Non-Stop Internet was designed as a distributed and modular software from day one.
Distribution provides the ability to collect data locally in each network location and merge them into a single routing policy, making routing consistent across a large AS with many data centers.
Modularity also provides the ability to size each part of the system individually and cope with scalability targets.
Geographically Distributed System
A typical NSI setup is made of a NSI Server and NSI Devices.
The NSI server holds the global functions of the system:
- Graphical User Interface
- Statistical analysis and aggregation algorithms
- Routing Decision Engine
The NSI Devices are distributed in data-centers. They are coordinated by the NSI Server through secure and resilient communication channels. They handle all local communication with the edge routers and remote networks:
- Performance probing
- SNMP collection
- Flow collection
- CLI automations
- BGP sessions
- Collecting topology information
- Sending outbound routing orders
- Route Preference Protocol inbound routing
- Other side features, e.g. TFTP, syslog, configuration versioning…
Having a distributed systems allows for:
- DC local performance evaluation
- DC local failures and packet loss detection
- Global routing policy, e.g. managing global transit CDR
- Sharing destination probes
Note: BORDER 6 also operates cloud servers for collecting, sharing and updating information on NSI servers. This allows for automation of tasks as well as providing our customers with global data analytics.
Physical or Virtual
NSI elements are Linux kernel based and were designed to operate onto x86_64 platforms. They can run on virtual environments (KVM, VMware), commodity server hardware or the BORDER 6 appliance.
NSI appliance is a 1 rack-unit packaged PC server designed for 100% availability and high I/O performance. For environments with no easy access to virtualisation hosts & less than 5Gb/s IMIX total traffic, the BORDER 6 NSI appliance is recommended.
Virtualisation of the NSI modules and specific hardwares are preferred for environments with higher flow density than IMIX and high bandwidth, that require more I/O performance.