Peer Fusion Clusters Have Great Performance.

Peer Fusion clusters multiply the performance of one peer by the count of peers in the cluster:

• Cluster Performance.

A file stored on a Peer Fusion cluster is striped across every peer in the cluster. A ten-gigabyte file stored on a ten-peer cluster will be divided into ten one-gigabyte stripe files. I/O requests are simultaneously received by all the peers and only the peers relevant to the I/O footprint will participate. The peers that are not participating in a given I/O request are free to participate in other simultaneous requests. All the peers know mathematically which peer stores what file blocks so there is no added communication required. This is a key reason for the scale-out performance of the system: the overhead increase is negligible as more peers are added to the cluster.

In a Peer Fusion cluster each new peer adds its I/O, network and computational bandwidth so that data can move faster into and out-of the cluster. The peers will do half the work when a file is striped across a twenty-peer cluster vs a ten-peer cluster.



• Peer Performance.

Peer Fusion peers stripe data across their drives without the controller computing a checksum (RAID 0). This results in a higher sustained I/O throughput and better fault-tolerance because the cluster ensures resiliency. The peers use as many NICs as are available through the Peer Fusion proprietary network protocol ensuring reliable and efficient many-to-many messaging.



• Cluster Healing Performance.

Healing a cluster involves all the peers in the cluster working together to regenerate missing data. Healing is done on-the-fly as long as a quorum of peers remain available. The process of healing is driven by walking the file system tree so that only disk blocks corresponding to missing file blocks are repaired. The healing computations are distributed equally across all the available peers which increases performance. The Peer Fusion proprietary codec performing the computations is multi-threaded and processes multiple gigabytes of data per second. The time required to heal one peer is essentially the same time required to heal simultaneously any number of peers due to the Peer Fusion cluster architecture.