Hardware load balancing is a type of physical equipment specifically designed to manage and distribute user requests among a complex network of computers. They usually have privately owned operating systems running on their devices. In contrast, software load balancing techniques are compatible with any existing architecture [24]. Today, most of the major load balancers utilize x86-based computer networks. They don’t have any specific hardware requirements, they only need some pre-configuration for functioning properly. A major distinction between software and hardware load balancers is the scalability factor. While hardware load balancers are static and service operations require additional hardware from the data center, software options can scale elastically as demand increases and ultimately software solutions are customizable [25]. They can be individually configured and modified, and integrate seamlessly with intricate hybrid and multi-cloud architectures. This advantage is not available with hardware load balancing.

Administrators can distribute network traffic to other servers using software load balancing. To check client requests, load balancers check application-level information such as IP address, HTTP header, and the content of the request. After examining the servers, the load balancer selects the most appropriate one and forwards the request to that server [26]. Usually, software-based solutions are provided by the Application Delivery Controller (ADC), which runs on a VMs or on a standard server. Unlike HLD, software load balancer doesn’t require any dedicated load-balancing device even though it provides the same functionality. The load balancing software is able to operate on any regular or virtual server [27].

Scalability is the primary benefit of software-based load balancing over hardware-based load balancing. A software-based load balancer can dynamically and instantaneously adapt to changes in network traffic by adding or removing virtual servers based on demand [28]. Because they work in a wide variety of environments, they are even more scalable than hardware load balancers. Standard desktop OSs, remote servers, cloud services, bare metal, and containers can all be configured to work with them. Because hardware load balancers cannot be programmed, they are less flexible [29].

Software load balancers can help businesses save money, particularly when the companies employ load balancing as a service or lbaas. despite the fact that if an IT company buys its own software load balancers, it will count as OPEX rather than CAPEX. Software load balancers save time and money since they are simple to deploy and on demand [30]. Last but not least, load-balancing software provides an additional degree of protection by having the capacity to discard erroneous data packets even before they get to the server by residing between the client and the server.

More and more companies are resorting to cloud-native solutions to guarantee higher availability and better performance, since the amount of internet traffic and workloads continues to rise. Cloud computing can be defined as the process of dividing up workloads, traffic, and user requests across multiple servers running on a cloud environment [31]. This technique makes sure that every cloud resource has a load that it can reasonably handle, preventing computers or servers in a cloud environment from being either overloaded or underutilized. As a result, cloud optimization is enhanced [32].

With the aid of efficient load balancing, businesses can meet the demands of cloud-based applications meanwhile reducing latency, boosting reliability, optimizing performance, and preventing downtime. As-needed cloud load balancing is provided via Cloud Load Balancing as a Service, or LBaaS, which takes the place of dedicated, on-premises hardware [33]. LBaaS provides increased scalability since load balancing in the cloud can handle traffic spikes without requiring the reconfiguration of physical infrastructure. Connecting to the nearest servers guarantees higher availability, and LBaaS can save maintenance and investment costs in comparison to hardware-based solutions [34]. Table 2 provides a comparative analysis of different types of load balancing in a concise manner.

So, how does cloud load balancing work? Load balancers are used in the cloud between backend servers and client machines. They make use of algorithms that consider different factors like server load and geographic distance, to divide requests efficiently. These algorithms typically fall into two categories, namely static and dynamic [35]. Static algorithms include algorithms such as the Round Robin algorithm, which distributes requests among cloud servers in a basic, recurring sequence.

Weighted Round Robin provides a higher weight to servers that have more capacity, allowing them to handle a large volume of incoming application traffic. IP Hash is a service that maps a client's IP address to specific servers by performing a hash, or mathematical computation, on the address [36]. Dynamic algorithms include algorithms such as the least connections algorithm which distributes workload between the servers having the lowest number of active connections. least response time distributes traffic among the servers having lowest average response time and fewest connections. Lastly, least bandwidth sends requests to the servers consuming the lowest amount of bandwidth during a recent period, etc [14], [15].

Cost It can be expensive, It is generally more It can be cost-effective since especially for high- affordable than hardware users only have to pay for the performance models. load balancers. resources they use.

Performance It has high performance as It may have lower Usually depends on the cloud they are optimized for load performance as compared provider but it is designed to balancing tasks. to hardware load balancer, provide high performance to especially under heavy efficiently distribute incoming loads. traffic.

