Large-scale Data Centers Research Articles

Research on Resource Allocation Technology of Cluster Virtual Cloud Based on SDN

The ever-expanding demand for cloud computing needs to build a large-scale data center. How to operate the data center efficiently is an urgent problem to be solved. Statistical reuse is a prominent feature in cloud computing, and the utilization of physical resources can be improved by using virtualization technology. The core technology of network virtualization is to decouple the virtual network based on software from the underlying network based on hardware, and the proposal of Software Defined Network (SDN) provides a practical solution to solve this problem. Cloud computing puts forward the highest goal of sharing resources in the whole Internet, makes full use of computing/storage resources to the maximum extent, and is an effective method to integrate high-performance computing resources of the whole society. This chapter describes the background and significance of virtual resource allocation in cloud computing environment, studies the cluster virtual cloud resource allocation technology based on SDN, and analyzes how to realize reasonable cloud computing resource allocation to ensure network performance.

A Virtual Machine Migration Mechanism Based on Firefly Optimization for Cloud Computing

Background: Cloud computing is one of the prominent technology revolutions around us. It is changing the ways the consumer expends services, changing the ways the organization develop and run applications and is completely reshaping the old business models in multiple industries. Cloud service providers need large-scale data centers for offering cloud resources to users, the electric power consumed by these data centers has become a concrete and prudential concern. Most of the energy is dissipated in these data centers due to under-utilized hosts, which also subsidies to global warming. The broadly adept technology is virtual machine migration in cloud computing, therefore, the main focus is to save energy. Objective: Virtual Machine (VM) migration can reap various objectives like load balancing, ubiquitous computing, power management, fault tolerance, server maintenance, etc. This paper presents an energy-oriented mechanism for VM migration based on firefly optimization that reduces energy consumption and the number of VM migrations to a great extent. Method: A Firefly Optimization (FFO) oriented VM migration mechanism has been proposed, which allocates tasks to the physical machines in cloud data centers. It strives to migrates high loaded VMs from one physical node to another, which induces minimum energy consumption after VM migration. Results: The empirical result shows that the FFO based mechanism, implemented in the CloudSim simulator, performs better in terms of the number of hosts saved up to 13.91% in contrast to the First Fit Decreasing (FFD) algorithm and 8.21% as compared to Ant Colony Optimization (ACO). It reduced energy consumption up to 12.76% as compared to FFD and 7.78% as compared to ACO and, ultimately lesser number of migrations up to 52.49% when compared to FFD and 44.51% as compared to ACO. Conclusion: The proposed scheme performs better in terms of saving hosts, reducing energy consumption, and decreasing the number of migrations in contrast to FFD and ACO techniques. The research paper also presents challenges and issues in cloud computing, VM migration process, VM migration techniques, their comparative review as well.

Economic analysis of hydrogen-powered data center

The data center needs more and more electricity due to the explosive growth of IT servers and it could cause electricity power shortage and huge carbon emission. It is an attractive and promising solution to power the data center with hydrogen energy source. The present work aims to conduct an economic analysis on the hydrogen-powered data center. Configurations of hydrogen-powered and traditional data centers are compared and the differences focus on backup power system, converter/inverter, fuel cell subsystem, carbon emission, hydrogen and electricity consumptions. Economic analysis is conducted to evaluate the feasibility to power the data center with hydrogen energy source. Results show that electricity price increasing rate and hydrogen cost are the main factors to influence economic feasibility of hydrogen-powered data center. When the electricity price keeps constant in the coming two decades, the critical hydrogen price is about 2.8 U.S. dollar per kilogram. If the electricity price could increase 5% annually due to explosive growth of electric vehicles and economy, critical hydrogen price will become 6.4 U.S. dollar per kilogram. Hydrogen sources and transportation determine the hydrogen price together. Hydrogen production cost varies greatly with hydrogen sources and production technologies. Hydrogen transport cost is greatly influenced by distances and H2 consumptions to consumers. It could be summarized that the hydrogen-powered data center is economic if hydrogen could be produced from natural gas or H2-rich industrial waste streams in chemical plant and data center could not be built too far away from hydrogen sources. In addition, large-scale hydrogen-powered data center is more likely to be economic. Solar hydrogen powered data center has entered into a critical stage in the economic feasibility. Solar hydrogen production cost has restrained the H2 utilization in data center power systems now, since it could be competitive only when more strict carbon emission regulation is employed, hydrogen production cost reduces greatly and electricity price is increasing greatly in the future. However, it could be expected solar hydrogen-powered system will be adopted as the power source of data centers in the next few years.

