TRADES OFF BETWEEN SECURITY AND PERFORMANCE IN CYBER SYSTEMS

Review-2 Report

Bachelor of Technology in

Computer Science and Engineering

(Cyber Security)

by

Manchali Shashank (23951A62D2)

B Vishnu Sai Charan (23951A62J0)

Abdul Wasie (23951A6201)

Computer Science and Engineering (Cyber Security)

INSTITUTE OF AERONAUTICAL ENGINEERING

(Autonomous)

Dundigal, Hyderabad - 500 043, Telangana

October 2025
TRADES OFF BETWEEN SECURITY AND PERFORMANCE IN CYBER SYSTEMS

Mentor Name: Dr. Ranjith Kumar Gatla

Designation: Associate Professor

Team Details:
Team Leader: Manchali Shashank -23951A62D2
Team Member: B Vishnu Sai Charan -23951A62J0
Team Member: Abdul Wasie -23951A6201

I
 DECLARATION

I certify that

a. The work contained in this report is original and has been done by me under the guidance of my
supervisor(s).

b. The work has been submitted to any other Institute for any degree or diploma.

c. I have followed the guidelines provided by the Institute in preparing the report.

d. I have conformed to the norms and guidelines given in the Ethical Code of the Conduct.

e. Whenever materials such as data, theoretical concepts, figures, or text from other sources were used in
this project, proper credit has been given through in-text citations and detailed references. All external
sources have been duly acknowledged to maintain academic integrity. Permissions were obtained from
copyright holders wherever required. This ensures originality and ethical compliance throughout the
project.

Place: Signature of the Students

Date: Roll. No.: 23951A62D2

23951A62J0
23951A6201

II
 CERTIFICATE

This is to certify that the project report entitled Trades Off Between Security and Performance in
Cyber Systems submitted by Manchali Shashank, B Vishnu Sai Charan and Abdul Wasie to the
Institute of Aeronautical Engineering, Hyderabad in partial fulfillment of the requirements for the
award of the Degree Bachelor of Technology in CSE (Cyber Security) is a Bonafide record of work
carried out by him/her under my/our guidance and supervision. The contents of this report, in full or
in parts, have not been submitted to any other Institute for the award of any Degree.

Supervisor Head of the Department

(Dr. Ranjith Kumar Gatla) (Dr. K Rajendra Prasad)

Date:

III
 APPROVAL SHEET

This project report entitled Trades Off Between Security and Performance in Cyber Systems by
Manchali Shashank, B Vishnu Sai Charan and Abdul Wasie is approved for the award of the
Degree Bachelor of Technology in CSE (Cyber Security).

Examiners Supervisor(s)
Dr. Ranjith Kumar Gatla

Principal
Dr. L V Narasimha Prasad

Date:
Place:

IV
 ACKNOWLEDGEMENT

The satisfaction that accompanies the successful completion of any task would be incomplete without
the mention of the people who made it possible and whose constants guidance and encouragement
crown all the efforts with success I thank our college management and Sri M. Rajashekhar Reddy,
Chairman, for providing us the necessary infrastructure to carry the project work.

We express the sincere thanks to Dr. L V Narasimha Prasad, beloved Principal who has been a great
source of information for our work and to Dr. K Rajendra Prasad, Professor and Head, Department
of CSE (Cyber Security), for extending his support to carry out the project work.

We are especially thankful to our internal supervisor to Dr. Ranjith Kumar Gatla, Associate
Professor, Department of CSE (Cyber Security), for the project into successful one.

We take this opportunity to express our thanks to one and all who directly or indirectly helped me
in bringing this effort to present form.

V
 CONTENTS

I. Abstract 1

II. Introduction 2

III. Literature Survey 3

IV. Problem Statements 4

V. Objectives 5

VI. Existing System 6

VII. Proposed System 7

VIII. Methodology 8

IX. Expected Results 9

X. References 10
 ABSTRACT

This project investigates the critical trade-off between security and performance in modern cyber
systems. Implementing robust security measures, such as encryption and deep packet inspection,
introduces computational overhead and latency, negatively impacting system performance
(throughput, response time). Conversely, prioritizing performance increases system vulnerability. The
primary goal is to quantify this inverse relationship and propose an optimal, adaptive framework that
dynamically balances security efficacy with required performance.

