PCIe delivers performance of up to 1.5Tb/s through 97 lanes in full-duplex mode.
The Avago ExpressFabric PEX9700 series of switches simplify connectivity of data center systems, delivering the highest PCIe switching performance available inside the rack.
Avago said that while PCIe is the ubiquitous interconnect for internal devices, external device connectivity typically requires using other architectures. This wastes power, reduces performance, and increases costs. With these PCIe switches, devices within a hyperscale system, NVMe enclosure, and rack-scale-based subsystems can now communicate directly through the same high-performance PCIe fabric, turning those devices into a single high-performance, high-capacity compute system.
PEX9700 key new switching features include:
- Shared I/O – Enables endpoints to be shared among multiple hosts to maximize system resource efficiency and reduce cost and power
- Tunneled Windows Connection – Interconnect up to 24 (+1 management host) nodes on a single chip or cascade multiple chips to connect up to 72 nodes with no performance loss
- Embedded DMA Engines – Equips each port with a NIC-DMA engine that enables efficient data transfers between multiple hosts
- Flexible Port Configuration – Allows up to 23 endpoints to be combined with ports as wide as 16 lanes, providing nearly 16GB/s of bandwidth on each link
- Downstream Port Containment – Isolates errors in any endpoint from bringing down the entire system
- Dedicated Port Management – Enables software control of switch features for customized implementations using Avago-developed software modules
“PCIe is fundamental to next-generation data centers to manage the growth and complexity of data traffic,” said Tom Swinford, senior vice president and general manager, Data Center Solutions Group at Avago. “With PCIe technology at the core of nearly every product in the data center, fabric convergence within the rack using Avago ExpressFabric is a powerful approach for intra-rack connectivity to enable powerful performance and scaling of systems.”