Energy, exergy and computing efficiency based data center workload and cooling management

The rapidly rising computing workloads in data centers (DCs) have necessitated new approaches to ensure effective performance and resilience that minimize the associated cooling energy. The literature on thermally-aware workload management provides strategies to reduce this energy cost, while typically ignoring the reduction in cooling capacity due to thermodynamic irreversibility and computing performance per unit energy consumption. Hence, we provide an approach that considers coefficient of performance COPc, exergy efficiency ηex, and a new metric, computing performance ratio CPR. In contrast to existing methods that consider one-dimensional workload distributions, the temperature predictions from a physics-based zonal model are used to optimize cooling for two-dimensional workload distributions in a multi-rack DC. The investigation reveals physics associated with two-dimensional workload management for multi-rack DCs, provides a framework for trade-offs between COPc, ηex, and CPR, explains the influence of IT load factor LF on different objectives, and describes how parameters obtained from single- and multi-objective problems can vary. Our findings show that COPc, and ηex can be improved by up to 20% and 8% by regulating the chilled water temperature and airflow setpoints while increasing the LF degrades the CPR by 7.5%. These results enable an extended approach for heterogeneous LF management in large-scale DCs.

Moisture Computing-Based Internet of Vehicles (IoV) Architecture for Smart Cities.

Recently, the concept of combining ‘things’ on the Internet to provide various services has gained tremendous momentum. Such a concept has also impacted the automotive industry, giving rise to the Internet of Vehicles (IoV). IoV enables Internet connectivity and communication between smart vehicles and other devices on the network. Shifting the computing towards the edge of the network reduces communication delays and provides various services instantly. However, both distributed (i.e., edge computing) and central computing (i.e., cloud computing) architectures suffer from several inherent issues, such as high latency, high infrastructure cost, and performance degradation. We propose a novel concept of computation, which we call moisture computing (MC) to be deployed slightly away from the edge of the network but below the cloud infrastructure. The MC-based IoV architecture can be used to assist smart vehicles in collaborating to solve traffic monitoring, road safety, and management issues. Moreover, the MC can be used to dispatch emergency and roadside assistance in case of incidents and accidents. In contrast to the cloud which covers a broader area, the MC provides smart vehicles with critical information with fewer delays. We argue that the MC can help reduce infrastructure costs efficiently since it requires a medium-scale data center with moderate resources to cover a wider area compared to small-scale data centers in edge computing and large-scale data centers in cloud computing. We performed mathematical analyses to demonstrate that the MC reduces network delays and enhances the response time in contrast to the edge and cloud infrastructure. Moreover, we present a simulation-based implementation to evaluate the computational performance of the MC. Our simulation results show that the total processing time (computation delay and communication delay) is optimized, and delays are minimized in the MC as apposed to the traditional approaches.

HS6: An Efficient H-Code RAID-6 Scaling by Optimizing Data Migrating and Parity Updating

Caused by massive data, large-scale data centers suffer from insufficient storage capacity. This increases the risk of component damage and disk failure. To enhance storage capacity of data centers, scaling has turned out to be one of the most popular choices. RAID-6 is deployed extensively that has superior reliability, availability, and scalability compared to other RAID levels. However, higher reliability also means expensive parity data update cost. To address this issue, this paper proposes an efficient RAID-6 scaling scheme, HS6, for H-Code. The basic idea of HS6 is reducing the total parity data update cost and saving the total scaling time by optimizing data migration and parity data update. The properties of HS6 can be summarized as follows: (1) it utilizes horizontal data migration to eliminate the horizontal parity data update, and (2) it makes full use of original parity data to cut down the total parity data update cost. Numerical results and experimental data analysis indicate that: (1) HS6 decreases the data migration rate by 76.92% $$\sim$$ 95.31%, 55.56% $$\sim$$ 84.46%, and 14.29% $$\sim$$ 16.67% compared to Round-Robin, Semi-RR, and HCS, (2) HS6 saves the total scaling time by 62.05% $$\sim$$ 76.75%, 53.24% $$\sim$$ 73.94%, and 3.68% $$\sim$$ 6.79% against Round-Robin, Semi-RR, and HCS, respectively, under offline and outperforms Round-Robin by 59.99% $$\sim$$ 79.02%, Semi-RR by 53.04% $$\sim$$ 76.41%, and HCS by 0.56% $$\sim$$ 23.78% in total scaling time under four workloads, and (3) HS6 maintains almost the same user average response time with Round-Robin, Semi-RR, and HCS during scaling and after scaling.

