Performance evaluation and modeling of remote execution and caching cluster
Ladda ner
Typ
Examensarbete för masterexamen
Program
Publicerad
2021
Författare
AGUILAR ROMERO, MARÍA FERNANDA
Modellbyggare
Tidskriftstitel
ISSN
Volymtitel
Utgivare
Sammanfattning
The scope and application of distributed systems are an increasing trend by the
need for high-performance infrastructure. Many multi-branch organizations adopt
the resource-sharing approach to maximize the system’s utilization, especially in the
software developing industry. Computational techniques such as caching and clustering
are suitable to boost the performance of computer systems at any architectural
level. This project presents a performance evaluation, modeling, and analysis of a
system inspired by the case study of the industrial partner, Veoneer. The case study
is used for software development and implements Bazel, a suitable tool for multibranch
cooperation through remote caching and remote execution. Here, the remote
cache allows storage and sharing the outputs from all the compilations among the
branches, thus preventing the re-execution of tasks and reducing the response time
for developers. On the other hand, remote execution benefits from the computational
resources of different servers across the branches. However, the benefits are
at the cost of complexity, and systems start to degrade in performance, such as
long response times revealed from the system’s monitoring. Therefore, performance
evaluation is essential for planning and improvement.
This work uses operational analysis, queueing networks, and the universal scalability
law to evaluate, model, and predict the system’s performance, respectively. The
aim is to establish algebraic relations between the system’s factors and metrics to
understand the impact of workload and system capacity on performance to predict
its behavior. The results expose that the cache is over-utilized and benefits from
horizontal scalability. On the other hand, the study reveals that a cluster benefits
from vertical scalability but still has its scalability bounds to prevent that multithreading
effects (i.e. coherence, concurrency, and contention) affect performance
on parallel executions.
Beskrivning
Ämne/nyckelord
Performance evaluation , performance modeling , remote cache , remote execution , queueing modeling , scalability