Recently I worked on a small data center design of about thirty racks that involved a fair amount of Cisco network kit. It wasn't a complex network, didn't have any special performance, latency or reliability requirements, and didn't use converged networking. It took about a day to lay out the basic network requirements in ports, locations and rough cabling outline. The next step was to select the Cisco switch models for the data center...
network core. I prepared the following list of six Cisco data center switch options only to quickly figure out that going low-end was probably the best bet:
1. Cisco 3750 switch stack
2. Catalyst 4500 chassis switch
3. Catalyst 6500 chassis switch
4. Nexus 5000/2000 FEX switching
5. Nexus 7000/2000 FEX switching
6. Nexus 7000 switching
Cisco strongly promotes the last four options, but the first two options are cheaper and go far. After all, a stack of 3750 switches will have the port density and routing to support a lot of servers. What's more, the Catalyst 4500 is a very good choice for data center networking now that it supports a cross-bar fabric and the software quality has finally reached acceptable levels for critical applications. It also has good port density. This makes the last four options much more difficult to justify.
Read more of Greg's Ethereal Mind blog
Greg's tech notes on Juniper QFabric
Greg picks apart the Data Centre Fabric symposium
If the customer is willing to consider a stackable, top-of-rack solution that has good 10 Gigabit support, then the Nexus 5000 at the network core with a Nexus 2000 as the access layer is also a good solution. However, after fifteen years of marketing, the message about chassis performance and stability (real or imagined) means that this is a difficult product to promote. Many people will not consider rackable switches for a data center core.
Why Catalyst 6500 shouldn't be your data center network option
The Catalyst 6500 is a product at the end of its strategic life. The recent Supervisor 2T upgrades represent the sunset phase of this chassis lifecycle, and Cisco appears to be closing down the development of the Service Modules, which are a key business driver for many companies. Even though Cisco has announced 10 years of support, I believe that buying a Catalyst 6500 today is like buying an obsolete product.
Nexus 7000 still has issues
The Nexus 7000 is similarly priced to the Catalyst 6500 and is a new product with a long term future and many excellent features. However, Cisco's reputation for poor software quality means that even after three years on the market, many of Cisco's most faithful customers are reluctant to adopt the new product for fear of Cisco software bugs. Another concern is the slow delivery of features for the Nexus 7000 such as MPLS, Service Modules and BGP.
Read more Fast Packet bloggers
HP's OpenFlow is a move away from the “tyranny of CLI”
Are networking vendors doing enough to prevent software bugs?
With edge software overlays, is network fabric just for raw bandwidth?
Additionally, I find significant confusion between the Nexus 7000 and the Catalyst 6500. Many customers see the Nexus 7000 as too complicated, with high-end features that are not necessary.
Too many Cisco products, too much money spent choosing one!
Ultimately, Cisco has too many data center network products that have enormous overlap. While there are differences between each of these products in terms of performance, features and even reliability, it's also true that some of the differences that Cisco promotes appear to be shallow. By filling many niches and flooding competing products into the market, Cisco can dominate all product decisions.
In the end, the time spent selecting products was costly. The meetings, research and discussions included professional services time that added 30% of the purchase price of the equipment. No wonder customers are reluctant to buy new products.