Hi, Both have their use cases in application architecture (dynamic load balancing vs. fixed pipelining) and support is needed on both platforms. It depends on the application which model suites the best. In addition a schedule() driven application can use poll queues as internal storage, etc.
-Petri > -----Original Message----- > From: [email protected] [mailto:lng-odp- > [email protected]] On Behalf Of ext Jacob, Jerin > Sent: Monday, December 01, 2014 1:36 PM > To: [email protected] > Subject: [lng-odp] ODP pull vs push executuion model > > > I would like to understand the ODP _application_ expectation with ODP > pull(odp_schedule()) and push(polled queues) execution model. > > Is it expected that a confirming ODP platform implementation need to > support both models ? > or at least one model shall be supported by confirming ODP platform > implementation ? > > In Cavium perspective, We have two different category of SoCs in which we > are trying to implement ODP. > 1) Networking SoC with hardware accelerated odp_schedule() support. > 2) Enterprise server(ARMV8) SoC with Network Function virtualization > support and NFV capabilities are > inline with state of art enterprise SoC like Intel(DPDK) per socket. > > So in this context, My queries are > 1)Does it make sense to support(emulate in software) polled queues in > former category of SoC ? if yes what is the use case ? > 2)An confirming ODP implementation needs to emulate odp_shedule() in > software if HW supports is not available ? > Is it worth to emulate ? or application to use the push execution > model(polled queues) like DPDK ? > > Any thoughts ? > > - Jerin > > > _______________________________________________ > lng-odp mailing list > [email protected] > http://lists.linaro.org/mailman/listinfo/lng-odp _______________________________________________ lng-odp mailing list [email protected] http://lists.linaro.org/mailman/listinfo/lng-odp
