Fault tolerance

A system can be affected by outages of all the appliances at the same site.

With the right configuration, the System can tolerate such outages without losing availability at all.

To guarantee write availability when a site outage occurs, the fraction of nodes at any one site should not exceed the following equation:



( IDAWidth - WriteThreshold ) / IDAWidth

To guarantee read availability, the proportion of nodes at a site should be less than this equation:



( WriteThreshold - IDAThreshold ) / IDAWidth

By following these guidelines, even entire site failures do not impact access to data. It leaves site correlated failures as the only concern. To mitigate the risk or impact of correlated site failures, several things can be attempted. You can select sites that get power from different locations or live on different power grids. Sites should have multi-homed connections, to different internet backbones. The geographic distance that separates sites should also be sufficient to avoid issues that could relate to the same natural disaster. By following these steps, the chance of your data being affected by a correlated failure can be substantially reduced.

The use of Vault Profiles represents a departure from the pre-Concentrated Dispersal method for configuring vaults. Due to the various subtleties of Concentrated Dispersal systems, users no longer directly configure IDA Width, Threshold, and Write Threshold. Instead, the IDA configuration is selected based on the number of Slicestors, the type, the number of sites, mirror settings, and the user selected Vault Optimization.