marina@ai.toronto.edu (Marina Haloulos) (10/30/89)
Department of Computer Science, University of Toronto
(GB = Gailbraith Building, 35 St. George Street)
-------------------------------------------------------------
SYSTEMS SEMINAR
GB244, at 2:00 p.m., Tuesday 7 November 1989
Professor Nick Roussopoulos
University of Maryland
"Incremental Access Methods for Databases"
A new class of _I_n_c_r_e_m_e_n_t_a_l _A_c_c_e_s_s _M_e_t_h_o_d_s (_I_A_M) will be presented. IAMs
retain knowledge from prior database searches in an integrated and
persistent _c_a_c_h_e _m_e_m_o_r_y of access paths. The access paths point directly
to the qualifying records and, therefore, subsequent materialization of
records from memory caches requires no search. IAMs maintain and/or
augment the access paths using incremental algorithms operating on the
differential files of the input operands. Operating under appropriate
update strategies, IAMs achieve performances that are far above those of
today's re-execution based methods.
In this talk, I will outline the principles of IAMs, their materialization
from the cache memory, and a cost model for the incremental algebra. I
will then discuss three implementations of IAMs. The first supports views
of relational and deductive databases. The second utilizes incremental
algorithms for maintaining replicated data in a distributed architecture of
database servers and workstations connected via gateways. The last one
uses an IAM engine to support a very efficient Prolog-DBMS interface.