Scalability Scalability is limited since It can be easily scaled by It is highly scalable as they can enhancing capability adding more resources or easily accommodate changes in requires buying more upgrading the underlying traffic and resource demands.

hardware. hardware.

Maintenance It needs ongoing It may require ongoing Simplified management as the configuration and software updates and cloud provider takes care of maintenance but may maintenance which maintenance and updates but require specialized requires technical may have less control over knowledge. knowledge and configuration and automation tools. customization.

Flexibility It is generally less flexible as It provides moderate It offers unparalleled flexibility compared with other types flexibility as it can be due to its virtualized and of load balancing since it deployed on a variety of dynamic scalable nature and often comes with limited platforms and various configurable options.

configurations. environments including cloud-based infrastructure.

Example A hardware load balancer A startup with a growing To ensure smooth app can be used by a major user base can deploy a performance and fast responses, online retailer to split up software load balancer on a mobile app developer can incoming web traffic among a virtual machine hosted in employ a cloud-based load several web servers, the cloud, distributing balancer offered by any cloud guaranteeing fast response incoming requests among provider to distribute incoming time and a seamless online multiple application API calls among several shopping experience for its servers to handle increased backend servers.

users. traffic.

There are many types of cloud load balancing, a few of them have been covered below. First we have DNS or Domain Name System load balancing that relies on the DNS infrastructure to distribute incoming traffic among multiple servers or resources. It works by providing a domain name to multiple IP addresses, effectively directing clients to different servers based on various policies. It is relatively simple to implement, as it doesn’t require any specialized hardware or software. It provides basic load balancing and failover capabilities. It can also distribute traffic across geographically distributed servers, improving performance for users in different regions [37].

There are some disadvantages associated with DNS load balancing. It is restricted to DNS resolution time, which updates more slowly than other load-balancing methods. Response time, resource usage, and server health are not taken into account. Applications needing fine-grained load distribution or session persistence can find DNS load balancing unsuitable [17]. DNS load balancing can be used by a Content Delivery Network (CDN) to direct users to the closest edge server based on their geographical location, ensuring faster content delivery and reduced latency [38]. Next we have Global Server Load Balancing (GSLB) that is a technique used to distribute traffic across geographically dispersed data centers.Integrating DNS load balancing with other cutting-edge features like health checks, offers a more clever and effective way to distribute traffic. Unlike DNS load balancing, it provides load balancing and failover capabilities across multiple data centers or geographic locations, can also improve performance and reduce latency for users by directing them to the closest or best-performing data center and supports advanced features, such as server health checks, session persistence, and custom routing policies. But it comes with some disadvantages.

It can be more complex to set up and manage than other load balancing techniques [39]. It may also require specialized hardware or software, increasing costs. It can be subject to the limitations of DNS, such as slow updates and caching issues. A global company may utilize GSLB to split up incoming requests for its web applications among multiple data centers globally, guaranteeing optimal performance and high availability for customers in various regions. To obtain the highest potential speed, scalability, and reliability, multiple load balancing approaches are combined in a different sort of cloud load balancing called hybrid load balancing. To provide the most efficient and adaptable load-balancing technique for a given circumstance, it usually combines hardware, software, and cloud-based solutions. Main advantage of hybrid load balancing is that it can provide the best combination of performance, scalability, and reliability by leveraging the strengths of different load balancing techniques. Additionally, it provides a great deal of flexibility because it can be customized to fit particular demands and infrastructure, enabling businesses to modify and enhance their load-balancing plan as circumstances change.

It can be more difficult to set up, administer, and manage than single-technique solutions since it combines a variety of load-balancing strategies, and it might call for a greater degree of skill and knowledge of these techniques. Possibly more expensive because it requires a mix of cloud-based services, software, and hardware. A hybrid load-balancing approach can be used by any major online streaming platform. It combines DNS load balancing for global traffic management, cloud-based load balancers for scalable content delivery, and hardware load balancers in data centers for high-performance traffic distribution. Millions of users worldwide are guaranteed excellent speed, scalability, and dependability using this strategy. There are two more types of cloud load balancing that are not widely known. The first one is Transport Layer load balancing, popularly known as Layer 4 load balancing. It functions at the fourth tier of the OSI model, namely the transport layer. Based on data from the TCP or UDP header, including source and destination IP addresses and port numbers, it divides incoming traffic. Because it makes choices based only on limited information from the transport layer, it operates extremely quickly and effectively. It can also manage a large range of traffic kinds and protocols. It's comparatively easy to implement and administer. However, there are some drawbacks. It is less effective in some situations due to unawareness of application-level information.It might not be appropriate for applications that need fine-grained load distribution or session persistence since it doesn't account for resource usage, response time, or server health.

