⚡ electrical engineering

Optimal Control of Applications for Hybrid Cloud Services

This paper proposes models, optimization criteria, and control approaches to address processing bottlenecks and ensure Quality of Service for Big Data applications within hybrid cloud infrastructures.

Original authors: Evgeniy Pluzhnik, Evgeniy Nikulchev, Simon Payain

Published 2026-06-04

📖 4 min read☕ Coffee break read

CC BY 3.0

Original authors: Evgeniy Pluzhnik, Evgeniy Nikulchev, Simon Payain

Original paper licensed under CC BY 3.0 (http://creativecommons.org/licenses/by/3.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are running a massive library that holds millions of books, articles, and videos. Some of these books are kept in your own basement (your Private Cloud), while others are stored in a giant, rented warehouse across town (the Public Cloud). This setup is called a Hybrid Cloud.

The problem the authors of this paper are tackling is like trying to find a specific book in this split library system without causing a traffic jam. If too many people ask for books at once, the "hallways" (network connections) get clogged, and the delivery trucks (data packets) get stuck or lost.

Here is a simple breakdown of what they did and what they found:

1. The Problem: Too Many Requests, Not Enough Space

In the past, people tried to solve library organization problems using static math (like drawing a map of the library). But the authors say this doesn't work well for the cloud because the library is alive and changing.

The Dynamic Nature: Sometimes the basement is full, sometimes the rented warehouse is full. Sometimes the road to the warehouse is a highway, and sometimes it's a dirt path.
The Goal: They wanted to build a "smart traffic cop" system that watches these changes in real-time and directs traffic so everyone gets their book quickly.

2. The Solution: A Smart Control System with "Feedback"

The authors created a system that uses feedback loops. Think of this like a thermostat in your house.

How it works: The system constantly asks, "How busy are the servers right now?" and "How fast is the internet?"
The Math: They used a special type of math (differential equations) to predict how the system will behave. Instead of just guessing, they modeled the system like a moving vehicle that needs steering adjustments (control) to stay on the road.
The Criteria: They wanted to achieve two main things:
1. Speed: Get the data to the user as fast as possible.
2. Cost: Don't waste money on unnecessary resources.

3. The Traffic Cop: Cisco AVC

To manage the flow of data, they used a specific tool called Cisco AVC (Application Visibility and Control). Imagine this as a super-smart toll booth operator who can look at every car on the road and decide who gets to go first.

Sorting the Traffic: The system separates cars into different lanes based on what they are carrying:
- Lane A: Local traffic (basement to basement).
- Lane B: Public cloud traffic (basement to warehouse).
- Lane C: Everything else.
Prioritizing: Inside Lane B, it sorts further: "Web browsing" gets one speed, "Database requests" get another, and "File transfers" get another.
The Result: By dynamically adjusting the speed limits and lane widths, they stopped the "tailgating" (network congestion) that causes data to get dropped.

4. The Experiment: The "Articles" Library

To test this, the authors set up a real experiment using a database of scientific articles.

The Setup:
- Local Database: Stored the "metadata" (the title, the author's name, the date) in a standard server.
- Cloud Storage: Stored the actual heavy files (the full text of the articles, which could be up to 3MB each) in the Microsoft Azure cloud.
The Comparison: They compared how fast it was to get an article from a purely local system versus this new hybrid system.

5. The Findings

The Graphs: They showed a chart of network traffic. Without their smart system, the traffic looked like a wild rollercoaster with huge spikes (bottlenecks) and crashes (dropped data). With their system, the traffic smoothed out into a nice, steady line.
The Conclusion: The hybrid system worked great for handling large files. However, they found that old-school math (graph theory) isn't good enough to predict how fast these hybrid systems will run. You need the "dynamic" control system they built because the resources are constantly shifting.

In a Nutshell

The paper argues that managing a hybrid cloud is like managing a busy city with changing traffic patterns. You can't just use a static map; you need a live, smart traffic control system that constantly adjusts the flow based on real-time conditions. By doing this, they proved you can store huge amounts of data in the cloud while keeping access fast and reliable for the user.

