The present web page provides an overview of techniques involved to guarantee a maximum service availability. A high service availability is required for branch offices and regional/central site, as today's business cannot be run durably without access to information systems.

In regional and central sites, failure resistance is a must as hosted applications are vital to run business. Two routers are used and run in parallel. It offers a de-facto increased availability as:

• Hardware is duplicated; since the repair time remains significantly shorter than the device Mean Time Between Failure (MTBF), a network outage has a little chance to be originated from a hardware failure.
• Two different uplinks are used. Thus, uplinks can use different transport networks.
• The service is not interrupted when a maintenance operation is done

The Virtual Redundant Router Protocol (VRRP) permits stations on the LAN to view a single virtual router that is actually made up of a master and idle routers. In some old installation, VRRP is replaced by IRDP; the ICMP Router Discovery Protocol (IRDP) is used to advertise the default router to the LAN stations.

When using two redundant routers, two network access uplinks are available. Load sharing permits the use of both uplinks to fully optimise the access infrastructure. LAN stations send packets destined to the WAN towards the master virtual router. This router shares bandwidth load: half of the flows are redirected to the slave router (which in turn forwards packets to the WAN) and the remaining half is sent on the master router uplink.

To further increase failure resilience, applications can be hosted on a virtual server, which is actually made up of several computers. The OneOs router forwards incoming requests for the virtual server to a physical server.The forwarding is stateful, i.e. for new incoming packets, the session is detected and if recognized, a packet from a given session is always forwarded to the same server. A weighted session load balancing is done on the servers and the OneOs detects faulty servers to prevent the forwarding of packets to unavailable servers.

In branch-offices, the cost of dual uplinks and redundant routers is usually too high. A backup access such as ISDN or PSTN uplinks is used to backup the main uplink interface. A classical approach is to monitor the uplink status to determine that the main access is operational. As long as it remains in this state, the backup connection is not used. The weakness of this method though is that it assumes that the entire network is never faulty. Actually, there are multiple points of failure along the path from the branch-office up to the central site(s): the access network, LNS, backbone/routing, central site routers… may encounter a defect. OneAccess has implemented a monitoring technique that makes sure that an end-to-end connection between the central site and the branch-office exists. If the path is faulty, then the backup access is used.