for our access switches, as we expect little communication between hosts on the same switch stack. Traffic on the access switches that is destined for the data center, however, will be limited by the uplink capacities, and we might run into some scaling issues at the core as the company grows in size.
At that point, we would investigate modular chassis products, which typically have much faster backplanes.
