From: Shuah K. <shu...@hp...> - 2008-12-08 18:42:18
|
After an evaluation of the current openhpi SNMP functionality, it is evident that the SNMP trap capability is missing. The initial design and architecture appears to have originated from the intent to keep the HPI interface to clients and HPI hardware interface (plug-ins) the same for simplicity which is polling. However there seem to be no compelling technical reasons for sticking to the simple polling model for the plug-in and hardware interface. Moving toward an event driven model will improve performance and plug-ins are free to choose an event driven model. In the case of SNMP based plug-ins, taking the approach of a central trap distribution module offers the following benefits: (please note that the client to daemon interface will not change) • generic infrastructure code for trap reception using a single trap port as opposed multiple trap ports for each plug-in that wants to receive SNMP traps. • easier to configure hardware management software with a single trap port • Plug-in can continue to use polling or switch to traps as they see fit. Here is an high-level proposal for a central infrastructure level SNMP trap distribution module and its interface to plug-ins. Please review and give me comments and feedback. • OpenHPI will register a new trap port with IANA. Follow IANA port application process. • openhpi daemon opens socket to receive SNMP traps at the openhpi daemon SNMP trap port. A new ABI is added to hand traps to plug-ins. • Plug-ins needing to receive SNMP traps will implement this new SNMP Trap ABI routine. • When a trap arrives, daemon will call plug-in SNMP trap ABI routines until one of the plug-ins accept the trap or nobody claims the trap. This is the same as the ABI mechanism we have today. Please give me comments and feedback. The intent to get initial feedback and schedule the implementation work to be done sometime next year. -- Shuah |