Hi,
Traffic monitoring will use significantly less CPU if you disable the sFlow collection aspect of it. It takes ~50x as much CPU to process sFlow data from one port as it does to poll statistics from the port.
So you could either disable sFlow monitoring across the whole topology in favor of statistics polling, or you could be selective about it and only keep sFlow enabled on what you deem critical ports while configuring all of the less important ports for statistics polling. You can do this by:
1. Selecting somewhere in the tree that has a Traffic tab, which will likely be driven by the scope over which you want to configure ports for statistics polling and/or sFlow.
2. Depending upon where you select in the tree and your intent, multi-selecting all of the ports shown in the Traffic tab or some subset upon which you wish to operate.
3. Once the desired ports are selected right-click on the "Cfg" column, choose the "Manual" submenu, and then choose "Manually enable statistics" to limit the selected ports to statistics polling only. For ports you want sFlow on choose "Manually enable sampling and statistics". If there are any ports you don't need data from at all then you can lop off some more CPU consumption with the "Manually disable sampling and statistics" choice here.
I hope this helps.
And for what it's worth, in PCM 3.x the sFlow scalability is much improved (up to 30k sFlow ports and 70k polled ports). You have the ability to collect data from more ports at the PCM server than in 2.x (3k sFlow, 7k polled), and if that's still not enough scalability you can offload some of the CPU workload to a separate "agent" system that does the raw collection and reports summarized data to the server. If you like the traffic application is might be worth considering an upgrade.
Regards,
SVB
