@proceedings{APPLIED24,

title = {Tracing the Latencies of Ares: A DSM Case Study},

author = {Chryssis Georgiou and Nicolas Nicolaou and Andria Trigeorgi},

year  = {2024},

date = {2024-06-17},

urldate = {2024-06-17},

booktitle = {The Applied Workshop: Advanced tools, programming languages, and Platforms for Implementing and Evaluating algorithms for Distributed systems (ApPLIED24)},

abstract = {Distributed tracing is a method used to monitor applications by

tracking and visualizing requests as they move across various com-

ponents and services in a distributed system. Despite being widely

adopted in major cloud-computing applications, to the best of our

knowledge, distributed tracing has not been employed in Distributed

Shared Memory (DSM) emulations. In such emulations, typically, a

set of networked nodes (servers) maintain copies of the memory

data, and a set of clients (readers/writers) access the data by sending

messages to the servers. The main challenge in this environment

is to maintain the consistency of the data despite asynchrony and

failures. Traditionally, the latency of operations in DSM implemen-

tations has been evaluated through simple log-based strategies

providing a high-level performance analysis.

This paper introduces distributed tracing to DSM, in an attempt

to provide a fine-grain performance analysis, helping to identify

performance bottlenecks. To this respect, we use Ares as a case

study. Ares is a crash-tolerant DSM algorithm, providing atomic

consistency and supporting dynamic participation of networked

nodes. Our approach employs a set of tracing tools: Opentelemetry

for code instrumentation, Jaeger for telemetry data collection, and

Grafana for visualization.},

howpublished = {Proceedings},

keywords = {},

pubstate = {published},

tppubtype = {proceedings}

}