Mellanox Technologies introduced its data center Ethernet Cloud Fabric (ECF) technology based on its second generation, Spectrum-2 silicon, which can deliver up to 16 ports of 400GbE, 32 ports of 200GbE, 64 ports of 100GbE, or 128 ports of 50/25/10/1GbE.
Mellanox ECF combines three critical capabilities:
Packet forwarding data plane
- 8.33 Billion Packets per second – Fastest in its class
- 42MB Monolithic and fully shared packet buffer to provide high bandwidth and low-latency cut-through performance
- Robust RoCE Datapath to enable hardware accelerated data movement for Ethernet Storage Fabric and Machine Learning applications
- Half a million flexible forwarding entries to support large Layer-2 and Layer-3 networks
- Up to 2 Million routes with external memory to address Internet Peering use cases
- 128-way ECMP with support for flowlet based Adaptive Routing
- Hardware-based Network Address Translation
- 500K+ Access Control List entries for micro-segmentation and cloud scale whitelist policies
- 500K+ VXLAN Tunnels, 10K+ VXLAN VTEPs to provide caveat-free Network Virtualization
Flexible and fully programmable data pipeline
- Support for VXLAN overlays including single pass VXLAN routing and bridging
- Centralized VXLAN routing for brown field environments
- Support for other overlay protocols including EVPN, VXLAN-GPE, MPLS-over-GRE/UDP, NSH, NVGRE, MPLS/IPv6 based Segment routing and more
- Future-proofing with programmable pipeline that can support new, custom and emerging protocols
- Hardware optimized stages that accelerate traditional as well as virtualized network functions
- Advanced modular data plane and integrated container support enables extensibility and flexibility to add customized and application specific capabilities
Open and Actionable telemetry
- 10X reduction in mean time to resolution by providing a rich set of contextual and actionable Layer 1-4 “What Just Happened” telemetry insights
- Hardware based packet buffer tracking and data summarization using histograms
- More than 500K flow tracking counters
- Open and Extensible platform to facilitate integration and customization with 3rd party and open source visualization tools
- Support for traditional visibility tools including sFlow, Streaming and In-band telemetry
Marvell said its Ethernet Cloud Fabric incorporates Ethernet Storage Fabric (ESF) technology that seamlessly allows the network to serve as the ideal scale-out data plane for computing, storage, artificial intelligence, and communications traffic.
“The Spectrum-2 switch ASIC operates at speeds up to 400 Gigabit Ethernet, but goes beyond just raw performance by delivering the most advanced features of any switch in its class without compromising operation ability and simplicity,” said Amir Prescher, senior vice president of end user sales and business development at Mellanox Technologies, “Spectrum-2 enables a new era of Ethernet Cloud Fabrics designed to increase business continuity by delivering the most advanced visibility capabilities to detect and eliminate data center outages. This state-of-the-art visibility technology is combined with fair and predictable performance unmatched in the industry, which guarantees consistent application level performance, which in turn drives predictable business results for our customers. Spectrum-2 is at the heart a new family of SN3000 switches that come in leaf, spine, and super-spine form factors.”
The Spectrum-2 based SN3000 family of switch systems with ECF technology will be available in Q3.
With Mellanox, NVIDIA targets full compute/network/storage stack
NVIDIA agreed to acquire Mellanox in a deal valued at approximately $6.9 billion.
The merger targets data centers in general and the high-performance computing (HPC) market in particular. Together, NVIDIA’s computing platform and Mellanox’s interconnects power over 250 of the world’s TOP500 supercomputers and have as customers every major cloud service provider and computer maker. Mellanox pioneered the InfiniBand interconnect technology, which along with its high-speed Ethernet products is now used in over half of the world’s fastest supercomputers and in many leading hyperscale datacenters.
NVIDIA said the acquired assets enables it to data center-scale workloads across the entire computing, networking and storage stack to achieve higher performance, greater utilization and lower operating cost for customers.
