Next-stage high-density switch options: What's right for you?

All major switch vendors have rolled out new high-density, low-latency data center switches. Expert Glen Kemp looks at what makes them special.

While it's true that software-defined networking (SDN) will likely become the de-facto method to design and deploy networks at some point in the future, today's network architects have more pressing problems. Implementing SDN takes development and process change; none of which will occur overnight.

So here's the dilemma: sold on the concept of SDN, but operational realities stand in your way; your physical network edge needs a capacity injection right now. If you are in the luxurious position of having money to spend, which of the most recently announced high-density switch options will best meet your needs today and your SDN requirements tomorrow?

Vendors adopting silicon necessary to bridge physical, virtual worlds

The reality is there is not one switch shooting for this particular goal, but several. Many vendors have adopted Broadcom Corp.'s Trident II silicon; it's a 1.28 Tbps, Layer 2 and Layer 3 chipset that supports 96 10 Gigabit Ethernet (GbE) ports or 32 40 GbE ports within a single rack unit. With such a level playing field, vendors have to find new ways to differentiate themselves. A common chipset means that raw packet performance figures will be exceptionally close and the usual "speeds and feeds" alone won't be enough to allow vendors to differentiate their switch from another's. The price and performance ratio makes Trident II designs attractive as not only a top of rack (ToR) switch in a fancy SDN-flavored data center, but also as a dense core in a smaller network. Arista, Cisco, Dell, Extreme, HP and Juniper Networks are starting to ship products based upon the Trident II and are now all attempting to differentiate themselves. A rundown:

Arista Networks 7050X

So here's the dilemma: sold on the concept of SDN, but operational realities stand in your way; your physical network edge needs a capacity injection right now.

Arista Networks Inc. was one of the first vendors to have access to the Trident II silicon; the announcement that it had production customers using the 7050X platform stole some of the thunder from Cisco's Nexus announcement. This early availability is likely to pay dividends in terms of product performance, stability and ongoing support. Being at the cutting edge meant that Arista was able to shape the product in relative privacy and take onboard the lessons learned before a widespread product launch.

Cisco Nexus 3100

The Nexus 3100 is one of several Nexus high-density switch devices that use the Broadcom chipset. Perhaps unsurprisingly, Cisco has taken a different approach from the rest of the herd. Rather than rely upon adding value to reference designs, Cisco has developed custom application-specific integrated circuits to complement the Trident II. The goal is get the best of both worlds: high performance but with low power consumption and high reliability.

Yet, for many, the 3100 will be a portfolio sale; driven by large organizations and their constant craving for switch capacity with high discounts. The biggest selling point may be the Nexus operating system itself; SDN aside, it has become an important influence in the networking arena and the sister Nexus 1000v, and hardware Nexus platforms are very widely deployed.

Dell S6000

The S6000 is based upon the Dell-acquired Force10 FTOS platform and provides the desirable NSX integration features right out of the box. For large deployments, Dell's implementation of a Clos architecture, Active Fabric, uses the Z9000 data center core fabric switch as the spine. Additionally, the S6000 can be stacked with the S4810, S4820T and S5000 10GbE/40GbE switches. Perhaps the most compelling reason to buy the S6000 is Dell itself; having a one-stop shop for servers and networking gear has proven to be an attractive proposition to IT and procurement managers alike. This switch is likely to be the most heavily discounted: It will end up in the commoditized end of the market and Dell will be very successful for it.

Extreme X770

Extreme Networks Inc.'s Extreme Summit X770 sits at the top of the SummitStack range; as its name suggests, it can be joined in a virtual chassis of up to eight nodes. Helpfully, the nodes can be a mixture of older SummitStack devices; this allows an easy method to inject performance into an existing ToR deployment. While virtual chassis (the ability to join multiple nodes to create a single distributed switch) is by no means unique, it is very desirable and will reassure existing Extreme customers that their investment is protected. The X770 also permits the same flexible deployment options, allowing the administrator to choose the best balance of port utilization versus uplink capacity.

HP FlexFabric 5390AF

Beyond the initial announcement of the new switch in August 2013, as of January 2014 no further details have been released regarding the switch. Nobody ever said that integrating merchant silicon into a vendor's existing product range was easy, yet given that engineering samples have been circulating for some time, one would expect that messaging should be complete and in the public domain. Unless there is flurry of announcements and ship dates in Q1, HP may well miss the Trident II window of opportunity and lose ground in the high-performance ToR/edge space.

Juniper QFX5100

The Juniper Networks Inc. QFabric family was initially focused as a large-scale optical fabric platform, but with the introduction of the QFX5100, high-end features and scalability are now in the reach of customers with less demanding requirements. The QFX5100 virtual chassis technology allows up to 20 devices to operate as a single logical unit. This is many more than the Extreme SummitStack; however, capacity comes at the cost of a little flexibility: Only specific EX and QFX switches are supported in the Juniper virtual chassis. An unusual feature of the platform is that the Junos OS sits on top of a kernel-based virtual machine hypervisor running on an embedded Intel Sandybridge x86 platform. This is primarily for availability reasons, but a limited amount of CPU, disk and RAM resources are available to the end user. These virtual resources could provide embedded application delivery controllers or traffic monitoring tasks. For small deployments, this may be an unnecessary luxury, but if you consider environments with dozens or even hundreds of ToR switches, the "free" distributed processing could be used very creatively.

Differentiation not easy with Trident as foundation

Having looked at the specifications of all the major vendors using or planning to use the Trident II chipset, it is clear that the differentiation in the high-density switch marketplace is going to become very difficult. All of the products mentioned offer the same or very similar high performance, relatively low cost and provide a stepping stone to a full-blown SDN if required. Some boast exotic features, claim better reliability or are likely to have big discounts available. For many enterprises, the ultimate selection criteria will be limited to vendors with whom they already have a relationship. Therefore, selecting a switch technology will be less governed by speeds and feeds and more by less tangible factors, such as the quality of support and the effectiveness of local value-add partners.

That said, even if preferred vendor status quo is unlikely to change, what is shifting is the influence a single chip manufacturer can have on an entire industry. Features and protocols baked into silicon by Broadcom will pop up in network products from at least six major vendors. Support for standards relating to SDN (such as the virtual extensible local area network) suffers from the chicken and egg paradox; no one uses them because there is no hardware support; and there is no hardware support because no one uses them. Yet, as the influence to determine protocols migrates to silicon, the hardware becomes commoditized and ubiquitous. As a result, advancements like SDN become self-fulfilling prophecies as architectural and operational barriers start to fall away.

About the author:
Glen Kemp is an enterprise solutions architect for a U.K.-based managed services provider. He designs and deploys network and application security tools, including access control, remote access, firewalls and other "keep the bad guys out" technologies. He is also an experienced professional services consultant: delivering elephants and not hunting unicorns. His blogs can be found at sslboy.net and at the Packet Pushers Podcast. Follow him on Twitter @ssl_boy.

Next Steps

Cumulus touts open source

A discussion with Broadcom's CIO

ACI and Cisco's approach

This was last published in January 2014

Dig Deeper on Network Infrastructure