Users Guide
Table Of Contents
- Revision History
- Table of Contents
- Regulatory and Safety Approvals
- Functional Description
- Network Link and Activity Indication
- Features
- Software and Hardware Features
- Virtualization Features
- VXLAN
- NVGRE/GRE/IP-in-IP/Geneve
- Stateless Offloads
- UDP Fragmentation Offload
- Stateless Transport Tunnel Offload
- Multiqueue Support for OS
- SR-IOV Configuration Support Matrix
- SR-IOV
- Network Partitioning (NPAR)
- RDMA over Converged Ethernet – RoCE
- Supported Combinations
- Installing the Hardware
- Software Packages and Installation
- Windows Driver Advanced Properties and Event Log Messages
- Teaming
- System-level Configuration
- ISCSI Boot
- VXLAN: Configuration and Use Case Examples
- SR-IOV: Configuration and Use Case Examples
- NPAR – Configuration and Use Case Example
- RoCE – Configuration and Use Case Examples
- DCBX – Data Center Bridging
FeaturesNetXtreme-E User’s Manual
September 4, 2019 • NetXtreme-E-UG103 Page 27
VXLAN
A Virtual eXtensible Local Area Network (VXLAN), defined in IETF RFC 7348, is used to address the need for
overlay networks within virtualized data centers accommodating multiple tenants. VXLAN is a Layer 2 overlay
or tunneling scheme over a Layer 3 network. Only VMs within the same VXLAN segment can communicate with
each other.
NVGRE/GRE/IP-in-IP/Geneve
Network Virtualization using GRE (NVGRE), defined in IETF RFC 7637, is similar to a VXLAN.
Stateless Offloads
RSS
Receive Side Scaling (RSS) uses a Toeplitz algorithm which uses 4 tuple match on the received frames and
forwards it to a deterministic CPU for frame processing. This allows streamlined frame processing and balances
CPU utilization. An indirection table is used to map the stream to a CPU.
Symmetric RSS allows the mapping of packets of a given TCP or UDP flow to the same receive queue.
TPA
Transparent Packet Aggregation (TPA) is a technique where received frames of the same 4 tuple matched
frames are aggregated together and then indicated to the network stack. Each entry in the TPA context is
identified by the 4 tuple: Source IP, destination IP, source TCP port, and destination TCP port. TPA improves
system performance by reducing interrupts for network traffic and lessening CPU overhead.
Header-Payload Split
Header-payload split is a feature that enables the software TCP/IP stack to receive TCP/IP packets with header
and payload data split into separate buffers. The support for this feature is available in both Windows and Linux
environments. The following are potential benefits of header-payload split:
• The header-payload split enables compact and efficient caching of packet headers into host CPU caches.
This can result in a receive side TCP/IP performance improvement.
• Header-payload splitting enables page flipping and zero copy operations by the host TCP/IP stack. This
can further improve the performance of the receive path.
UDP Fragmentation Offload
UDP Fragmentation Offload (UFO) is a feature that enables the software stack to offload fragmentation of UDP/
IP datagrams into UDP/IP packets. The support for this feature is only available in the Linux environment. The
following is a potential benefit of UFO:
• The UFO enables the NIC to handle fragmentation of a UDP datagram into UDP/IP packets. This can result
in the reduction of CPU overhead for transmit side UDP/IP processing.