Technical Summary: Optimal Control of Applications for Hybrid Cloud Services

Problem Statement

The rapid development of Big Data necessitates the migration of processing systems and data structures into hybrid cloud environments to ensure Quality of Service (QoS). However, traditional optimization methods relying on graph theory and discrete mathematics fail to provide accurate assessments for cloud technologies. These classical approaches cannot adequately model the dynamic nature of hybrid infrastructures, which involve virtualization, dynamic resource reallocation, and variable bandwidth communication channels. Consequently, there is a critical need for new tools to describe database management systems that account for the dynamic character of data exchange, where every block of an application must be evaluated at each execution stage to prevent bottlenecks.

Methodology

The authors propose a control system architecture based on dynamic modeling with feedback, approximated by difference equations with vector control. The methodology is structured around four key components:

System Architecture: The proposed architecture integrates a SaaS portal for personalized access, an autonomous control framework with optimization modules (including QoS planners), and a hybrid IaaS layer combining private and public cloud services. The specific use case focuses on scientific and educational resources (multimedia, video, text) within an educational institution.
Mathematical Modeling: The system is modeled using linear stationary difference equations:
$x(t+1) = Ax(t) + Bu(t)$
Where $x$ represents the $n$ -dimensional vector of system states (e.g., virtual machine workload) and $u$ represents the $m$ -dimensional vector of controls. The model incorporates feedback loops to monitor current system states and query completion status, though the authors note that excessive feedback can congest narrow communication channels, necessitating optimization.
Network Control Mechanism: To manage QoS, the study utilizes Cisco Application Visibility and Control (AVC) technology. The network is segmented into three levels: local servers/private cloud, network switching/routing, and public cloud. Traffic is separated in two stages:
- Stage 1: Separation by IP addresses and ports (Private cloud vs. Public cloud vs. Other).
- Stage 2: Separation by application type using NBAR2 and Flexible Netflow (Web, Database, File, and Service applications).
  Based on this classification, the system dynamically sets QoS priorities and allocates bandwidth.
Hybrid Database Implementation: An experiment was conducted comparing a local MS SQL Server database against a hybrid model. In the hybrid model, metadata resides in the local SQL Server, while large article bodies (100 KB to 3 MB) are stored in Microsoft Azure Storage.

Key Contributions

Dynamic Control Framework: The paper introduces a shift from static graph-theory models to dynamic difference equation models with feedback for managing hybrid cloud resources.
Traffic Management Strategy: A specific two-stage traffic separation and dynamic bandwidth allocation strategy using Cisco AVC is detailed to handle the bottlenecks inherent in hybrid cloud data exchange.
Hybrid Storage Architecture: The authors demonstrate a practical implementation of a hybrid database where metadata and large binary objects (BLOBs) are decoupled, storing heavy data in the public cloud while maintaining relational integrity locally.

Results

The study presents empirical results derived from monitoring a corporate network over a year and conducting query performance tests:

Network Load Distribution: Empirical histograms of channel load showed that the proposed control algorithm significantly altered the traffic distribution. The "tails" and peaks in the load distribution were reduced (as seen in the comparison between Figure 3a and 3b), indicating a reduction in dropped packets and an improvement in data delivery quality.
Storage Efficiency: In the database experiment, the local database (containing all data) occupied approximately 26,667 MB of memory. In contrast, the hybrid database (metadata only) occupied only 47.08 MB on the local server, with the remaining ~27 GB stored in the cloud.
Query Performance: The experiment demonstrated that applying graph theory for time evaluation of database queries in this context is ineffective because the resources used in local server processing differ fundamentally from those in external cloud storage.

Significance and Claims

The paper claims that the development of these control models and optimization criteria is essential for providing QoS in hybrid cloud services handling Big Data. The authors assert that their approach successfully addresses the economic and technical issues of porting applications while maintaining lossless data access speeds.

The primary significance lies in the validation that dynamic control systems with feedback are necessary to manage the bottlenecks of hybrid infrastructures, which static models cannot resolve. The results suggest that offloading large data bodies to public cloud storage while retaining metadata locally is an effective strategy for managing storage costs and resource allocation, provided that a robust, dynamic QoS control mechanism (such as the Cisco AVC-based system described) is in place to manage the network traffic between these environments. The paper concludes that for hybrid cloud query processing, traditional theoretical assessments must be replaced by dynamic, state-aware control mechanisms.