Performance Required During a Rebuild from a Failure is impacted only by read rebuild operations. Unlike other storage systems, writing performance is unaffected by failures and rebuilds, since Weka systems continue writing to functioning backends in the cluster. However, read performance can be affected, because the reading of data from a failed component has to be performed from the whole stripe. This requires a simultaneous operation and an instant priority rebuild for data read operations. If, for example, one failure occurs in a cluster of 100 backends, performance will be affected by 1%; however, in a cluster of 100 backends with a stripe width of 16, performance will be reduced by up to 16% at the beginning of the rebuild. Naturally, the cluster size can exceed the stripe width or the number of failure domains. Consequently, for large clusters, it is recommended that the stripe width does not exceed 25% of the cluster size, e.g., for a cluster of 40 backends, 8+2 protection is recommended so that if a failure occurs, the impact on performance will not exceed 25%.