xy - Fotolia

Get started Bring yourself up to speed with our introductory content.

What you need to know about switch upgrades

Next-generation switches have a lot of capabilities, especially as vendors retool their product lines to support application awareness and other features. But make sure you get the switch you need.

You may not be doing software-defined networking (SDN) in your data center this year or next, but you will be within the lifespan of the next data center switch you buy. You should have your eyes on that future, but it shouldn't be the only thing guiding your selection of a new switch. Other things to focus on include the standard L.C.D. triumvirate (latency, capacity, density), virtualization support, fabric support, security features and application support.

L. C. D. triumvirate

Organizations continue to shift to a service-oriented architecture (SOA) model for the applications they buy and develop. Moving to SOA takes conversations and data transfers among application components that previously took place in memory on a single server and puts them on the network. Consequently, low latency becomes more desirable every day. Your next switch should be able to push packets through with latencies under 4 microseconds; really low latencies are under 1 microsecond.

As for capacity, 10 Gigabit Ethernet (GbE) has some years left in it, but 40 GbE and up will serve you better if you are aggressively virtualizing the next 30% of your workload, are still increasing virtual server density on your host servers, or are using Ethernet-based storage (such as iSCSI, FCoE, ATAoE and so on).

But 40 GbE and up will serve you better if you are aggressively virtualizing the next 30% of your workload.

Density isn't as concerning as it was before serious server virtualization and the advent of blade servers, and while these technologies have slowed the explosion in physical servers that are needed to connect, they have not eliminated increases in the number of ports required. Instead, systems that might have had one con­nection in the past are now getting two; those that might have had two are getting four; and so on. (And, of course, not everyone uses blades.) Add in connections for management networks and the migration of storage, and port density still makes a difference. So keep your eye on densities above the typical 48 per 1 U; double that is possible.

Fabric and virtualization support

In addition to the table stakes of network vir­tualization -- flexible support for VLANs -- your next switch needs to support newer flavors like VXLAN and NVGRE. These can be used to overlay a virtual network on a physical switch structure that is unaware of them, but participating switches can improve services, reliabil­ity and performance.

Supporting the virtual networks overlaid on the physical one is part and parcel of being the network fabric -- a single giant distributed switch -- underlying a flattened data center architecture. As noted earlier, flattening also pushes IT to higher densities and lower latencies, so fabric-oriented switches emphasize them as well as consolidated management (managing all the switches or cards as a single entity).


In addition to old standbys -- VLANs, port isolation, access control lists and authenti­cated access to the network -- watch for more general security services -- whether they are high-level, like DDoS detection, or low-level, like packet source verification or encryption offload for servers. Expect each piece of your infrastruc­ture to be more flexible as a piece in your de­fense-in-depth strategy.

Application support

Lastly, look for application awareness in switches -- features that differentially help improve the performance of specific applica­tions. The most straightforward way is to work robustly with class-of-service and quality-of-service markings on packets, to ensure those tagged for lossless delivery or ultra-low latency get the kind of special attention they need. Newer switches may provide more robust prioritization based on protocol detection, and include services such as rate guarantees, rate limits and packet reorganization for optimal transmission.

As always, make sure data center switches have a management platform geared toward minimal hands-on time and maximal automation. And with SDN on its way, make sure new gear is OpenFlow-ready so each device can play whatever role is required of it.

Next Steps

Cisco switch signals new landscape

Extreme easing the way to 40 Gigabit Ethernet

Curtain opens for 100 Gig uplinks in Arista ToR switch

This was last published in July 2014

Dig Deeper on Network Infrastructure

Join the conversation

1 comment

Send me notifications when other members comment.

Please create a username to comment.

Are you actively assessing whether to upgrade your data center switch architecture?