Page 57 - Sirius_Annual_Report_2021

Shiji Bijo
Shiji Bijo
Main research findings:
Multicore architectures aim to improve execution speed of software through parallel computations. Large-scale multicore systems have massively parallel hardware architectures. The develop- ment of parallel programs which can exploit the multicore architecture is
a non-trivial task for the
Temitope Ajileye
Temitope Ajileye
Abstract: Many RDF systems support reasoning with Datalog rules via materiali- sation, where all conclusions of RDF data and the rules are precomputed and explicitly stored in a preprocessing step. As the amount of RDF data used in applications keeps increasing, processing large datasets often requires distributing the data in a
software industry. Analysing the effect of different archi-
tectures during software development helps to uncover
cases that may affect the performance. This thesis studies
at the theoretical level how the underlying architecture and
data movements between cores and memory systems may
influence the program performance. The main contribution
of this thesis is a detailed formal model of multicore archi-
tectures along with an associated proof-of-concept analysis
tool. One of the main challenges in characterising the core-
memory communication patterns in multicore systems is
the consistency of shared data in different memory levels. evaluate our materialisation algorithm against two state- The formalisation is used to guarantee consistency of shared of-the-art distributed Datalog systems and show that our data for all architectures that can be expressed in our formal technique offers competitive performance, particularly when
cluster of shared-nothing servers. While numerous distri- buted query answering techniques are known, distributed materialisation is less well understood. In this paper, we present several techniques that facilitate scalable materia- lisation in distributed RDF systems. First, we present a new distributed materialisation algorithm that aims to minimise communication and synchronisation in the cluster. Second, we present two new algorithms for partitioning RDF data, both of which aim to produce tightly connected partitions, but without loading complete datasets into memory. We
model. The formal model captures interactions between
cores and memory and the tool provides a model-based
simulation environment to examine the effect of different
multicore architectures on performance during parallel or superior to the state of the art min-cut partitioning, but executions. computing the partitions requires considerably less time
the rules are complex. Moreover, we analyse in depth the effects of data partitioning on reasoning performance and show that our techniques offer performance comparable
and memory.
SIRIUS ANNUAL REPORT 2021 | 57