Adopting A Business-Oriented Approach To Bandwidth Management.
The first challenge lies in understanding all of the factors that contribute to a user's perception of the service levels they are receiving. With the exponential growth of content-rich applications and Internet computing demands, users often perceive that there is insufficient bandwidth available to completely satisfy the needs of the business. Users measure bandwidth in terms of the response times they experience. While related, however, response times are not an accurate measurement of bandwidth and capacity. Many networks where users experience slow response times actually have more than enough total bandwidth; the bottleneck may well be caused by another component such as a server, an application, storage device or the user's own PC.
The first stage of bandwidth management, then, is to accurately measure the service levels being provided--from the users' standpoint. The enterprise network management system is effective in collecting performance data for any given device, but in order to make any informed decision regarding the adequacy of the service levels delivered, this raw data must be rolled up and interpreted in terms of business metrics as experienced by each user. While calculating service levels manually is very time-consuming, there is a new generation of software tools now available to help the IT executive assess and track actual service levels according to meaningful business metrics--and to identify whether bottlenecks are equipment, application, or bandwidth-related.
How Much Bandwidth Is Enough?
The goal of managing network capacity is to have the right amount of bandwidth in the right place at the right time for the right set of users and applications. Network components have two common characteristics: a finite transmission capacity and a measurable utilization or load. Bandwidth management requires that the load/capacity relationship of key facilities be understood, in order to ensure bandwidth sufficient to keep a business or a service functioning properly and profitably. Each environment is different. To ensure enough bandwidth, bottlenecks must be identified and dealt with on a case-by-case basis, based on business priorities. In each organization, the service levels for application performance will vary based on the delivery priority of particular information applications--and this delivery priority must meet the objectives of the business model.
Six steps should be taken to measure and plan bandwidth capacity (Fig 1).
* Identify current bottlenecks
* Provide increased capacity to meet current demands
* Project the historical demand levels into the future requirement
* Put plans in place to supplement current capacity before future bottlenecks occur
* Eliminate or downsize any costly, underutilized facilities to reduce waste
* Balance demand across time in order to more efficiently use available capacity
Obtaining the information necessary for performing these steps successfully can be a challenge. Here are some of the problems and common mistakes, as well as the new approaches available to address bandwidth management issues.
Traditional bandwidth management tools are not designed for continuous capacity assurance, nor are they integrated into the overall management infrastructure. They are typically stand-alone point products that collect raw statistical information. Raw data, in and of itself, is not useful for business analysis. For example, a graph that displays InOctet and OutOctet values over time doesn't really tell you whether there is adequate bandwidth to meet the needs of the business. Simply having that information does not illustrate how well the route, from which the data was collected, is functioning in terms of its primary purpose--enabling information flow throughout the enterprise. What's needed are appropriate business metrics for really understanding capacity issues.
Bi-Directional Data Links And Circuits
A data link provides a connection between two or more computers. A circuit is a virtualized data link. It is important to note that most communication systems are highly layered and what appears to be a link to one person may be considered a circuit by another. From a capacity planning point of view, all such connections are considered links.
A link contains one or two facilities. A half-duplex or shared link like Ethernet has a single capacity, or bandwidth, that is shared among all attached devices and is treated as a single facility. A full-duplex link, on the other hand, contains two independent facilities, each of which has its own capacity and load. Capacity planning must be performed or a facility basis; in no case should the capacities or loads be aggregated among the link's facilities. It is entirely possible for one facility of a link to be saturated while the other is underutilized. Aggregating the two would produce a facility with a mid-level load that does not correspond to anything in the real world and could obscure potential problems or bottlenecks.
The nominal capacity of a full-duplex link is really the capacity of each facility. For example, 20Mbps can be transferred over a full-duplex Ethernet (10Mbps) link. If this is not taken into consideration, it can seriously distort any capacity planning activities.
How Busy Is It?
