Back-Tracing over Distributed Persistent Storage Without Transactions

Orthogonally persistent automatic storage management provides a useful abstraction over stable secondary storage, in contrast to manual storage management which is complex, error-prone and tedious. The provision of the persistence abstraction over a cluster of storage nodes provides this abstraction in conjunction with mainframe-level aggregate IO performance, necessary for the support of high performance physical simulation algorithms. A distributed persistent garbage collector should exhibit the properties of safety, completeness and scalability in the face of distributed cycles of garbage. Of the few extant distributed secondary storage garbage collectors, even fewer still exhibit any of these properties beyond safety. The aim of this research is to make detailed a comparison between two collectors believed to be both safe complete and scalable; to make such a comparison requires that both collectors be implemented in the same context. This paper describes the transformation of an existing distributed garbage collector based on backtracing and transactions to a form where it does not depend on transactions for safety or completeness, without adversely affecting safety or com-