This major North American service provider was able to gain very high control of its network during a critical period of huge traffic growth, and to maintain high reliability along with huge cost savings using the Cisco MATE portfolio.
Overview
Cisco worked with a major North American ISP that was facing some critical network upgrades to accommodate rapid growth in both subscribers and traffic. The customer turned to Cisco to provide an integrated network planning and design system to enable the company to anticipate the impacts of such growth.
Using MATE Design and MATE Collector, the customer created a model of its network and simulated various traffic growth scenarios, their impact on the network, and the capacity planning that was necessary to accommodate that growth.
Early in the analysis, in estimating short-term growth, the ISP found that it would soon have a total of 25 links in its network that would exceed its baseline performance standard of 50 percent utilization on any link. In fact, the ISP's Washington-to-Philadelphia link would become totally congested.
The sample network model shown in Figure 1 indicates that result. Link traffic of 50 percent or more is colored yellow, and 100 percent or greater is colored red.
Figure 1. Short-Term Traffic Growth Projections
Note: Twenty-five links have larger than 50 percent utilization.
Although under normal operation up to 80 percent utilization of a link is quite functional, the ISP's capacity planning policy was to upgrade any link that exceeded 50 percent utilization to provide margin for potential failures and consequent traffic re-routing. That policy had served the company well up until then, but the expected rapid growth would make the policy prohibitively expensive to maintain. Under this policy, 25 links would have to be upgraded to meet the short-term traffic growth demands.
Finding the Primary Cause of Congestion
The customer decided to examine the impact of network failures in more detail, to see if the upgrade-at-50 percent policy was too aggressive. With Cisco MATE Design, the customer modeled various failures on the network and examined the effects on routing and the traffic demands on circuit capacities.
For example, Cisco MATE Design showed the ISP that a failure of the JNC<-->m3 link would cause only three circuits to become congested (Figure 2). Further analysis of other possible failures showed similar results, indicating that perhaps the previous policy was indeed too aggressive and too costly.
Figure 2. Impact of Failure of JNC<-->m3
Note: The red square around the DET site indicates intra-site link congestion.
Worst-Case Analysis
Rather than examining all possible failures one at a time, the customer chose to use the Simulation Analysis function of MATE Design to examine all possible failures automatically. The company chose to consider all single-circuit failures and have MATE Design tally the worst-case link utilization for every link under all such failures. (Figure 3).
Figure 3. MATE Design Simulation Analysis; Single-Circuit Failures
The effect of all possible single-circuit failures was calculated automatically, and displayed to show "worst-case" utilization (Figure 4). This analysis showed that under all possible single-circuit failures only 8 circuits were vulnerable to congestion, not 25 as originally thought.
Figure 4. Worst-Case Failure Analysis
Note: Only eight circuits subject to congestion under failure.
A new policy of upgrading only those circuits vulnerable to any possible single-circuit failure reduced the cost of maintaining a robust network from 25 upgrades to only 8 upgrades.
Determining Root Cause
A further analysis of the MATE Design results led to yet more savings. The customer examined the MATE quantitative tables (Figure 5) and found that most of the congested links became congested when the CHI<->DET link failed. So, rather than upgrading all the congested links, the customer decided to duplicate the CHI<->DET link and eliminate it altogether as a potential single-circuit failure.
Figure 5. MATE Design Results for Simulation Analysis
Note: Column "WC Failure" shows cause of "WC Util Sim" congestion values (sorted, max first).
A re-run of the Simulation Analysis showed that, with the added CHI<->DET circuit, only three other circuits were vulnerable to congestion under any possible single-circuit failure in the network (Figure 6).
Figure 6. Worst Case Analysis With Extra CHI<-->DET Link
Note: Only three circuits now need to be upgraded.
Upgrading the network for full reliability under any single-circuit failure then required only the upgrade of 3 existing circuits and adding a new one, instead of the originally planned 25 circuits.
Conclusion
At the current price point of 10 Gbps links on the backbone, the customer estimated a savings of over US$2.5M in the first year as a result of using Cisco MATE Design and Collector. The ROI for the MATE portfolio was immediate.