Identify Performance Saboteurs

Instructions to identify traffic and applications that undermine others' performance

Some applications and protocols tend to burst — rapidly expand to consume great quantities of bandwidth, undermining the performance of applications that are, perhaps, more urgent. This behavior doesn't necessarily imply unruly users or bad applications. The applications might be critical but tend to consume greater than an appropriate share of bandwidth (Microsoft Exchange, for example). Or they might be important, but not interactive and not particularly urgent (FTP, for example). Or they might be recreational applications (music downloads, for example).

   Steps:

1. PacketWise's adaptive response feature automatically monitors for conditions of interest, detects potential problems, and optionally notifies somebody and/or takes corrective actions if a problem is detected. For this context, you can use the adaptive response feature to have PacketWise automatically detect new applications that use an unacceptable amount of bandwidth. You get to define "unacceptable" any way you like.

   If you want to be notified automatically of new bandwidth-greedy applications without needing to check manually, see Monitor Bandwidth of New Applications.
   
   

2. Display the Network Performance Summary and examine the Top Ten pie chart. It shows the top ten traffic classes — those that generated the most traffic over time. Play with the time period to get a feel for which traffic classes are consistently top bandwidth consumers.
   

3. Examine the usage statistics for all traffic classes shown on the Monitor Traffic page. Search for classes whose peak approaches your link capacity.
   
   
   
   In this example, note that MPEG-Audio is taking 3.1 Mbps. Unless there's a business need for it to take such a large portion of the link, it's a good candidate for PacketWise's control features.
   
   Although the Top Ten pie chart provides insight about which applications use the most total bandwidth, it doesn't highlight applications that tend to spike. An application that uses a small amount of bandwidth consistently over time can end up in the Top Ten, even though it causes few performance problems for other applications. However, an application that sporadically spikes to consume the whole link for 5 or 10 minutes imposes a much more disruptive problem, even though it might not make the Top Ten.
   

4. Display a Network Efficiency graph and a Peak Utilization graph for the link as a whole. Then display the same graphs for each traffic class you identified in the previous step that had spikes near capacity.
   
   

   
   
   

5. Compare the class graphs to the link graphs. Did class spikes impact the link? Did they bring down efficiency? Which classes caused the problems?

6. Repeat the previous step for traffic classes in your Top Ten pie chart.
   
   If you have enough information about users who use the most bandwidth resources, you can stop here.
   

7. Track the hosts that contribute most to any traffic class by setting PacketWise's Top Talkers and/or Top Listeners features on the traffic class of interest. Top Talkers displays the top 20 generators of a class' traffic, and Top Listeners displays the top ten recipients of a class' traffic. Once set, both of them have an associated icon that appears in the Monitor Traffic window ( for Top Listeners and for Top Talkers).
   
   A sample Top Listeners list for an HTTP class appears below.
   

Examples of how to use Top Talkers (TT) and Top Listeners (TL):

• Use TT on HTTP to reveal the most popular websites.
• Use TL on a traffic class for the three most popular sports sites to reveal your top sports fans.
• Use TL on the current music-download application's traffic class to reveal those that retrieve the most music.
• Use TT on the same class to reveal servers most tapped as music sources.
• Use TT on an Oracle traffic class to reveal the most active Oracle servers.

Remember that PacketWise's monitoring and analysis features are only the first half of managing performance. Control features can solve the problems that analysis only identifies.