This article was accepted into the corpus but its outbound wikilinks were never NER-processed — typical at the deepest BFS hop or when the run's entity cap was reached. No expansion funnel to show.
| Data Center Fabric | |
|---|---|
| Name | Data Center Fabric |
| Type | Network architecture |
| Use | High-performance data center networking |
Data Center Fabric Data center fabric refers to an architectural approach to interconnect compute, storage, and management resources within a modern Amazon Web Services, Microsoft Azure, Google Cloud Platform, Facebook, Google, Microsoft, Apple Inc., Netflix or Alibaba Group scale facility using high-density switching, programmability, and automation. It evolved from developments in Ethernet, InfiniBand, Fibre Channel, and innovations by vendors such as Cisco Systems, Juniper Networks, Arista Networks, Huawei, Hewlett Packard Enterprise to support workloads driven by Big data, Machine learning, High-performance computing, and web-scale services. Fabric designs aim to deliver low-latency, high-throughput, and deterministic behavior for tenants including hyperscale operators like OpenAI, Tencent, and Baidu.
A fabric implements a flattened, Clos-like interconnect inspired by the Banyan network and the Benes network, integrating layer 2 and layer 3 functions across spine and leaf layers employed by providers such as Equinix, Digital Realty, and OVHcloud. It leverages protocols and standards including Ethernet, Internet Protocol, Border Gateway Protocol, Segment Routing, Virtual Extensible LAN, and Virtual Private LAN Service while interfacing with storage fabrics like Fibre Channel and compute fabrics like PCI Express. Operators deploy fabrics to support services such as Content delivery network, Platform as a Service, and container orchestration systems like Kubernetes and Docker Swarm.
Typical components include leaf switches, spine switches, top-of-rack units, white-box switches from vendors like Arista Networks or ODMs used by Facebook, network interface cards from Intel Corporation or Broadcom, and specialized accelerators such as NVIDIA DPUs. Control plane elements incorporate routing daemons derived from Quagga or FRRouting and management systems influenced by OpenStack and ONAP. Fabric architectures often integrate telemetry and observability stacks modeled after Prometheus and ELK Stack to provide metrics, tracing, and logging, and make use of orchestration from Ansible, Terraform, and Chef.
Design patterns include Clos (leaf-spine), fat-tree pioneered in research from University of California, Berkeley and implemented by hyperscalers, two-tier and three-tier hierarchical models used by Verizon and AT&T, and flattened approaches seen in Google's Jupiter network. Overlay models using VXLAN, NVGRE, and GRE permit multi-tenant isolation influenced by work from VMware and Cisco Systems while underlay fabrics use MPLS or Segment Routing for traffic engineering. Emerging designs incorporate programmable data planes via P4 and switches from the Open Compute Project.
Management relies on intent-based networking concepts popularized by Arista Networks and Cisco Systems and uses controllers such as OpenDaylight and proprietary systems from Juniper Networks and Nokia. Orchestration integrates with cloud management platforms like OpenStack, VMware vSphere, and Google Kubernetes Engine to provision virtual networks, security groups, and service function chaining as seen in NFV deployments championed by ETSI. Automation frameworks often combine Ansible, Puppet, and Terraform alongside telemetry consumers like Grafana and AIOps solutions from Splunk or Elastic NV.
Fabrics target predictable latency and jitter for workloads from Stripe payments, streaming at Netflix, and inference at OpenAI by employing congestion control algorithms such as Data Center TCP and techniques like ECN, priority flow control, and RDMA over Converged Ethernet influenced by Mellanox Technologies. Scalability is achieved via Clos scaling principles studied in Carnegie Mellon University and MIT research, while reliability borrows from carrier-grade practices used by AT&T and Verizon including fast reroute, link aggregation, and in-service software upgrades (ISSU). Load balancing at scale uses hardware-based equal-cost multipath and software load balancers inspired by HAProxy and NGINX.
Security mechanisms integrate tenant isolation methods such as VRFs, VXLAN segmentation, and microsegmentation popularized by VMware NSX and companies like Illumio. Access control leverages AAA systems from Cisco Systems and identity providers like Okta, and fabrics implement threat detection with partners including CrowdStrike and Palo Alto Networks. Regulatory and compliance considerations reference standards organizations like ISO, NIST, and initiatives by PCI Security Standards Council for protecting tenant workloads in colocation facilities operated by Equinix and cloud regions run by Amazon Web Services.
Deployment models span on-premises enterprise installations at Goldman Sachs and Walmart, hybrid architectures connecting to Azure Stack and AWS Outposts, and public cloud fabrics inside Google Cloud Platform and Microsoft Azure. Use cases include AI training clusters at NVIDIA partner sites, HPC at national labs such as Lawrence Livermore National Laboratory and Los Alamos National Laboratory, financial trading platforms in New York Stock Exchange data centers, content delivery by Akamai Technologies, and multi-tenant hosting by DigitalOcean and Linode.
Category:Computer networking