Experience with the PerDiS Large-Scale Data-Sharing Middleware
- Cite this paper as:
- Shapiro M., Ferreira P., Richer N. (2001) Experience with the PerDiS Large-Scale Data-Sharing Middleware. In: Kirby G.N.C., Dearle A., Sjøberg D.I.K. (eds) Persistent Object Systems: Design, Implementation, and Use. Lecture Notes in Computer Science, vol 2135. Springer, Berlin, Heidelberg
PerDiS is a distributed persistent middleware platform, intended to ease the distributed sharing of long-term data. Its users belong to geographically-distant and non-trusting entreprises. It targets CAD applications for the building industry: data sets are large and pointer-rich; simultaneous reads and updates are supported; there is no central database; migrating legacy applications is accessible to unskilled programmers. A number of real applications have been either ported or written specifically for PerDiS.
The following design decisions were essential to the scalability and to the useability of PerDiS. Isolation (transactions) decouples users from one another. The system design provides different granularities. The programming abstraction is a fine-grain, persistent, isolated shared memory, with objects, invocations and URLs. The system mechanisms are coarsegrained, loosely-coupled and optimistic. Fine-grain application entities are encapsulated into coarse-grain system entities, respectively clusters, domains, transactions, and projects.
Unable to display preview. Download preview PDF.