NVIDIA 网络打造超低时延证券解决方案.pdf

1、NVIDIANVIDIA网络打造超低时延证券解决方案陈龙，高级市场开发经理2020.December#page#构建最快的交易系统低时延网卡低时延交换机connectsLinkyou toXmorel低时延线缆降低拥塞时延#page#时延分布ServerSwitchDOWN-ModuleQUEUEMAC+PHYCPUPassive Copper0.3nsDDRSerdes100-200nsDDR DIMTCP/IP5-15USActive Copper 1ns10-100GbpFiber 10ns-1Tbps2nsSwitch64-128Gbps200-400n510-100Gbp10-100

2、GbpainpowainpowMAC+PHY2nsPassive Copper 0.3nsPassive Copper0.3ns100-200nsSINSerdesActive Copper insActive Copper ins100-200nsFiber 10nsFiber 10nsPropagationLatency#page#高速转发和时延关系（1）10G/25G/100G128GbpsStoreForward Latency100080600h4bb.h4002000212102425Gbps100Gbps10Gbps#page#高速转发和时延关系（2）目由于在线交易时，最易发生的

3、“微爆流”，使网络交换机由于拥堵延退传输甚至高达2毫秒Congestion_Latency(us）锁剪丽领题物钢加的住B88838BBR#page#三种低时延方案NVIDIAVMA-兼容性好平衡性高RoCEInfiniBand极致低时延#page#VMA硬件卸载降低转发延时Standard ApplicationStandard Applicationwithout NVIDIA VMAwith NVIDIA VMA跳过Kernel层User ApplicationUser ApplicationUser SpaceVMA运行在用户态，精简了协议栈Socket APISocket API的处理

4、，带来处理时延的大幅降低VMATCPUDPKernelBypassNIC Driver应用透明用于上层软件开发不感知VMA，开发与TCP/IP无差别对端无需做任何修改即可通信#page#NVIDIAVMA-追求极致低时延KERNELVSVMACONNECTX-5UDP LATENCY苏0Kernel2.943VMA1.0261432512102464173256sLatency Ping-Ping vs.#connections Multicast with VMA Socketxtreme1.6861.6631.6321.581501.361.3871.361.3331591.1311002

5、01010002030nVIDL#page#NVIDIAVMA-一举多得对于金融交易系统，VMA带来众多的好处不仅使得时延大幅降低，而且稳定，同时释放了宝贵的CPU资源，提升服务器的应用价值，葆护好用户的投资LowerLower Latency80%LATENCY*LatencyLessThroughput85%r CPU utilizationLowerCORESLOWERLess90%Lower JitterJITTER*LOWEHGb/sHighest throughput100PER PORTHIGHERComparing to Kernel performance#page#ROCE

6、进一步释放CPU资源Server-InitiatorServer-InitiatorServer-TargetServer-TargetApplicationApplicationbaforApplicationApplicationSocketsSocketsNIC DriverNIC DriverRoCERNICSRNICNICNIC武一市#page#ROCE保证低时延的同时进一步释放CPU资源ROCEWithout RDMAWith RDMA and Offload25G100GRoCELatency天73253%CPU-88%CPUUtilizationUtilization27060