The Bridge - Issue 3, 2021 - 17

Feature
Analyzing Service Mesh Performance
do not have the tooling necessary for or refuse to
publish performance data. Such tests can be arduous
to create and sustain a capable harness, a point-intime
consideration (none of the elements under
measurement are static), and misinterpreted. These
challenges are considered in the next sections.
III. Service Mesh Perfomance Considerations
As the software-defined networking layer of microservices,
service mesh encompasses multiple aspects of critical
functions for applications, such as circuit breaking,
health checks, and packet operations. Analyzing the
permutations of these configurations is an impossible
task without a suitable test harness. A service mesh
management plane can be such a tool. As the multimesh
manager, Meshery is capable of provisioning ten
different service meshes, workloads atop the meshes,
generating load using Nighthawk and analyzing that
load. No other tool capable of performing these tasks
end-to-end exists. Meshery is a Cloud Native Computing
Foundation project originally created by Layer5
[4]. Nighthawk is a Layer 7 (HTTP/HTTPS/HTTP2)
performance characterization tool created by Envoy
community [5].
A. How Are You Measuring?
Consider the simple set of steps to execute performance
tests in a simple Kubernetes-based cluster:
1. Setup your cluster, service mesh, and application
under test.
2. Pick a benchmarking tool that can measure Layer 4
or Layer 7 performance.
3. Configure your test setup for performance, doing
so in context of other constraints that you might
need to uphold (e.g. resiliency characteristics of your
service deployment).
4. Choose the protocol of interest such as HTTP,
HTTPS, HTTP1/2, gRPC, or NATS.
5. Identify key performance indicators (KPIs) of
interest - Transactions per second (TPS) or percentile
latencies, etc.
6. Decide on the test duration such as 60 seconds,
5 minutes, or 1 hour.
7. Choose the number of requests per second (RPS).
8. Execute the test.
9. Mark down requests per second, latencies,
throughput, and any other output provided by
benchmarking tools.
Note that Kubernetes is an open-source platform for
managing cloud-native applications.
B. What are you measuring?
Performance of a service mesh can be described
across multiple dimensions covering some or all of
these core functionalities of a service mesh. So, which
dimensions are the linchpins of performance? Which
metrics are key indicators of performance? Outside of
the different types of performance tests, performance
management concerns include the need for performance
and overhead data under a permutation of different
workloads (applications) and different types and sizes of
infrastructure resources. Hence, it is crucial to understand
what is being measured in a service-mesh-based
deployment. Certain critical considerations are missing
from the simple methodology previously described.
Figure 1 gives examples using the Service Mesh
Performance (SMP) standard which is a specification
developed by within the cloud computing community [6].
Ultimately, the goal of any performance tests is to ensure
repeatable measurements and obtain consistent results
across multiple test runs. A list of test considerations
includes the following.
TRAFFIC CONSIDERATIONS
* East-West traffic
- between two pods within the same or two
different Virtual Machines (VM).
- between two pods within the same or two
different bare metal nodes.
- combination of above with choice of user-space or
kernel-space networking stack on the host node.
* North-South traffic
- Throughput and latency of traffic flowing in and out
of a single VM or across a single bare metal node.
DEPLOYMENT CONSIDERATIONS
* Number of hops between traffic source and traffic
destination with load balancers, API gateways, ingress
controllers, security components such as firewall,
deep packet inspectors, and so on.
* Operating system settings.
* Hardware settings such as BIOS options, power
management features, NUMA awareness, platform
resource management, hardware accelerators, and
so on.
HKN.ORG
17
http://www.HKN.ORG
The Bridge - Issue 3, 2021

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