Intel’s newly launched Ethernet 800 series network adapter meets the high-speed transmission needs of high-performance storage, and how to use Intel Ethernet 800 series network card with ADQ acceleration technology to provide higher With high performance and lower latency, today we will continue to explain the key technologies of Ethernet 800 series network adapters in detail.
QSFP28, single port or dual port
PCle3.0 and 4.0, ×16
Ethernet Port Configuration Tool (EthernetPort Configuration Tool, EPCT)
Application Device Queues (ADQ).
Dynamic Device Personalization (DDP).
Support RDMA iWARP and RoCEV2
The Ethernet 800 series innovatively improves application efficiency and network performance to optimize high-performance server workloads such as NFV, storage, HPC-AI, and hybrid cloud.
Improve cloud application performance
Deliver high bandwidth on demand for cloud workloads (including edge services, web servers, database applications, cache servers, and storage targets) and increase application throughput.
ADQ uses enhanced traffic-oriented technology to improve the predictability of application response time
DDP enhances packet classification capabilities, and can provide up to 3 times the throughput for certain cloud workloads
Support RDMA iWARP and RoCE v2 transmission protocol at the same time, provide high-speed and low-latency connection for storage target
Optimize telecom load
Provide packet classification and sequencing optimization for high-bandwidth network and communication workloads (including mobile core, 5G RAN, and network equipment)
DDP supports existing and new specific protocols in the communication field, and can increase the packet processing efficiency of certain NFV workloads by 3 times
Support IEEE 1588v2 Precision Time Protocol (Precision Time Protocol), which can achieve high-precision time synchronization in 5G RAN deployment
Enhanced data plane development kit (DataPlane Development Kit, DPDK) support, improve the packet processing speed
Flexible configuration of multiple ports with EPCT
E810-CQDA1 and CQDA2 network adapters support a variety of system configurations to meet the workload requirements of users. The combination of more available ports and speeds simplifies verification and deployment.
Using EPCT, Intel E810 (dual port or single port) can be programmed into many different physical network adapters with a maximum throughput of 100Gbps
Key technologies of Intel Ethernet 800 series network adapters:
Improve the predictability of horizontal expansion.With the horizontal expansion of modern data centers, providing elastic and predictable application-level performance has become one of the key challenges. ADQ technology greatly reduces performance jitter by establishing dedicated queues for critical loads, and significantly improves the scalability and predictability of applications.
Lower jitter increases the predictability of application response time, which not only allocates more computing servers to tasks, but also allows more users to access the system and provides a better end-user experience. Even small applications can benefit from higher consistency, making it easier to meet SLAs.
ADQ uses a method of optimizing the application thread to the device data path to achieve data control, transmission and rate limiting for specific applications. This dedicated queue and the ability to adjust network traffic not only improves application performance, it also reduces latency and improves throughput.
Use Dynamic Device Personalization (DDP) to improve packet processing efficiency
By filtering data packets and combining with enhanced DPDK, DDP can support advanced data packet forwarding and efficient packet processing in scenarios such as cloud and NFV.
The Intel Ethernet 700 series introduced DDP technology for the first time, and some protocols can be added to the default protocol set in the firmware. The Intel Ethernet 800 series firmware loads an enhanced DDP configuration file during the driver initialization stage, as well as many protocols for specific workloads, thereby gaining greater flexibility. When there are multiple 800 series adapters in the system, the pipeline on each adapter can be independently programmed using different DDP configuration files.
IEEE1588 Precision Time Protocol (PTP)
The Intel Ethernet 800 series supports two versions of IEEE 1588 PTP v1 and v2, provides nanosecond time accuracy, and can accurately report the receiving time of each packet. This level of time accuracy helps ensure precise synchronization of network deployments from 5G RAN to financial services, industrial automation, and energy monitoring.
Improve throughput and reduce latency through remote direct memory access (RDMA)
RDMA provides high throughput and low latency performance for modern high-speed Ethernet by eliminating the three main network resource overheads of TCP/IP stack process, memory copy, and application context switching. Intel Ethernet 800 series network adapters support all mainstream storage transmission protocols, including iWARP, RoCE v2, and NVMe over TCP.
