Polling processing is comprised of a specified number (default 10) of polling threads and several other threads. The polling threads operate in parallel, while the other threads do not. The overall polling rate that can be supported is limited by the single copy threads. If there are many concurrent threads running or if the total amount of CPU available is not adequate, polling may be adversely affected. A faster CPU or slower polling rate would be required.
The minimum requirement for polling is that the total single-threaded polling CPU seconds as specified in Single-threaded Polling CPU seconds does not approach the chosen polling interval. Single-threaded Polling CPU seconds can be used with the number of managed interfaces and managed ports obtained or estimated in Single-threaded Polling CPU seconds.
| Operating system |
IP Availability Manager |
IP Availability Manager and IP Performance Manager (AM-PM) |
||||
|---|---|---|---|---|---|---|
| Per interface |
Per unmanaged port |
Per managed port |
Per interface |
Per unmanaged port |
Per managed port |
|
| Linux |
0.000037 |
0.000000 |
0.000305 |
0.000148 |
0.000000 |
0.001629 |
| Solaris |
0.000222 |
0.000000 |
0.001830 |
0.000888 |
0.000000 |
0.009774 |
Actual SNMP polling is done by the polling threads; if they are not keeping up due to device latency, more may be added. “Calculate SNMP polling thread utilization” on page 149 will help you determine whether the number of polling threads needs to be adjusted.
To calculate the total amount of CPU needed per polling cycle use Total polling CPU seconds in conjunction with the number of Managed Interfaces and Managed Ports obtained or estimated.
| Operating system |
IP Availability Manager |
IP Availability Manager and IP Performance Manager (AM-PM) |
||||
|---|---|---|---|---|---|---|
| Per interface |
Per unmanaged port |
Per managed port |
Per interface |
Per unmanaged port |
Per managed port |
|
| Linux |
0.000319 |
0.000000 |
0.001618 |
0.001055 |
0.000000 |
0.011634 |
| Solaris |
0.001914 |
0.000000 |
0.009708 |
0.006330 |
0.000000 |
0.069804 |
While polling will not improve by increasing the processing power above its minimum requirements, it does have an absolute requirement for CPU in order to keep up with specified polling rates. SNMP polling is augmented by pinging all IPs. If the ping fails, the SNMP requests are suppressed for related interfaces and ports, thus preventing a sudden influx of timeouts. By default, pinging occurs every 20 seconds. SNMP polling defaults to every 4 minutes.
The amount of parallelism possible in the polling subsystem varies, but it is recommended that the total polling CPU should not exceed 100% of the CPU. This is the effective limit on how much topology a server can support. Appendix B, “Hardware Specifications,” provides specifications of servers measured.
