V.А. Kozmidiady

Principles of construction of the centralized depositories in which it is possible to keep some tens of petabytes of information are examined in the article. Requirements to such depository: - a maximal volume of the kept information can be about 10 PB. The easy scaled of volume must be well-to-do ot the hundreds of TB to 10 PB; - access to the depository must be carried out both through a network and locally; - access time must be not worse, than at the existent systems of storage. Possibly, it is possible to assume sublogarithmic growth of this time with depending on a volume; - high fault-tolerance must provide work in the mode 24/7/365 (24 hours a day, 7 days a week, 365 days a year). Failures and refusals must not result in the loss of data; - a cost of storage of data in a depository must be of the same order, as the cost of storage on an ordinary hard disk. Difficulties of storage of data of such volume are examined in the distributed file systems. For achieving the indicated volumes it is suggested in a number of cases to give up the observance of standard of POSIX. Basic deviations followings from POSIX: - refusal of directory tree of the file system and passing to the flat model, i.e. refusal of the file system hierarchy. Refusal of support of hard and symbolic links; - refusal of file name and passing to the identifier of data set. This identifier is constructed at creation of data set and returns an user; - assumption, that different users in the same moment of time can see different content; - limitation of nomenclature of functions, workings with a depository. Exemplary composition: creation of data set with pointing of his content, reading of data set, elimination of data set, modification of data set by replacement of content. In addition, it is suggested to provide reliable and trouble-free storage of data due to maintenance for every set of information a few copies. In the article principles of work of the distributed depository of information and operation of access are offered to the depository. The special attention is spared providing of depository fault-tolerance, i.e. to the conduct of depository at the fall of nodes and their recovery.