Font Size: a A A

Research On Packet I/O Technology For General-purpose Multi-core Network Processor

Posted on:2013-02-06Degree:MasterType:Thesis
Country:ChinaCandidate:L TangFull Text:PDF
GTID:2298330422474261Subject:Software engineering
Abstract/Summary:PDF Full Text Request
With the development and maturation of microelectronic and microprocessortechnology, the parallel processing performance of multithread in general-purposeprocessors has been largely improved. Designing network processors (especially for therouters in the edge/access network) based on general-purpose multi-core processorcould utilize the proven techniques in the domain of general-purpose microprocessors(GPP), employing the merits of relative low cost, short developing time and lowfailing possibilities. Therefore, it becoming a hot research topic in the architectureresearch of network processor (NP). The combination of GPP and hardware acceleratingengine of packet processing have two advantages: the programming flexibility could beimproved by embedding GPP; on the other hand, the high packet processingperformance can be obtained by the network-processing accelerating engine (NPE)implemented in hardware However, packet data transmission between GPP and NPE isthe bottleneck which limits the improvement of NP’s performance. Facing with thisproblem, a novel technology of packet I/O, called PacketDirect, has been proposed forthe general-purpose multi-core network processor in this thesis.1. The packet transmitting and receiving mechanism of PacketDirect are proposed,the Run-to-Completion model of NP packet processing is fully utilized by transmittingpacket flow in the form of link list and processing packets with polling mechanism.Packet transmission procedure can be simplified and the descriptor management andinterrupt processing cost in traditional packet I/O could be avoided.2. The resource-optimized management mechanism of packet buffer descriptors isproposed. By using packet blocks a basic unit when allocate and manage packet buffer,hardware storage requirements are efficiently reduced in descriptor management engine.The number of descriptors can be avoided to be the bottleneck of the packet I/Operformance in the whole system.3. The performance and resource model of PacketDirect technology and generalpacket I/O technology are established, including packet transmission delay,transmission rate and descriptor management storage. Theoretically analysis shows thatPacketDirect has lower delay, higher transition rate and lower storage usage comparedwith traditional packet I/O technology.4. the prototype system of PacketDirect has been implemented based on theCompassNP platform. Based on the system, the experiments and tests are carried on thepacket I/O.The experimental results show that PacketDirect can provide high-speedtransmission channel for the general-purpose multi-core network processors.In summary, PacketDirect technology has been proposed in the thesis to solve thebottleneck of packet I/O in general-purpose multi-core network processors. The technology has been implemented and embedded in the network processor platformbased on the independent multicore CPU. Therefore, the research has theoretic meaningand practical value for network processor design.
Keywords/Search Tags:network processor, PacketDirect, packet I/O
PDF Full Text Request
Related items