Tools utilized in the methodology will include Wireshark for network analysis and latency
measurement, Apache JMeter for load and stress testing to simulate traffic and quantify performance
degradation, and a Python/MATLAB environment for developing the mathematical modeling and the
proposed adaptive security control framework. The expected result is a set of data-driven guidelines
and a dynamic mechanism to achieve a justifiable and measurable compromise.

Keywords: Trade-Off between security and performance, adaptive frameworks, Wireshark, Apache
JMeter, Python/MATLAB, data-driven guidelines, Dynamic mechanism, Adaptive security.

1
 Introduction

The current landscape of information technology is built upon interconnected cyber systems, encompassing
everything from large-scale Cloud Computing Platforms (CCP) and critical national infrastructure to resource-
limited Internet of Things (IoT) devices. These systems operate under a stringent dual mandate: they must
deliver high-speed, reliable service (Performance) while maintaining absolute protection against malicious actors
(Security).Performance Metrics dictate the system's effectiveness and User Experience (UX), focusing on
qualities like throughput (data processed per unit time), latency (response delay), and system utilization.
Security Metrics dictate the system's resilience, focusing on confidentiality (encryption), integrity (data
verification), and availability (defense against Denial-of-Service attacks).While both qualities are crucial for
system success, they inherently consume the same finite resources (CPU cycles, memory, bandwidth), setting the
stage for an inevitable trade-off.

The central dilemma for system architects is that strengthening security directly imposes a measurable cost on
performance. This relationship is not trivial; it forms the Problem Context for this project:
Security Overhead: Implementing robust security mechanisms, such as switching from a weaker AES-128
encryption to a stronger AES-256, or enabling full Deep Packet Inspection (DPI) in a firewall, requires
significantly more computational power and processing time. This overhead translates directly into increased
latency and reduced throughput.
The Static Dilemma: Most organizations currently rely on a static security posture, setting controls based on
the maximum perceived threat. This approach forces the system to operate under heavy, often unnecessary
performance degradation even during periods of low threat or low load.
The Need for Optimization: Prioritizing performance by using minimal security exposes the system to
catastrophic breaches. Conversely, prioritizing maximum security compromises operational efficiency. The lack
of an adaptive, quantitative model to manage this inverse relationship prevents systems from achieving a true
optimal balance—i.e., maintaining a minimum required security level with maximum attainable performance at
any given moment.

2
Literature Survey

3
 Problem Statement

Cyber systems, particularly those encompassing the Internet of Things (IoT), Industrial Control Systems (ICS), or
other resource-constrained networks, face a fundamental conflict between Security and Performance. While
modern Machine Learning (ML) and Deep Learning (DL) models (e.g., CNN, LSTM) achieve high security (e.g.,
100% detection accuracy for intrusions), their computational complexity, high processor overheads, and energy
consumption significantly degrade system performance. The key challenge is that existing high-accuracy
Intrusion Detection Systems (IDS) are not scalable or efficient enough for real-time deployment on devices with
limited battery, CPU, and bandwidth, leading to either poor performance or a compromise in security.

4
 Objectives

The primary objective is to develop and evaluate a novel IDS approach that optimizes the security-performance
trade-off.

O1: Develop an Optimized Model: To design a novel or hybrid ML/DL model architecture for NIDS that
maintains high security metrics (Detection Accuracy/Rate, low False Positive Rate) while explicitly reducing
computational complexity.

O2: Quantify the Trade-off: To define and implement a Trade-off Metric that mathematically combines security
(e.g., accuracy) and performance (e.g., CPU utilization, latency, energy consumption).

