Network Integration: Explained

There once was a time when network engineers made spaghetti from network cables. These were dark times. Admins kept to the shadows - their proper place, so they were told - fighting back the invisible forces of the internet. Spaghetti; their only defense for that is what tools Kalpana, the gods of old, had offered them. 

New dawn showed across the horizon. Network admins could come out from their holes, for a new light had pierced the black sky. They now demanded fealty to the new chassis for that chassis expunged spaghetti from data centers across the land. 

This was the dawn of network integration. The old might of Kalpana and Packard Bell was replaced by Cisco and HP. Today, we will tell the tale of Network Integration: Explained.

An Overview of Network Integration

In this video, Jeff Kish explains what Cisco UCS is and why blade servers have significantly impacted network integration in today’s IT world.

Network Integration: Explained

A traditional server rack holds various pieces of equipment. For example, it might rack a UPS, a couple of servers, a tape backup machine, and a storage appliance. At the top of each rack, a switch connects all of these devices to the business network. 

Everything in IT is redundant, however. Each server, tape backup machine, and storage appliance may have multiple network connections. Each device may have two network cables running from each machine to the switch at the top of the rack. If there are neighboring racks with switches, some of those redundant network connections will span to those switches, too. 

This model creates a lot of issues, however. First, running redundant network connections between racks makes, what I call, network cable soup. Any front-line IT tech that has had to trace a network connection from a port on the business floor to the switch knows how much of a nightmare this scenario is. At best, these network cables are bound together in a tight bundle. At worst, network cables are strewn about with no labels. No matter the scenario, tone generators typically pass crosstalk between lines, making tracking cables significantly harder. 

That’s not to say that network engineers have it any better. Though front-line techs might need to trace network runs through cable soup, network engineers need to track configurations for each of those switches carefully.  Each switch is administrated separately without the help of a master controller application. 

That’s the old way. We have newer, better ways to rack servers in a data center. 

Companies like Cisco have created fully integrated products.  These tend to be called blades. Of course, this is an oversimplification of the equipment in question. So, let’s discuss this further. 

In a blade system, we start with a chassis. Think of the chassis as the frame of a car. Its structure is designed from the ground up to hold the engine, gas lines, and other components together. The chassis of a server rack has spaces to mount each component, such as a server, power supplies with distribution blocks to each of those blade servers in the chassis, and something called a fabric interconnect. The fabric interconnect is where the magic happens. 

The fabric interconnect is like a fancy switch. Each blade server in the chassis connects to the fabric interconnect through internal traces in the chassis. The fabric interconnect has virtualized network adapters in it. Because of this, each blade in the chassis thinks they have its own network adapter. The network adapter lives outside of the server, however.

Each fabric interconnect operates in end-host mode. End-host mode is a fancy way for the fabric interconnects to operate like a switch but identify as a network adapter to the upstream devices in the network. 

That means each device in the blade chassis doesn’t need redundant network connections. Each blade server connects to the fabric interconnect in the chassis. The fabric interconnects then connect to the switch upstream with a single network cable (two if you want redundancy). A single 24-port switch could service more than 12 blade racks with redundant connections. Each blade rack can handle 4+ blade servers. That translates to only needing a minimum of 12 network cables (per this example) to connect more than 48+ servers.

Now that’s amazing! Each fabric interconnect can run 10Gbps or faster, too. 

Learn More About Cisco UCS With CBT Nuggets

We’ve only covered the beginning of what Cisco offers us. That’s not to mention other vendors like HP, too. Learning the Cisco UCS architecture requires more than a 1,000-word article. If you’re interested in continuing this conversation, consider taking an online CCNP course. CBT Nuggets has online training for Cisco UCS training available right now.