Transactional Massive-Parallel DBMSs: A New Wave.
A possibility to build unlimitedly scalable cluster-based systems has lead to strong activation of research and development of "shared-nothing" architectures of data management systems. Two camps has been established: "NoSQL" that refutes the main principles related with DBMSs and "one size doesn’t fit all" that emphasizes needs of systems’ specialization saving the most important features of DBMS. The most interesting seems to be a confrontation of these camps in the area of "transactional" data management systems. Based on the CAP-"theorem" of Eric Bruwer, representatives of the camp of NoSQL declines to support traditional features of ACID of database transactions. This paper discusses the essence of the Bruwer’s "theorem" and proves that this theorem does not any relation with the ACID features. The paper also overviews the most interesting modern research projects provided classic ACID-transactions in parallel shared-nothing environments and the soundest approaches that partially relaxes requirements of ACID by purely pragmatic reasons (but not at all in relation with the CAP "theorem").
Proceedings of the Institute for System Programming, vol. 20, 2011, pp. 189-251.
ISSN 2220-6426 (Online), ISSN 2079-8156 (Print).Full text of the paper in pdf (in Russian) Back to the contents of the volume