EAMA: Efficient Adaptive Migration Algorithm for Cloud Data Centers (CDCs)

The rapid demand for Cloud services resulted in the establishment of large-scale Cloud Data Centers (CDCs), which ultimately consume a large amount of energy. An enormous amount of energy consumption eventually leads to high operating costs and carbon emissions. To reduce energy consumption with efficient resource utilization, various dynamic Virtual Machine (VM) consolidation approaches (i.e., Predictive Anti-Correlated Placement Algorithm (PACPA), Resource-Utilization-Aware Energy Efficient (RUAEE), Memory-bound Pre-copy Live Migration (MPLM), m Mixed migration strategy, Memory/disk operation aware Live VM Migration (MLLM), etc.) have been considered. Most of these techniques do aggressive VM consolidation that eventually results in performance degradation of CDCs in terms of resource utilization and energy consumption. In this paper, an Efficient Adaptive Migration Algorithm (EAMA) is proposed for effective migration and placement of VMs on the Physical Machines (PMs) dynamically. The proposed approach has two distinct features: first, selection of PM locations with optimum access delay where the VMs are required to be migrated, and second, reduces the number of VM migrations. Extensive simulation experiments have been conducted using the CloudSim toolkit. The results of the proposed approach are compared with the PACPA and RUAEE algorithms in terms of Service-Level Agreement (SLA) violation, resource utilization, number of hosts shut down, and energy consumption. Results show that proposed EAMA approach significantly reduces the number of migrations by 16% and 24%, SLA violation by 20% and 34%, and increases the resource utilization by 8% to 17% with increased number of hosts shut down from 10% to 13% as compared to the PACPA and RUAEE, respectively. Moreover, a 13% improvement in energy consumption has also been observed.

Cloud ruins: Ericsson's Vaudreuil-Dorion data centre and infrastructural abandonment

ABSTRACT The past decade has seen the accelerated growth and expansion of large-scale data centre operations across the world to support emerging consumer and business data and computation needs. Built out rapidly, these emergent digital infrastructures carry the promise for new local industrial futures, all while their paths to obsolescence are shortened. Their lifespans are dependent on financial speculation, shifting corporate strategies, and advances in consumer technology. In this article we track the promise and afterlife of an abruptly abandoned data centre constructed by the global telecom giant Ericsson in Vaudreuil-Dorion, a town near Montréal, Québec, Canada, in order to expand emergent debates about digital ruination. Employing site visits, press reports, and qualitative interviews with architects and staff involved with the data centre's development in Sweden and Canada, we propose ‘cloud ruins’ as a sensitising concept to capture some of the specific meanings and material articulations that the abandonment of global data infrastructures may evoke in local contexts. Simultaneously familiar and novel, cloud ruins anticipate an emergent landscape of post-digital ruination that unfolds in the built environment in peripheral communities, part of the global logistical cities from within which our contemporary understandings of digitalisation are produced.

MONet: heterogeneous Memory over Optical Network for large-scale data center resource disaggregation

The memory over optical network (MONet) system is a disaggregated data center architecture where serial (HMC)/parallel (DDR4) memory resources can be accessed over optically switched interconnects within and between racks. An FPGA/ASIC-based custom hardware IP (ReMAT) supports heterogeneous memory pools, accommodates optical-to-electrical conversion for remote access, performs the required serial/parallel conversion, and hosts the necessary local memory controller. An optically interconnected HMC-based (serial I/O type) memory card is accessed by a memory controller embedded in the compute card, simplifying the hardware near the memory modules. This substantially reduces overheads on latency, cost, power consumption, and space. We characterize CPU–memory performance by experimentally demonstrating the impact of distance, number of switching hops, transceivers, channel bonding, and bit rate per transceiver on the bit error rate, power consumption, additional latency, sustained remote memory bandwidth/throughput (using industry standard benchmark STREAMS), and cloud workload performance (such as operations per second, average added latency, and retired instructions per second memcached with YCSB cloud workloads). MONet pushes the CPU–memory operational limit from a few centimeters to tens of meters, yet applications can experience as low as 10% performance penalty (at 36 m) compared to a direct-attached equivalent. Using the proposed parallel topology, a system can support up to 100,000 disaggregated cards.

