Infinetics has designed and implemented Long Hop™, a high performance network connection architecture that provides significantly better tradeoffs between bandwidth, latency, fault tolerance and cost than previously known approaches. Long Hop networks are more cost efficient than a fat-tree, supporting at least 50% more servers using the same network hardware while providing equivalent bandwidth and improved latency of at least 25%. Oversubscription and fault tolerance can also be smoothly varied to meet cost and performance goals. Long Hop’s regular wiring patterns maximize the number of short connections and minimize the number of long distance connections between server rows or containers, thereby reducing cabling costs and simplifying expansion by reducing or eliminating rewiring. Long Hop networks well serve the needs of modern data centers by providing consistent bandwidth under varying load conditions via extensive multi-path support and automatic load balancing, allowing mobile workloads to be placed without regard to available bandwidth, and allowing services to be isolated in a multi-tenant environment to match the business needs of customers.

Long Hop achieves this by combining patent-pending wiring methods with a simple high-performance, TCAM-efficient forwarding method well suited to the current generation of network switch hardware and SDN controllers supporting OpenFlow or proprietary southbound APIs. The wiring method starts with a highly symmetrical framework well suited to economical, fault tolerant switching, and overlays another symmetrical network of short cuts that simultaneously amplify available bandwidth as well as reduce latency. The end result gives the ability to design and construct networks that precisely meet implementation-specific business needs

System Overview

Centralized components

  • Data factory
  • Control plane executive

Distributed components

  • Infinetics-enabled switches
  • Soft switch for hypervisors

Data Center Wiring

Single switch layer
  • No aggregation layer switches
All switches are peers
  • All switches have ports for servers
  • Highly regular wiring patterns
Many short interconnects, fewer long interconnects
Switches can be physical or within hypervisor

Control Plane Executive

Central controller of switches and hypervisors
  • Sends forwarding table updates to switches
  • Receives local topology and traffic load from switches
Universal Boundary Manager (UBM)
  • Abstracts the physical network to a unified and hierarchical logical group with rights-based inheritance for security and QoS parameters
  • Controls visibility of hosts and services
  • Provides a single “Firewall” around perimeter of entire layer 2 network managing routing decisions for quality assurance and security enforcement
  • Scriptable policy management based upon time-of-day, congestion and application type

Data Factory

Stores network configuration
  • Security, segmentation & QoS policies
  • Host location
Collects network status and traffic patterns
  • Topology updates from switches
  • Link utilization
Redundancy & Synchronization
  • Tunable replication/distribution
  • Supports industry standard interfaces and tools

Hypervisor Soft Switch

Supports all Infinetics hardware switch features
  • Hugely scalable flat layer 2 network
  • Security
  • Traffic management
  • Broadcast suppression
Supports topologically-enhanced multi-homing
  • Multiple NICs absorbed into switch topology
VM motion and life cycle management
Highly portable implementation

Copyright © 2009-2013 Infinetics Technologies, Inc. All Rights Reserved. Contact Us +1 877.438.1010       Legal