RoCE (RDMA over Converged Ethernet):
RoCE v2 replaces the InfiniBand physical layer and data link layer with Ethernet, runs on UDP/IP, and can be routed on the IP network.
iWARP, based on IETF standard protocol:
Deliver RDMA on top of the common TCP/IP protocol. iWARP RDMA operates through a standard network and transport layer, and works with all Ethernet network infrastructure. TCP provides flow control and congestion management without the need to establish lossless Ethernet. iWARP is a highly routable and scalable RDMA implementation.
Protection, detection, recovery
Zero Trust is designed based on the concept that before granting access, you need to verify any content that is trying to connect or gain access.
The Intel Ethernet 800 series provides flexible protection for the platform through three security mechanisms of protection, detection and recovery, as well as the Hardware Root of Trust. Through this design, its built-in fault detection can protect the firmware and key device settings, and perform automatic device recovery to ensure that the device returns to its original programming state.
Intel Ethernet 800 series adapters are based on Ethernet Controller E810, and the overall characteristics are as follows:
- Conforms to the current 0.9 version of the PCle 4.0 standard
- Concurrency of 256 non-posted requests
- Basic mode VF is compatible with Intel Adaptive Virtual Functions Specification (Intel Adaptive Virtual Functions Specification)
- Tx/Rx queue
-2048Tx queue and 2048 Rx queue
-Queue dynamically allocated to functions and VSls
-2048 interrupt vectors, allocated to queues and other use cases in a flexible way
-Multiple interruption adjustment scheme
- Control Queues (Control Queues, also known as Admin Queues)
-For PF-VF and driver-driven mailbox queues
-Management queue for software-firmware control flow
-Sideband Queues for software access to E810 internal IP
- 256 Tx doorbell (DB) queues
- 512 Tx completion queue
- Each Tx queue has a quantum descriptor (QD) queue, and quantum information is also embedded in the Tx doorbell
- Programmable Rx descriptor field
- Enhanced Data Plane Development Kit (DPDK)
- General function
-Analysis, exchange, ACL, classification, package modification and other stages
-Programmable data packet processing pipeline
-Based on configuration files
-Propagation priority between stages
-Resolvable up to 504B Baotou
-Based on analytical graphs
-Programmable analysis engine
- Binary Classifier (VEB Switch)
-768 switch port (VSls)
-Programmable forwarding rules
-8K programmable TCAM entries
-The width of the tile can reach n*40b
- Classification filter
-Statistical distribution based on hash
-Based on Intel EthernetFD flow classification
-Flow-based iWARP and RoCE flow identification
-Insert (Tx), delete (Rx), and packet VLAN modification
-L3 and L4 checksum and CRC
- Realize host virtualization through VMDQ and SR-IOV
- Up to 256 SR-IOV virtual functions
- Stateless offloading of tunnel packets (network virtualization support)
- Malicious VF protection
- Virtual Machine Load Balancing (VMLB)
- Advanced packet filtering
- VLAN supports VLAN tag insertion, stripping and packet filtering, up to 4096 VLAN tags
- VXLAN, GENEVE, NVGRE, MPLS, VXLAN-GPE with Network Service Headers (NSH)
- Intel Ethernet Adaptive Virtual Function drivers
- Support iWARP and RoCEv2
- 256K queue pairs
- Send queue push mode
Note: When E810 is configured for >4 port operation, RDMA is not supported.
- WFQ transmission scheduler has nine programmable layers
- Share pipeline, avoid starvation
- QoS through 802.1p PCP or Differentiated Services Code Point (DSCP) value
- Package adjustment
- SMBus operation up to 1Mb/s
- NC-SI 1.1 interface conforming to DMTF, 100Mb/s
- MCTP over PCIe and SMBus
- Realize enterprise-level management solutions through local BMC
- SNMP and RMON statistical counters
- Watchdog timer
- PLDM over MCTP; PLDM monitoring; PLDM firmware upgrade: PLDM for RDE
- Firmware management protocol support
- Support PCI power management status D0, D3hot, D3cold
- APM WoL support in Do, D3hot and D3cold
- Timestamp of each Rx packet
- Optional Tx packet timestamp
- Support IEEE PTP1588v1/2
- Synchronize time stamps with other local platforms
- Signed UEFl option ROM compatible with HTTPS boot
- Hardware-based root of trust
- Identity verification based on NVM Read and Power On
- Built-in firmware/key setting failure detection and automatic device recovery