Data center-class switches: A buyer's guide
A collection of articles that takes you from defining technology needs to purchasing options
When it comes to choosing the right data center-class switch for your business, it can be difficult to decide where to start. While most data center-class switches generally perform the same tasks, several factors -- other than price -- need to be considered in order to find the right approach for your needs.
Data center-class switches come in all shapes and sizes in terms of port types and port capacities. Sizes range from small, 1-rack unit switches, to massive blade chassis switches that fill multiple equipment racks. It's up to you to figure out the number of Ethernet copper/fiber and Fibre Channel ports you need currently, as well as into the future.
Each vendor's hardware scales differently, and your physical port needs may be a better fit using one vendor's hardware over another. In addition to port numbers, you must determine whether you need 10 Gbps, 40 Gbps, or perhaps even 100 Gbps fiber interconnects. Your data center may also require support for redundant links and port aggregation between switches and blade servers. Once you factor all of this in, you'll realize your physical port count requirements multiply rapidly.
Another aspect to consider is whether you plan to use a top-of-rack, end-of-row configuration, or a fully centralized switch architecture. Much of this depends on the size of the data center and how the cables are configured. If cabling for an end-of-row configuration is already in place, it usually makes sense to reuse the existing wiring and deploy the larger capacity end-of-row switches.
If this is a new data center or if new cabling will be used, you may want to consider top-of-rack options, which offer more flexibility in terms of physical port coverage across a data center.
If you already manage physically separate local area networks and storage area networks and you are hoping to converge the two, you need to identify the vendors and switch models that offer LAN and SAN convergence hardware and software.
It's important to take a look at your current SAN investment to see what hardware will be compatible with data center-class switches from a SAN and Fibre Channel over Ethernet (FCoE) perspective. If you are unable to run a fully converged FCoE network on the SAN side, you must consider how many legacy Fibre Channel ports will be required.
Some data center-class switches no longer offer native Fibre Channel, so it's important that this be identified early on so you can focus on devices that support it. Additionally, you need to look at your physical cabling plant within your data center to gauge the true cost of converging two networks into one.
This may be a reason to consider distributing your data center-class switches as opposed to centralizing them in one physical location. Recabling inside a data center is expensive and extra steps should be taken to reuse existing cabling whenever possible.
More and more data flows today are east-west; that is, they remain within the walls of the data center with traffic flowing between virtual and physical servers. The vast majority of this data is between application servers and their back-end databases and network-attached storage repositories.
Because of this, the data center-class switch has a big role in moving all this east-west traffic with little latency and delay. This is why it's critical to get an accurate estimate of current and future data flow mappings and their throughput requirements. Switch data throughput capacity can vary dramatically, so it's critical to right size a switch with sufficient backplane throughput capacity to avoid bottlenecks.
Power and cooling requirements
The largest recurring cost in a data center is your electricity bill. The power required to run all network and server equipment creates an enormous output of heat, and the air conditioning required to eliminate high temperatures draws a lot of power on its own.
Reducing data center power draw is not simply an environmental question; it can save the company a substantial amount of money. The amount of energy used by data center-class switches varies from one vendor to the next. Models that emphasize low-consumption hardware components will be far more efficient and less costly to operate year over year.
Hypervisor virtual switch integration
Many data center-class switches simply rely on the virtual switch capabilities built into hypervisor software offered by vendors like VMware and Microsoft.
But other network hardware and software vendors run their own proprietary virtual switching software that fully integrates with the physical data center-class switch. The advantage of ensuring your physical switches are integrated with your hypervisor software comes when you need to move a virtual machine from one blade server to another. When a move like this occurs, the physical switch port and security policies applied to the physical data center-class switch port interface automatically transfer with it. The key factor to consider is the frequency with which VMs will be moved. If moves are frequent, the automated and integrated switching between the physical and virtual switch worlds may be an important feature.
There is something to be said about maintaining a network using end-to-end hardware from a single vendor. Unified design and configuration guides, as well as ease of troubleshooting, are distinct advantages when going with a single-vendor approach.
Additionally, some proprietary data center-class switch features can only be leveraged when upstream switches are from the same vendor. This is especially true with advanced data center redundancy features.
You also have to start thinking about software-defined networks and how your current investments can be leveraged if and when SDN takes hold. Unless you used a truly open SDN platform, a single-vendor design is going to be a wise choice in terms of future-proofing your investment.
On the flip side, going with a single vendor's end-to-end approach can have drawbacks. Vendor lock-in is the most obvious disadvantage if proprietary features are used in production. If those proprietary features are never ratified into open standards, you may be stuck operating a closed system and have few upgrade options without a complete architecture redesign.
Evaluate and consider a vendor's ability to support your data center-class switches in terms of hardware replacement, software upgrades and configuration, and troubleshooting assistance.
Data center-class switch vendors with smaller market share may have a more difficult time meeting strict hardware replacement timeframes. On the other hand, they may offer a more personal level of support service.
All support agreements have benefits and drawbacks, and most vendors excel at one or two areas of support but lack in others. The key is to identify what vendor support categories you will most likely leverage and find vendors that shine in those areas of support.
Weighing the criteria to make your decision
You need to consider a great deal when purchasing data center-class switches. It's important to note that no single vendor or switch is perfect for every situation. The key is to identify the right questions in order to narrow the field based on your specific needs. Then based on this information, you can evaluate vendors based on the quality of support they provide and their paths toward future-proofing your investment.
These criteria are meant to guide you toward your goal of finding the ideal data center-class switch for your organization.
Check out the 14 questions you should ask SDN vendors
Time to choose: Cisco vs. HP switches
Check out this guide to help deal with increasing data center-class switching automation
Will top-of-rack switching be the demise of copper cabling?