V.A. Raikhlin, R.Sh. Minjazev
Multiclusterization is considered as a way to revolutionary increase of the parallel databases` scalability and failure-resistance, oriented on analytical data processing. Episodic data update in a specially allotted time and a high proportion of complex queries such as select-project-join is peculiar to DBMS of conservative type. Transition to the multiclusterization is investigated by the example of DBMS cluster-component usage. But conclusions must be sufficiently general, first of all for mono-cluster regime and for identified serious limitations of scalability with conservative databases and temporary dominance phase of the barrier synchronization of the initial processing of data read from the disk. The latter conclusion obliges the developer to focus on the organization of an asynchronous distributed database systems.
For major also can be attributed next results.
The scalability of distributed databases of the conservative type increased many times during the transition to multiclusteriza-tion.
Routerization with the definite value of the parameter distribution is reasonable
Conclusion about the preferred choice of a symmetric architecture underlying DBMS Clusterix.