The conversion from one set of servers to another, more capable, set of servers was supposed to be a slam dunk....
In fact, the process was so routine that there was no plan to extend a service outage past a maximum of three hours. However, when the data center technicians began the process, they discovered that none of the cables connecting the old servers were labeled, and the old servers had not been mounted properly in the equipment racks. Once the cables were disconnected, with difficulty, some of the cables were found to have broken plugs and, in any case, it was hard to figure out where they needed to be plugged into the new boxes.
To make a long story short: An outage that was supposed to take only about 20 minutes ended up taking five hours. Furthermore, in the panic to get service restored, temporary cables were strung between equipment racks, leaving a rat's nest of wiring on the floor and necessitating additional service outages later to clean up the mess.
To make a long story short: An outage that was supposed to take only about 20 minutes ended up taking five hours.
This is a true story, and the sad part is that the servers in question were mission-critical, and their outage caused a considerable amount of damage to the company's revenue generating activities.
Data center cabling a critical issue
Cabling, it turns out, is a critical part of the data center's internal network. It is absolutely necessary to not only know where each cable goes, but to be able to locate specific cabling when equipment is reconfigured or changed. This means that every data center should have a cabling inventory and a cabling inventory management system. Data center managers must be able to obtain specific information on each and every cable within seconds, including the type and vendor of the cable, termination points, cable-race locations and date of installation.
The good news is that it isn't necessary to build such a system from scratch using spreadsheets (although if that is the most expedient option to get started, it is better than nothing). There are cable management systems that range from the very basic to the very complex. Some also tie into network management systems to provide immediate access to inventory information if a networked device is indicated in a fault situation. Many are part of larger data center infrastructure management (DCIM) applications. IBM and Rackwise, to mention only two, have rather extensive offerings in this space. There are also adequate open source solutions, such as openDCIM, which can be used to get started.
Moving ahead with cable management strategy
First and foremost, the cables themselves must be labeled. At a minimum, each cable termination should be tagged with the cable's unique identification number and some indication of its purpose. Remember, when there is a fault, it is critical to be able to find a specific cable quickly and unambiguously.
Data centers should also have a specific cabling architecture: one that allocates specific cable races to specific functions. For example, interconnection cables between routers, switch and firewall devices should be segregated from cables devoted to external communications. (The author is reminded of a data center where a single twisted pair was hung from the ceiling to support external customer access to an automated service ticker processor. Every time the fans came on, the twisted pair would swing in the breeze and customers' access would drop.) Additionally, a fault-tolerant architecture should be sensitive to such things as cross-talk and noise, as well as heat dissipation.
Cabling tips and tricks
Everything you wanted to know about cabling
Lower your energy bills
Here's how to cut alien cross-talk
Is this a lot of work? Yes. Is it necessary? Do the math: A custom-built cable, depending on length, can be several hundred dollars. A cable failure, however, can lead to thousands of dollars of lost revenue -- or at the very least lost productivity -- as technicians struggle to find the cable and replace it. Think of a good cabling plan as an insurance policy that can literally pay for itself the first time a failure occurs.
So, how does one tell if a data center is at risk? Take a tour of the data center and examine the cabling. Are cables clearly marked and color-coded? Are cable races neat and cables segregated by function? Are cables off the floor? If the answer to any of these questions is no, then the data center has a problem.
Inventory and tag, then name
Where to start? The best place is to inventory and tag every cable. If a cabling nomenclature is needed, the Web is full of practical guides to develop labeling conventions. Once everything has been located and labeled, it is time to input the cabling into an inventory system -- even a spreadsheet, although that is not optimal. Ultimately, one will want a more capable system that allows for additions, moves and changes without major input overhead.
Finally, once the inventory is operational and the cabling has been identified, there should be an ongoing effort to test and reroute cables to match function and type. The process of improvement should be continuous and will result in a significantly improved cabling infrastructure in a surprisingly short time frame.
The choice is simple: Either take pains to clean up data center cabling, or pay for inefficient or poorly marked cabling in the future. Generally speaking, the cost will be higher if you wait, and that delay could be very damaging to your business.