To begin meaningful capacity and bandwidth planning, it is necessary to assess how busy a facility is, but how is "busy" measured? What we are really asking is: "How close is the facility to saturation?" Saturation occurs when a facility reaches its full operating-capacity--the maximum level at which it will perform normally under continuous operation. For example, an aircraft may have a maximum speed, but will also have a cruising speed that is somewhat below its maximum. Similarly, nearly all technologies have an operating capacity somewhat below their maximum capacity.
Human activities do not automatically spread out across the hours in a day. For the most part, the work needs to be done when the people are ready to do it. For that reason, it's not the average load that is important, but the peak load during the busiest period for the facility. Facilities are nearly always sized based on peak load characteristics. Sizing based on anything less results in unhappy users or system breakdowns.
The load of a typical real-world facility is extremely variable. Consider the case of a T-1 link. If several people sampled its load at different intervals, each would see a different peak-load metric for the link. Each of these different answers would be correct, but none would be helpful. We need to reframe the question "what is the peak load?" to include an appropriate time factor. Peak-hour load tends to be the most practical metric for assessing bandwidth availability and managing capacity because most activity affecting bandwidth typically takes place during normal business hours.
After measuring the activity levels of a facility for a period of time, a pattern can be detected in the peak-hour load trending. The peak-hour load is much easier to project than the raw load because it consolidates all the samples for a day into a single metric and smoothes many of the perturbations. Peak-hour load should not be used directly for projections, however, because of business-cycle considerations such as weekends, month-end-closings, and so forth. High and low samples should consequently be removed from consideration and projection should be made based on the midrange of peak-hour load samples. By using this method, peak-hour load is projected over various periods to determine a projected saturation date--which should, then, be compared with the service-cycle time for the given facility. How long does it take to perform an upgrade? Remedial action can then be initiated for any facilities whose projected saturation is sooner than the service-cycle time.
Once the peak-hour load is available, it is easy to identify the facilities that are underutilized. Any facility with a long-term peak-hour load below a given downgrade threshold should be considered a candidate for downsizing or removal. The downgrade threshold can be set to indicate the level at which the next lower increment of capacity becomes more economical or at which the facility should be eliminated or redeployed.
The final step in the band-width-planning process is to look for ways in which the demands can be shifted in time or reduced through organization or procedural change. Facilities with extremely high peak-hour load relative to average load are good candidates for demand balancing. Comparing the average load of a facility with its peak-hour load yields a metric that can be referred to as the "balance factor."
Demand balancing can be achieved in many ways, most of which affect the work-process and procedures within the organization (Fig 2). Examples include rescheduling batch jobs, deferring data entry, staggering work shifts, or placing usage restrictions on individuals during peak usage times. Once the demand on a facility is balanced, the peak-hour load should be significantly reduced, allowing for the possibility of significantly reduced costs by eliminating, downsizing, or deferring the upgrade of the facility.
Determining the balance factor is the most difficult of the capacity-planning activities and requires information on the daily-usage pattern of the facility, the business activities of the organization, and the ability to institute organizational change. It is, however, an important part of the process. During this step, a highly effective integrator can produce significant savings for the organization while increasing infrastructure responsiveness.
Peter Cruz is the director of product planning and management at Opticom, Inc. (Andover, MA).
Six ways to plan bandwidth capacity Managing Bandwidth: Steps to Take 6 Balance demand across time in order to more efficiently use available capacity 5 Eliminate of downsize any costly, underutilized facilities to reduce waste 4 Put plans in place to supplement current capacity before future bottlenecks occur 3 Project the historical demand levels into the future requirement 2 Provide increased capacity to meet current demands 1 Identify current bottlenecks
|Printer friendly Cite/link Email Feedback|
|Title Annotation:||Technology Information|
|Publication:||Computer Technology Review|
|Date:||Mar 1, 2000|
|Previous Article:||1394 Grows Up.|
|Next Article:||Seagate's Viper Drive.|