WHY UPGRADE?
There are plenty of reasons to upgrade a switching infrastructure. When asked what technical drivers would induce them to upgrade, respondents to our survey cited more capacity in the access, distribution, and core layers, and more capacity in the data center as their top technical reasons for upgrading. The need to improve fault tolerance wasn't far behind.
The top business drivers among IT professionals we surveyed were to improve network security, support real-time communications, and improve fault tolerance and network flexibility (see chart, "How important are these switch upgrade factors?", below).
Performance is very important. In this Rolling Review, we considered a few factors that affect performance, such as how devices are interconnected and what types of switches (stackable or chassis) are used at the edge and the core. In addition to raw throughput, TacDoh's plan for VoIP requires a low-latency network.
Redundant 10-Gbps links were common in all the designs. The difference lies in whether multiple links can be aggregated, accumulating bandwidth as needed. Extreme's uplinks were active/passive, which limited us to 10 Gbps. Designs from Foundry and 3Com aggregated multiple 10-Gbps uplinks for increased bandwidth and high availability, which provided plenty of room for growth. Alcatel-Lucent's edge switches can support two 10-Gbps ports each, but the company specified a single core in its design, so adding more uplinks would only add more capacity--the sole core switch is a single point of failure.
HP ProCurve's design includes redundant, aggregated 1-Gbps uplinks to the core/server switches. This provides only 2-Gbps capacity, but the 5400zl chassis line can be upgraded to 10 Gbps in the future. This fit within the requirements of our RFI, so the lack of 10 Gbps doesn't detract from the design. The inclusion of 10 Gbps from other vendors means fewer upgrades in the future.
Comparing stackable switches versus chassis at the edge is really a comparison of cost, flexibility, and performance. The stackables from all the vendors are remarkably similar. All boast a bidirectional ring topology so traffic can pass in either direction on the ring, up and down the stack, via the shortest route. Switches can be added anywhere in the stack up to eight high, for a maximum density of 384 ports in 48-port increments. Stacking access switches, which Alcatel-Lucent's, Extreme's, and 3Com's designs offer, is a lower-cost way to add ports incrementally.
Foundry's SuperX chassis supports up to 206 Gbps ports, while HP ProCurve's 5406zl access switches have a maximum capacity of 144 Gbps ports. We would have to add whole new chassis plus uplinks to add capacity, which is more costly than adding a stacked switch. The main difference, however, is the capacity between hosts on the same switch. Many stacked switches, such as those from 3Com, only push 48-Gbps bidirectional traffic: 24 Gbps up and 24 Gbps down over the stacking cable. With two cables, that's 96 Gbps total. Foundry's SuperX, however, can pump 510 Gbps over the chassis backplane.
At the access layer, even a seemingly paltry 24 Gbps is more than enough capacity in situations where there will be little connectivity between hosts. The added capacity is required at the core, however, where all traffic is aggregated and redistributed. All the designs except 3Com's had chassis solutions at the core. 3Com has chassis switches, but chose a stacked core for the TacDoh network design. The capacity offered by the 3Com design would fit the needs stated in the RFP, but we foresee having to migrate to a chassis in the future.
Discover the benefits of scalable backbone and network services for application and cloud providers.
Network automation is gaining momentum. Here's how you can drive this powerful technology to its full potential.
Subsea cable alternatives can provide a level of comfort for those concerned about disruptions. Realistically, they will continue to play a small, but critical role in the short term.
