The Data Center: What Is a Good Network? - Systems Management News On The Web

Sort By :DateRelevance

DEPARTMENTS

HOME

TOP STORIES

ON THE WEB

BZ MEDIA

Printable version

The Data Center: What Is a Good Network?
By John Rath

September 15, 2008 —

A recent question on LinkedIn asked, “What is a good network?” Normally I would have dismissed that because it was too vague to even think about answering. The answers mainly focused on the “-ities” (as I call them): reliability, scalability and flexibility. Don’t get me wrong, those are all good qualities, but they are also generic marketing terms that can be interpreted a number of ways.

Most of the respondents on LinkedIn answered in the context of a local business or business branch that had a T1 (1.544Mb/sec) or other account with a carrier. When I think about what a “good” network is, I naturally migrate to what would be good for a data center environment.

Start with OC-3 (155Mb/sec) and work up to OC-192 (9,953Mb/second) and beyond. While the size of the pipe may differ for office vs. data center environments, don’t underestimate the office connection. As it relates to my previous cloud computing columns, if your applications, platform or entire infrastructure is in the cloud, it won’t do a bit of good if your employees can’t reach it because the office network is down. Complete a thorough analysis of how office connectivity should be architected based on the needs of the business.

The LinkedIn answers also confirmed my initial thought that the real answer to the question depends on what the business needs or wants to accomplish with its network. A few columns back, I wrote about the massive networks Microsoft and Google have in place to connect their equally sized data centers. Unless you have a few billion dollars lying around, I think one of the top ingredients of a good network is knowledge of and working relationships with carriers.

There are some excellent opportunities to work with dark fiber providers and connect data centers that way. The need for dark fiber vs. a connection from various carriers again goes back to what you will use the network for and what kind of uptime, control, price, security and scalability you need. Large companies such as Ford Motor, Bank of America, Bausch & Lomb and Gannett have their own fiber networks, an approach that fits their needs better.

When dealing with a carrier or agent to help build a dark or lit fiber network, the type of network you choose will depend on how you will need it to perform. Some companies may want a Synchronous Optical Network (SONET) deployment or a dense wavelength-division multiplexing (DWDM) network in order to accommodate multiple circuit types and plenty of room for growth.

SONET seems to be on its way out in favor of DWDM and other newer technologies with greater capacity. AT&T is reported to have deployed 50,000 wavelength miles of OC-768 (40 Gb/sec) in its Internet/Multiprotocol Label Switching (MPLS) backbone network. Wavelength-division multiplexing offers service in OCx varieties as well as 1Gb, 2.5Gb and 10Gb.

A Cisco presentation that included the slide “Why DWDM?” made me laugh. Two of the three bullet-point answers obviously addressed different audiences. One was “subject to relaxed dispersion and nonlinearity tolerances”; the other was “capable of graceful capacity growth.”

A few years back, new functionality for long-haul DWDM equipment appeared: reconfigurable optical add-drop multiplexing (ROADM). ROADM aids in the planning of bandwidth by letting the network operator perform configurations as needed, instead of only at the initial implementation. It also allows for remote configuration and reconfiguration.

A third optical transport protocol is MPLS. Often thought of as a Layer 2.5 protocol in the Open Systems Interconnection model, MPLS provides an efficient means for managing applications and moving information between locations. Its advantage is that it prefixes packets with an MPLS header, for class-of-service and network traffic prioritization.

MPLS is replacing Asynchronous Transfer Mode and frame relay, and it is a strong contender now and will be for the future. Foundry Networks (recently acquired by Brocade) announced that its new software and firmware for NetIron routers will let service providers integrate Layer 2 protocols with MPLS services, with the intent of scaling Ethernet services across Layer 2 metro and MPLS networks.

John Rath is an independent consultant and blogger at Datacenterlinks.com. He can be reached at johnsr4@gmail.com.

Related Search Term(s): cables, cloud computing, data centers, networking

Share this link: http://www.sysmannews.com/link/32851

EMAIL THIS ARTICLE

SEND FEEDBACK

MORE COLUMNS