The Bridge - Issue 3, 2021 - 16

Feature
Analyzing Service Mesh Performance
By Lee Calcote, Layer5; Mrittika Ganguli, Intel; Sunku Ranganath, Intel; Otto Van Der Schaaf, Red Hat
I. Introduction
As a forthcoming, ubiquitous layer of cloud native
infrastructure, service meshes offer deep and uniform
control and visibility into the topology and state of
ephemeral microservices. Managing the myriad
configurations of cloud native infrastructure is greatly
facilitated by a service mesh, but succinctly summarizing
and characterizing the performance of your service
mesh in context of your unique workloads and your
infrastructure of choice is a challenge unto its own [1-3].
We explore how to model your service mesh topology
and optimize for your ideal configuration in context
of how much you value properties of resiliency,
performance, throughput, latency, and so on before
you deploy to production. Readers will understand how
distributed performance analysis offers unique insights
on the behavior of microservices and their efficiency
of operation, see examples of how common types of
workloads perform under specific service mesh functions,
and be empowered with analytical tooling that can be
used to make optimized configurations.
We provide core, memory and I/O combinations based
on workload needs with insights into workload analysis
which can influence the efficiency of the service mesh
and overall performance of the cluster.
THE BRIDGE
II. Characterizing the Complexity of
Combinatorial Analysis
Consider that the more value you try to derive from
your service mesh, the more work that you will ask it to
do. Said another way, an analysis of the architecture of
a service mesh - with its distributed proxies - and the
functionality it offers will lead to the question, " What
overhead is running my service mesh incurring? " This is
not an easy question to answer as the permutations of
configuration between your infrastructure, service mesh,
and applications are innumerable and any change to one
of them affects their collective performance.
How would you describe the performance of your
service mesh and that of your clusters and their
workloads? Are you imagining a wall of line charts with
metrics capturing golden signals? The act of articulating
the performance of your service mesh requires you to
characterize the state of your systems and the overhead
incurred by your infrastructure and to consider what this
means to your users.
Moreover, when performance is characterized,
the analysis is subjective to the specific workload,
infrastructure, and instruments used for measurement.
Given the variety of this measurement challenge,
the providers of most service meshes and their data
plane proxies (if there is a third-party component)

https://hkn.ieee.org/

The Bridge - Issue 3, 2021

Table of Contents for the Digital Edition of The Bridge - Issue 3, 2021