Wednesday, February 22, 2006

Administration

A portion of the bandwidth used by any particular node in a P2P backup system has to be dedicated to administrative purposes. After a service level agreement is negotiated, a node must constantly ensure that his partner is available and that its data is available on that partner's shared storage. While the former is relatively negligible, ensuring the data is available could be rather expensive. The Samsara and/or Pastiche papers suggested using a computed hash of the data to ensure the data is stored on the host, but Lillibridge points out that this will not ensure the partnering node will provide the data when it's needed. The benefit, which is the unused bandwidth, gained from not supplying the data despite the steep cost of having to hold the data anyway seems rather small but still creates a potential impact to the reliability of the system.

Thus, the added administrative cost is the size of the data block that needs transferred, the bandwidth (time) used in performing the transfer, and the frequency with which this needs to occur for all partners of a paritcular node.

Presumbly, this cost is fixed, and hence assumed, for any service level agreement between two partners. It should, however, be chosen so as to minimize the cost of the administration of the partnership.

No comments: