Up to Speed

Rethink data center requirements

Midsize organizations should emulate big company solutions on a smaller scale.

by Stephen Garrison

IT managers in midsize companies may not know it, but their problems are similar to those in large companies, particularly concerning their data centers. One common problem is the need to modify systems to comply with regulations such as Sarbanes Oxley (SOX). Like their big company counterparts, IT managers in midsize companies are tasked with ensuring financial information is consistent and accurate.

IT managers are working to reduce data duplication, simplify auditing and achieve compliance. They are also consolidating data by moving servers out of business units into the data center. Many of these servers have been underutilized. As a result, IT is turning to virtualization software to squeeze all the processing power possible out of their banks of CPUs.

Data consolidation aids in regulatory compliance, while virtualization optimizes investments in server hardware and software. These techniques, however, present their own problems. Where a server room may have had 50 to 100 servers, now it has 500 to 1,000. Rather than highly distributed data and servers, companies have centralized data and servers, creating new traffic patterns and server loads that can strain existing data center networks.

A data center switch should be capable of operating under heavy loads, with no impact on traffic and without failing.

As data centers grow, network performance and reliability become crucial. Having placed all their eggs in one basket, so to speak, IT managers should ensure that congested network links do not keep users from doing their jobs - or that a network disruption does not cripple the business entirely.

As IT managers in mid-tier companies re-engineer their data centers, they need to re-evaluate their network infrastructure. A clear understanding of the requirements for a data center switch can help IT managers determine if their existing network gear is up to the task and provide guidelines for evaluating alternatives.

As IT managers in midsize enterprises expand their data centers, their network needs to echo those of large companies, just on a smaller scale. Consequently, the characteristics they should look for in a data center switch are similar.

The ideal data center switch is a high-performance, high-density, scalable and highly resilient device. Within the data center, the network should be so reliable as to be invisible to users. A data center switch should be capable of operating under heavy loads, with no impact on traffic and without failing.

Software stability is as critical as hardware redundancy in ensuring always-on operation. Software should have some degree of modularity and memory protection to ensure adjacent processes do not corrupt each other. In addition, line cards and other key components should be hot swappable.

Once data and servers are consolidated, the number of users operating against a single data set or server can increase dramatically. Data center switches should be high performance, providing line-rate throughput on all ports. The switch architecture should be non-blocking and deliver low-latency switching - even under a full load - regardless of packet size.

A data center switch should have a small form factor, yet provide hundreds of Gigabit Ethernet (GbE) interfaces for server connection. For scalability, a data center switch should be designed with excess capacity, allowing for substantial growth without the need for a forklift upgrade. It should also have modular 10-GbE interfaces, allowing IT to add high-speed interfaces, as needed, for trunking or core uplinks. A scalable, high-density switch allows for a simpler network design with fewer devices to manage; requires a smaller overall footprint; uses less power and cooling; and accommodates growth.

Stephen Garrison is vice president of marketing for Force10 Networks, San Jose, Calif. Garrison received a bachelor's degree in ceramic science from Alfred University and a master's in materials science and engineering from MIT. He sits on the board of directors for the Ethernet Alliance.

 For more information (click here)