嵌入式网络协议引擎ETA加速网络应用程序处理

1、嵌入式网络协议引擎(ETA)加速网络应用程序处理日期：2003-121概要今天的网络环境和网络操作系统Linux 的tcp/ip stack 的性能嵌入内核的tcp 加速模块ETA的现实应用2今天的网络环境和操作系统310G的以太网络带来什么通用处理器必须每67奈秒处理一个数据包 超过通用处理器处理极限协议占用太多时间4NP是唯一的解决方案吗?Intel Ixp1200带来好处网络处理器对软件系统结果带来的问题灵活性和高性能的矛盾5系统中断 内存拷贝缓冲区管理系统呼叫Tcp/ip协议应用程序只能占用3%谁占用的cpu的时间.6Linux 的tcp/ip协议7内核的resource manage

2、r (ckrm)控制 diskIO 的资源分配控制网络的IO分配目前还不成熟8我们的解决方案区分出独立的协议处理层固定一个cpu来专门处理协议制定一个异步的通讯接口新的socket库来利用ETA9ETA System ArchitectureLANStorageIPCETA Host InterfaceIP StorageDriverKernelApplicationsUser Socket ApplicationsSocketProxyNetwork stackVirtualized, asynchronous queuing and event handlingEngine Archite

3、cture & platform integrationNetwork FabricPacket Processing Engine10Direct Transport InterfaceETAPacket Processing EngineNICApplication (Kernel or User)Adaptation LayerDTIEventQueueDTIRxQueueDTITxQueueAnonymousBufferPoolDTIDoorbellShared Host MemoryAppBuffersNIC11DTI Operation Model DTI operations:C

4、onnection requests (Connect, Listen, Bind, Accept, Close, )Data transfer requests (Send, Receive)Misc. operations (Set/Get Options,)EVENT AEVENT BEVENT CEventQTxQOP AOP CRxQOP BOP DDTIDoorbellProcess OperationServiceDoorbellDe-Queue OperationDescriptorPostCompletionEventPostETA InterruptEvent (if wa

5、iting)Host ApplicationAdaptation layer12ETA PPESoftwareGigabit NICs (5)ETA Host InterfaceKernelTestProgramCPU 0Host2.4 GhzCPU 1PPE2.4 GhzOff-the-shelf Linux ServersHostMemoryClientsClientsClientsClientsTestClientsKernelAbstraction LayerETA Test Environment132P SMP ProfileProcessing requirements in m

6、ultiple componentsTCP/IP is the largest single component, but is small compared to totalThe copy overhead is required to support legacy (synchronous) socket semanticsInterrupts and system calls are required in order to time-share the CPU resources14ETA Profile (1 host CPU + 1 PPE)Processing times are compressedIdle time represents CPU resource that is usable for applicationsAsynchronous queuing interface avoids copy overheadInterrupts avoided by not time-sharing CPUSystem cal