To ensure that players are evenly divided among the available game servers for seamless gameplay, online gaming platforms primarily use it to distribute game server traffic based on IP addresses and port numbers. Next we have application layer load balancing, commonly known as Layer 7 load balancing. As evident from its name, this type of load balancing works on the application layer of the OSI model, i.e., the seventh layer. To distribute incoming traffic more effectively, it considers application-specific data, including HTTP headers, cookies, and URL routes. It offers load balancing that is more intelligent and precise since it takes application-level data into account.

It also supports advanced features, such as session persistence, content-based routing, and SSL offloading, and can be tailored to specific application requirements and protocols However, because it needs a more thorough examination of incoming data than Layer 4 load balancing, it is slower and takes more resources. Specialized hardware or software may also be needed to manage application-level traffic processing and inspection. When compared to other load-balancing strategies, it is more difficult to configure and maintain. Layer 7 load balancing can be used by a web application with several microservices to route incoming API calls according to the URL path, guaranteeing that each microservice receives only the requests it is in charge of processing.

Cloud computing uses technology to support people and systems. The primary goal of cloud computing is to share resources, software, and information online in order to reduce capital and operational costs, improve response and data processing times, improve system stability, and allow for future system modifications. As a result, there are many technical issues that need to be addressed, such as server consolidation, fault tolerance, high availability, and scalability. Researchers are trying to find ways to maintain load during VM migration and avoid overloading systems [5]. Some other future challenges include the possibility of VM-based malware, targeted attacks (shared-tenancy environment), service provider trust, vendor lockouts, data security if there are many layers, problems with botnet hosting,etc. [14], [19].

Eventually, the Internet of Things (IoT) will shape the future of cloud computing, with the obvious challenge of preventing it from becoming the Internet of Over Things (IoOT) and above [18]. Research needs to be done to exhaust challenges posed by the Internet underneath and serious new cyberattack threats will go after new connected devices that are more susceptible. Another key issue in load balancing in cloud computing is how to manage complexity as networks grow exponentially every day. Managing cryptocurrencies is another future problem associated with the cloud environment, because future currency markets are slowly moving towards digital currencies [20].

SI methods by processing requests, their exchange and availability of many VMs can contribute to resources the use of which has increased and reduced the response time so the integration of remote data centers SI techniques with clouds in the common terrain will be the focus of future studies. Therefore, it is important to provide SI techniques such as autonomous multi-objective scheduling to optimize the competing objectives in the least possible time [13].

The increasing number of clients and networks in this era is making it challenging to access records. Cloud distribution systems allow the customer to apply all of their services and execute their applications for less expenditure than they would need to pay for the underlying structure [21]. The computational component is probably overloaded with multiple users requesting to run their programs. There are several techniques in present structures that use stability masses to obtain either the initial node frame or dynamic node load [22]. The dynamic state of the nodes and system rearrangement so as of the burden percent are nevertheless taken into account within the gift machine algorithms. The existing technique's dynamic algorithms consequently do not provide better output, response times, lower prices, or less delays. Performance optimization, reliability, security and scalability are some of the key challenges associated with load balancing [23].

This research has a goal of answering a set of questions that were identified from literature surveys, these questions need to be answered before going into the load balancing process. Some of these questions have been answered in literature while others still remain unanswered. The questions are given as follows:

I. What causes the load unbalancing problem? This question addresses the causes behind load unbalancing problems. It requires a thorough study of each of the factors leading to load unbalancing.

II. Why is load balancing so important in cloud computing? This question explores the problems and challenges faced by cloud service providers.

III. How do different types of load balancers (software-based, hardware-based, cloud-based) compare in terms of performance and scalability?

IV. What are the main load balancing algorithms used in modern load balancer systems?

After these questions, it is important to dig deeper into the realm of load balancing as these will help us to understand things clearly. By carefully examining and answering each question, we aim to gain a better understanding of our topic [8]. Through this process, we aim to gain a comprehensive understanding that comes only from beyond surface-level observations. This section serves as a roadmap for this study, as it gives us a glimpse of where we are headed and what we might find along the way.

After comparing and studying different load balancers, we have prepared Table 3 comparing different load balancers available in the market.