New XMC modules interface 10-Gig Ethernet to PCI Express
Acromag’s new XMC-6260 and XMC-6280 mezzanine modules provide a 10-gigabit Ethernet (10GbE) interface solution for data-intensive, real-time embedded computing systems. Ultra-high performance is achieved using a TCP/IP offload engine (TOE) ASIC connected to a PCI Express Gen2 x8 interface. The XMC-6260 has dual XAUI 10GBASE-KX4 ports and supports conduction-cooling or -40 to 85°C operation. The XMC-6280 features four SFP+ ports for fiber or copper cables. Applications include high-speed data storage, image collection/transfer, distributed control networks, and board-to-board interfaces.
Acromag’s network interface modules pair a purpose-built, multi-protocol processor with up to four channels of 10GbE connectivity. This solution offloads TCP overhead tasks from the host CPU to maintain maximum computing speeds. To meet the needs of data-intensive real-time applications, these fully integrated network interface cards (NIC) employ the Chelsio T4 processor, which has four XGMAC (10GbE) interfaces and supports up to 1M connections. Five gigabits of DDR3 memory enhances the number of virtual connections. The T4 chip provides full offload support for TCP, UDP, iSCSI, and Fibre Channel over Ethernet (FCoE). Other functions include high-performance packet switching, traffic filtering and management. By relieving the host CPU of these network processing tasks, very low Ethernet latency and high-level determinism are reliably achievable.
An 8-lane PCIe host interface provides a high-speed connection to the system processor. With support for 5Gbps data rates, the PCIe interface delivers up to 32Gbps of bandwidth to the server. This connection enables stateless offloads, packet filtering (firewall offload), and traffic shaping (media streaming).
The TOE ASIC has hundreds of programmable registers for protocol configuration and offload control. As a result, these modules can offload TCP processing per connection, per server, per interface. They can also globally and simultaneously tunnel traffic from non-offloaded connections to the host processor for the native TCP/IP stack to process. Additionally, the modules provides a flexible zero-copy capability for regular TCP connections, requiring no changes to the sender, to deliver line rate performance with minimal CPU usage.
The cards integrate a high-performance packet switch, which allows switching of traffic from any of the input ports to any of the output ports (wire-to-wire), and from any of the output ports to any of the input ports (host-to-host).