SPECTRUM Throttling Concepts
Throttling is a tool to control the volume of traffic to an SNMP agent.
Spectrum Throttling regulates how many SNMP management requests are sent by Spectrum to a network device or application. It allows a Spectrum user to dictate the number of outstanding requests allowed for any given SNMP agent. (It does not create a delay between requests.)
If the number of outstanding requests is at the throttle limit, the next request is queued. When the number of outstanding requests falls below the limit, a request is dequeued and sent.
One throttle entry represents all of the components models for one physical device. Because Throttling attaches to the Device Model Type, model types derived from an SNMP-managed Device can use the throttling tool. When throttling is enabled, requests are handled in the order that they are received, queued to preserve that order, and no model has precedence over another.
Essentials Of Enabling/Disabling Throttling
The internal attributes associated with Throttling are:
Throttling attribute id 0x11f79 - Throttling is enabled when Throttling equals TRUE; disabled when Throttling equals FALSE. Most model types have throttling disabled by default.
Throttle_Count attribute id 0x11f39 - The Throttle_Count attribute's value represents the maximum number of outstanding requests allowed for the SNMP agent. A zero value disables throttling for the model and transitions the Throttling attribute to FALSE.
When Is Throttling Appropriate
There is no concrete guideline for applying throttling to an SNMP agent. The Spectrum user determines what degree of throttling is appropriate. This includes when throttling should be enabled and the appropriate limit for outstanding requests. Generally, if an SNMP agent is unable to respond to the typical volume of Spectrum requests, an SNMP Management Lost orange alarm or a Lost Contact red alarm is asserted on the models sharing that Network_Address. When this occurs it may be appropriate to use throttling.
An SNMP agent may have limited resources for handling requests from a network management application including such things as physical memory limitations on the device, excessive load on the device, and reduced priority for handling management traffic. The extra load from throttling may not be appropriate for SNMP agents that cannot handle the rate of Spectrum's SNMP management requests. Often these SNMP agents respond immediately to a request, but fail as the request rate increases. Remember that consistently high traffic volume causes requests to be queued and degrades the turn-around time. You may experience performance loss and delayed updates to model views.
If an SNMP agent is receiving management requests from several SpectroSERVERs or other management applications then throttling must account for this and the Throttle_Count value reduced appropriately. Throttling is applied per SpectroSERVER. If an SNMP agent is receiving management requests from more than one SpectroSERVER or other management application then throttling must account for this. Using multiple managers may require the Throttle_Count value to be reduced, which can cause the previously described delays.