O3: Comparative Evaluation: To validate the proposed system against established, state-of-the-art models (like
standalone CNN/LSTM/SVM) to demonstrate an improved trade-off balance, achieving comparable detection
accuracy with significantly lower resource overhead.

5
 Existing System

The current state-of-the-art, as highlighted by the base papers, exhibits limitations that necessitate a trade-off:

Resource Intensiveness of High-Security Models: Advanced models like Convolutional Neural Networks
(CNN) and Long Short-Term Memory (LSTM), while achieving high detection accuracy (up to 100%), are
complex and require significant processing resources, leading to performance bottlenecks.

Performance Metrics are Overlooked: Many previous studies focus primarily on detection accuracy and neglect
critical performance metrics like CPU/processor overhead, energy consumption, and real-time latency, which are
essential for practical deployment, especially in resource-constrained IoT systems.

6
 Proposed System

The proposed system will be a Lightweight Hybrid Intrusion Detection System (LH-IDS) designed for
optimized performance on constrained devices while preserving detection accuracy.

Hybrid Architecture: Combine a powerful but resource-intensive DL component (for high accuracy) with a
low-overhead ML component (for initial traffic filtering/feature reduction). This could involve an efficient
Feature Selection layer followed by a specialized, pruned neural network or a hybrid ML classifier.

Dynamic Resource Allocation: The system will include a mechanism for dynamic model selection or
resource-aware classification, potentially using a simple, fast classifier (e.g., optimized KNN/SVM) for
normal traffic and activating the complex DL model only for suspicious traffic batches.

Targeted Optimization: The final model should be explicitly optimized for speed (low processing
time/latency) and low memory footprint, not just accuracy.

7
 Methodology

1) Dataset Acquisition and Pre-processing: Use a relevant, modern network intrusion dataset (e.g., CIC-
IDS2017, Bot-IoT, or UNSW-NB15) which reflects current threat types and network dynamics.

2) Efficient Feature Engineering: Employ Principal Component Analysis (PCA) or a similar

dimensionality reduction technique to drastically cut the number of features, prioritizing those that
maintain high discriminative power against attacks while reducing processing time.

3) Model Development (LH-IDS): Implement the proposed hybrid model architecture, focusing on a
compact design (fewer layers, fewer neurons) compared to the standard high-accuracy models found in
the literature.

4) Trade-off Evaluation and Quantification:

▪ Security Metrics: Measure Detection Accuracy (ACC), True Positive Rate (TPR), and False Positive
Rate (FPR).
▪ Performance Metrics: Measure CPU Utilization/Overhead (%), Energy Consumption
(J/packet), and Average Detection Latency (ms).

8
 Expected Result

The project is expected to yield the following results:

1. Improved Trade-off Point: A validated LH-IDS model that achieves a significantly better
security-performance trade-off score ($S$) than existing models.

2. Quantifiable Efficiency Gain: A demonstration that the LH-IDS reduces the computational
overhead (CPU/latency) by a specific percentage (e.g., 30–50%) compared to the
highest-accuracy benchmark model (e.g., CNN/LSTM) while maintaining an acceptable
detection accuracy (e.g., >95%).

3. Deployability: Proof of concept showing that the LH-IDS model is small and fast enough to
be deployed on a simulated or real resource-constrained device (e.g., a Raspberry Pi or IoT
gateway).

9
 References

1) Irfanullah Khan, Junaid Khan Shah Hussain Bangash, Muhammad Asad Iftikhar, Khalid
Hameed, and Waqas Ahmad. "Intrusion Detection Using Machine Learning and Deep
Learning Models on Cyber Security Attacks." VFAST Transactions on Software Engineering,
Volume 12, Number 2, April-June 2024, pp:95-113.

2) Junaid Arshad, Muhammad Ajmal Azad, Roohi Amad, Khaled Salah, Mamoun Alazab, and
Razi Iqbal. "A Review of Performance, Energy and Privacy of Intrusion Detection Systems
for IoT." Electronics, Volume 9, Number 4, 629, April 2020.

10