Assess Wasted Network Capacity

Instructions to detect, quantify, and pinpoint bandwidth wasted due to retransmissions

Retransmissions, traffic that must traverse the network multiple times for successful arrival, should optimally be as close to zero as possible. But when router queues deepen and cause dropped packets, retransmissions spike. When latency increases the frequency of time-outs, retransmissions spike. When a busy IP network behaves precisely as designed under heavy loads, retransmissions spike. Buying bandwidth to support a high rate of retransmissions is a costly, wasteful situation.

These instructions show you how to play detective with your network's retransmissions. However, you should remember that well-managed bandwidth prevents network inefficiency. PacketWise's control features, including TCP rate control, help avoid router queues, dropped packets, and retransmissions so that you will have less inefficiency to investigate.

Steps:

1. Examine the Network Efficiency graphs for the Inbound and Outbound traffic classes by displaying the Network Performance Summary.
   
   
   
   PacketWise's Network Efficiency graph shows the percentage of your bandwidth wasted by retransmissions. PacketWise knows the rates of throughput and retransmissions for any traffic class, and calculates percentages of wasted bandwidth. With the Network Efficiency graph, you can track the current retransmission rate or explore its history. You can focus on the traffic that is of interest: your link as a whole, an application, a protocol, a subnet, a user, a server, or a web destination.
   
   Suppose you notice your Outbound class' network efficiency is fairly steady at about 90 percent.
   

2. Try a few recent time periods with the report's interval control settings to see if the inefficiency figure is a steady one.
   
   For example, you might note that Outbound's efficiency figure was at 97 percent or higher until the past couple days.

3. Select several critical traffic classes and display each Network Efficiency graph.
   Consult examples of applications' efficiency graphs and their interpretations.
   
   Suppose one of your efficiency graphs indicates that all classes but one show less than 4 percent retransmissions. One, the Outbound/Oracle class, shows 80 percent network efficiency or, alternatively, 20 percent retransmissions. Definitely too high.

4. At this point, some of PacketWise's other analysis features would be appropriate. For example, you could examine Oracle's average and peak usage rates on the Monitor screen, and analyze server and network response times for Oracle by examining the Transaction Delay graph. (Note: The Transaction Delay graph is not available on ISP models.)
   
   
   

For example, you might note that Oracle's network delay has been consistent and within limits. But its server delay started climbing a couple days ago.

5. In addition to examining the Network Efficiency graph, you could display the Bytes Transmitted graph that plots the number of bytes transmitted and retransmitted together, on one graph. The result can be quite enlightening. If you have no spikes in transmissions without a parallel spike in retransmissions, then high usage is not the source of congestion problems, but error conditions (such as down or overloaded servers) are.
   
   
   

6. Choose your next investigative technique based on your current findings.
   
   In the example, because the server delay was high, a good choice would be to enable Worst Clients and Worst Servers on the Oracle traffic class. (Note: The worst hosts feature is not available on ISP models.) Another option might be to divide the Outbound/Oracle class into several child classes based on client subnet, server IP address, or any other criteria that you suspect might distinguish traffic with a higher percentage retransmissions. For example, if you created child classes with different server IP addresses, you would discover which servers had higher delays (the same function as PacketWise's Worst Servers) and higher retransmissions. For additional ideas, peruse the list of PacketWise graphs.
   

The adaptive response feature's ability to automatically detect problems can be helpful in this context. You can keep tabs on your network efficiency at all times without frequent investigations. See Analyze Efficiency with Traffic Performance Agents.