Appeared in the Proceedings of the Summer USENIX Conference, June 1991, pages 17-29 Management of Replicated Volume Lo cation Data in the Ficus Replicated File System y Thomas W. Page Jr., Richard G. Guy, John S. Heidemann, Gerald J. Popek, Wai Mak, and Dieter Rothmeier Department of Computer Science University of California Los Angeles fpage,guy,popek,johnh,waimak,
[email protected] Abstract Existing techniques to provide name transparency in distributed le systems have b een de- signed for mo dest scale systems, and do not readily extend to very large con gurations. This pap er details the approach whichisnow op erational in the Ficus replicated Unix ling envi- ronment, and shows how it di ers from other metho ds currently in use. The Ficus mechanism p ermits optimistic management of the volume lo cation data by exploiting the existing directory reconciliation algorithms which merge directory up dates made during network partition. 1 Intro duction Most Unix le system implementations separate the maintenance of name binding data into two levels: within a volume or lesystem, directory entries bind names to low-level identi ers such as ino des; a second mechanism is used to form a sup er-tree by \gluing" the set of volumes to each other. The traditional Unix volume sup er-tree connection mechanism has b een widely altered or replaced to supp ort b oth small and large scale distributed le systems. Examples of the former are Sun's Network File System NFS [13] and IBM's TCF [12]; larger scale le systems are exempli ed by AFS [7], Decorum [6], Co da [14], Sprite [9], and Ficus [2, 10].