Optimized for Multi-core Processor Servers - Intel® Ethernet Flow Director
Today’s data centers depend on the multiprocessing, high performance capability of servers to increase system throughput, responsiveness, and reliability through the introduction of additional hardware threads, CPUs, or cores. But in a multiprocessing environment, it is essential to ensure a coordinated affinity of protocol processing and network applications on the same target cores. This affinity significantly reduces contention for shared resources, minimizes software synchronization overheads between cores, and enhances cache efficiency.
Receive Side Scaling (RSS)
Resolves the single-processor bottleneck by enabling the receive side network load from a network adapter to be shared across multiple cores. RSS enables packet receive-processing to scale with the number of available cores. However, RSS has a limitation. it cannot steer an incoming network flow to the same core where the application process resides. RSS does not maintain the Traffic Flow>Core (Application) relationship. If an application is running on one core, while RSS has scheduled receive traffic to be on another core, poor cache efficiency and significant core-to-core synchronization overheads will result. The overall system performance can be significantly degraded. The Intel® Ethernet Flow Director and the Application Target Routing (ATR) service found in Intel’s Ethernet controllers, is an advanced network offload technology that provides the benefits of parallel receive processing in multiprocessing environments that automatically steer incoming network data to the same core on which its application process resides. Intel Ethernet Flow Director and ATR preserve the Traffic Flow>Core (Application) relationship. As a result, Intel Ethernet Flow Director and ATR can significantly lower latency and improve CPU usage.
Intel Ethernet Flow Director enables administrators to define “signature filters” and the ATR service on the Intel Ethernet controller uses these filters to ensure that all packets in a TCP flow are processed by a single core. This intelligent offload capability supports advanced filters that direct receive packets by their flows to different queues and enables tight control on routing a flow in the platform. It matches flows and CPU cores for flow affinity and supports multiple parameters for flexible flow classification and load balancing.
Best Choice for Virtualization
Intel leads the industry in virtualization by being the first to provide virtualization for all the major operating systems and working with the OEMs to implement virtualization not only on the adapter but also on the platform.
Intel® Virtualization Technology for connectivity (Intel® VT-c)
Intel® Ethernet Controllers includes Intel® Virtualization Technology for connectivity (Intel VT-c) to deliver virtualized I/O performance optimizations and Quality of Service (QoS) features designed directly in to the controller’s silicon. Working in conjunction with virtualization optimized drivers, PCI-SIG* Single Root I/O Virtualization and Sharing (SR-IOV) can be used to help reduce I/O bottlenecks, and improve the overall server performance.
Hypervisor BYPASS using SR-IOV
Bypassing the hypervisor and allowing direct hardware access by virtual machines, reduces CPU overhead, reduces latency, and increases network throughput. Most of the current hypervisor releases have been enabled to partition a single physical Ethernet controller into multiple virtual Ethernet controllers that can be used directly by VMs by taking advantage the PCI-SIG* SR-IOV standard. The use of these virtual controllers, known as Virtual Functions (VF), enables additional QoS features in the controller’s silicon to manage and direct traffic such as traffic isolation, port partitioning with bandwidth allocation and on-chip VF-VF switching.