Proceedings of the 29th Annual Hawaii International Conference on System Sciences- 1996 Distributed Persistent Object System with Uniform Representation of Pointers and Its Garbage Collection

In persistent object storage, one of the most influen-tial overheads at application runtime is caused by the conversion of pointers for transparency among vari-ously represented pointers. Em’sting systems require some representations of pointers, which causes conver-sion of pointers and runtime over...

Full description

Bibliographic Details
Main Authors: Yamamoto Kouhei, Inohara Shigekazu, Miyazawa Hajime, Uehara Keitaro Hara Masahiro
Other Authors: The Pennsylvania State University CiteSeerX Archives
Format: Text
Language:English
Subjects:
Online Access:http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.100.835
http://csdl.computer.org/comp/proceedings/hicss/1996/7324/00/73240012.pdf
Description
Summary:In persistent object storage, one of the most influen-tial overheads at application runtime is caused by the conversion of pointers for transparency among vari-ously represented pointers. Em’sting systems require some representations of pointers, which causes conver-sion of pointers and runtime overheads, because of the limitation of 3%bit processors ’ virtual address space, and efficient garbage collection (GC) of distributed persistent objects. In order to remove overheads of pointer conversion, we propose to use an Indirect Back Pointer(IBP) for a distributed persistent object storage on recent 64-bit processors. IBP is an auxiliary data structure for part-by-part compacting GC. Virtual spaces of 64-bit pro-cessors are so large that systems can exploit the linear mapping of the storage and uniform representation of pointers that do not need the conversion of pointers. IBP maintains a list of inter-partition pointers that is sufficient for part-by-part compacting GC. Because IBP is separated from the pointers that are used by applications, IBP and uniform representation of point-ers are compatible. In this way, the proposed system achieves compacting GC and no overheads for conver-sion of pointers at application runtime. 1