Addressing Large Distributed Collections of Persistent Object: The
J. Eliot B. Moss · 1989
THE MNEME PERSISTENT OBJECT STORE PROJECT HAS AS ONE OF ITS PRIMARY GOALS TO SUPPORT COOPERATIVE, CONCURRENT, AND RELIABLE USE OF LARGE, DIS- TRIBUTED COLLECTIONS OF OBJECTS. IN OUR CASE, DISTRIBUTION IS INTENDED TO MEAN NOT ONLY PHYSICAL DISPERSION, BUT ALSO SOME AUTONOMY OR INDEPENDENCE OF SUBCOLLECTIONS OF OBJECTS---THAT IS, THE OBJECT SPACE IS UNDER SOME DEGREE OF DECENTRALIZED MANAGEMENT. PROVIDING INDEPENDENCE OF SUBCOL- LECTIONS OF OBJECTS HAS INTERESTING IMPLICATIONS AS TO HOW TO ADDRESS AND RETRIEVE OBJECTS EFFICIENTLY. HERE WE DESCRIBE THE NEED FOR AND BENEFITS OF INDEPENDENT SUBCOLLECTIONS OF OBJECTS, EXAMINE THE ADDRESSING IMPLICA- TIONS, AND SHOW HOW THESE IMPLICATIONS HAVE AFFECTED THE DESIGN OF MNEME. MOST PARTICULARLY, WE ARGUE AGAINST THE APPROPRIATENESS OF A LARGE FLAT STORE OF BYTES OR EVEN OF OBJECT IDENTIFIERS, AND IN FAVOR OR RICHER, MORE FLEXIBLE, STRUCTURES. WE ALSO CONTEND THAT COMPARABLE PERFORMANCE CAN BE ACHIEVED BY CAREFUL IMPLEMENTATION OF THE RICHER STRUCTURES, WITH CONSIDER- ABLY MORE FUNCTIONALITY AND FLEXIBILITY THAN FLAT STRUCTURES PROVIDE. THE COST IS INCREASED COMPLEXITY OF THE SUPPORTING SOFTWARE.