5 Major Drivers of OpEx in the Enterprise Network

Michael Bushong is the vice president of marketing at Plexxi. This column is part one of a two-part series looking at the costs factors and controlling costs in your networking infrastructure.

Capital costs account for nearly two-thirds of the purchasing decision for networking equipment, according to research by IDC. But over the life of the gear, the total cost of ownership (TCO) is dominated by ongoing operational costs – both administration and maintenance of the device. Many TCO models exhaustively look at all sources of expense, but it’s also important to note the key drivers behind OpEx.

When looking to evaluate your budget for equipment, consider these five major drivers of OpEx in your network infrastructure:

1. Number of Devices Under Management
The largest primary driver of cost is the total number of devices under management. Each device will drive space costs based on its size and the price-per rack-unit for a particular environment. Similarly, each device will contribute to power and cooling costs.

The total number of devices also impacts ongoing operational costs. For example, the total number of spare chassis and line cards required on-hand will scale linearly with the number of devices deployed, as will the carrying costs for these spares. These carrying costs will increase with the number of different platforms within an architecture, as each platform family requires its own spares.

Additionally, administrative overhead has a correlation with the number of devices. Each device represents an element that must be ultimately provisioned, monitored, troubleshot, audited, and secured. To some extent, these management costs can be partially offset by provisioning tools (as with DevOps-type tools), automation frameworks, and network controllers that reduce the total number of administrative touch points.

Beyond the easily quantifiable drivers, there is an overarching complexity contribution to ongoing costs. It is impossible to model, but complexity is positively correlated with the number of devices under management. As the number increases, so too will complexity – along with the costs required to manage it.

2. Number of Ports Under Management
While the number of devices is a good proxy for environmental costs and administrative overhead, it is the number of ports under management that drives cabling and provisioning costs. The most basic cost tied to ports is the physical cabling required to interconnect the ports. For architectures that utilize many fabric ports, this additional cabling provides connectivity through the fabric but does not increase the total number of servers attached to the network.

How networking gear is cabled also impacts long-term operational costs. In some architectures, the interconnect ports are taken from the same pool as the server ports. For every interconnect port, the server capacity of the switch is reduced by one. For large, distributed architectures that require non-blocking paths through a core switch layer, this can represent a significant percentage of available server facing ports. The result is a larger number of devices to meet overall server port requirements.

Beyond the cabling and power costs, the total number of ports under management servers as a reasonable proxy for provisioning, monitoring, and maintenance costs. Each port represents another entity that must be managed.

3. Number of Administrative Touch Points
While the environmental costs will grow linearly with the number of devices, the ongoing operational costs can be mitigated somewhat by reducing the number of administrative touch points in the network.

To some extent, the rise of software-defined networking is a response to the rising operational costs tied to network growth. Controller-based architectures are designed to provide central points of control through which entire networks can be managed, reducing the number of administrative touch points in the network.

By providing a single point from which all devices can be provisioned, monitored, and troubleshot, the overall effort required to do so is greatly reduced. This has the added benefit of driving down human error – the single largest source of network downtime in most networks.

A single point of administration also lends itself well to providing better network visibility. By collecting distributed data and presenting it from a single point, the network is better documented, making troubleshooting tasks shorter and more straightforward. This ultimately impacts metrics like Mean Time to Insight (the time it takes to correctly diagnose and triage new issues) and Mean Time to Repair.

4. Number of Integration Points
Capability in isolation is useless. Ultimately, data center networking gear must be integrated with surrounding infrastructure to provide any real value. That surrounding infrastructure certainly includes other networking devices, but integration extends well beyond network interoperability.

Network infrastructure must be integrated with surrounding compute, storage, and application components. Each integration requires time and money. Accordingly, the number of points at which these integrations must be executed will be a cost driver. For architectures requiring device-by-device integration, costs can be high. Those that handle integration through central points will contribute lower cost. These costs are incurred at both the time of integration as well as at any subsequent change.

Beyond the sheer number, most of these integrations require some exchange of data. If each supporting tool is responsible for harvesting information separately, the effort to integrate will be higher. To the extent that architectures can provide a common means of extracting data from the system, these costs can be lowered.

5. Number of Management Models
Beyond just the number of devices that must be managed, the number of disparate ways in which those devices are managed is also a cost driver. Where architectures are standardized around a single device type or family of devices, there is typically one management model. The single operating system environment lends itself well to developing and leveraging a single set of training materials, provisioning models and templates, standard operating procedures, and supporting processes like auditing and change management.

Costs associated with these types of tasks will tend to scale linearly with the number of management models within a data center. Companies that reduce management complexity will see reduced ongoing operational costs.

Controlling these cost drivers should be a primary objective when designing all data centers. While specific device characteristics and capabilities can mitigate costs to some extent, the most significant contributor to ongoing operational costs is the underlying data center architecture. Accordingly, data center architects should consider the long-term cost impacts of architectural designs.

In part two of this series, I will examine specific ways you can plan and architect your network to control costs. 

Industry Perspectives is a content channel at Data Center Knowledge highlighting thought leadership in the data center arena. See our guidelines and submission process for information on participating. View previously published Industry Perspectives in our Knowledge Library.