Closer: Scalable load balancing mechanism for cloud datacenters

Cloud providers (e.g., Google, Alibaba, Amazon) own large-scale datacenter networks that comprise thousands of switches and links. A load-balancing mechanism is supposed to effectively utilize the bisection bandwidth. Both Equal-Cost Multi-Path (ECMP), the canonical solution in practice, and alternatives come with performance limitations or significant deployment challenges. In this work, we propose Closer, a scalable load balancing mechanism for cloud datacenters. Closer complies with the evaluation of technology including the deployment of Clos-based topologies, overlays for network virtualization, and virtual machine (VM) clusters. We decouple the system into centralized route calculation and distributed route decision to guarantee its flexibility and stability in large-scale networks. Leveraging In-band Network Telemetry (INT) to obtain precise link state information, a simple but efficient algorithm implements a weighted ECMP at the edge of fabric, which enables Closer to proactively map the flows to the appropriate path and avoid the excessive congestion of a single link. Closer achieves 2 to 7 times better flow completion time (FCT) at 70% network load than existing schemes that work with same hardware environment.

Minimizing energy on homogeneous processors with shared memory

Energy efficiency is a crucial desideratum in the design of computer systems, from small-sized mobile devices with limited battery to large scale data centers. In such computing systems, processors and memory are considered as two major power consumers among all the system components. One recent trend to reduce power consumption is using shared memory in multi-core systems, such architecture has become ubiquitous nowadays. However, implementing the energy-efficient methods to the multi-core processor and the shared memory separately is not trivial. In this work, we consider the energy-efficient task scheduling problem, which coordinates the power consumption of both the multi-core processor and the shared memory, especially focus on the general situation in which the number of tasks is more than the number of cores. We devise an approximation algorithm with guaranteed performance in the multiple cores system. We tackle the problem by first presenting an optimal algorithm when the assignment of tasks to cores is given. Then we propose an approximation assignment for the general task scheduling.

Energy efficient virtual machine migration approach with SLA conservation in cloud computing

In the age of online workload explosion, cloud users are increasing exponentialy. Therefore, large scale data centers are required in cloud environment that leads to high energy consumption. Hence, optimal resource utilization is essential to improve energy efficiency of cloud data center. Although, most of the existing literature focuses on virtual machine (VM) consolidation for increasing energy efficiency at the cost of service level agreement degradation. In order to improve the existing approaches, load aware three-gear THReshold (LATHR) as well as modified best fit decreasing (MBFD) algorithm is proposed for minimizing total energy consumption while improving the quality of service in terms of SLA. It offers promising results under dynamic workload and variable number of VMs (1–290) allocated on individual host. The outcomes of the proposed work are measured in terms of SLA, energy consumption, instruction energy ratio (IER) and the number of migrations against the varied numbers of VMs. From experimental results it has been concluded that the proposed technique reduced the SLA violations (55%, 26% and 39%) and energy consumption (17%, 12% and 6%) as compared to median absolute deviation (MAD), inter quartile range (IQR) and double threshold (THR) overload detection policies, respectively.

Server consolidation: A technique to enhance cloud data center power efficiency and overall cost of ownership

Cloud computing is a well-known technology that provides flexible, efficient, and cost-effective information technology solutions for multinationals to offer improved and enhanced quality of business services to end-users. The cloud computing paradigm is instigated from grid and parallel computing models as it uses virtualization, server consolidation, utility computing, and other computing technologies and models for providing better information technology solutions for large-scale computational data centers. The recent intensifying computational demands from multinationals enterprises have motivated the magnification for large complicated cloud data centers to handle business, monetary, Internet, and commercial applications of different enterprises. A cloud data center encompasses thousands of millions of physical server machines arranged in racks along with network, storage, and other equipment that entails an extensive amount of power to process different processes and amenities required by business firms to run their business applications. This data center infrastructure leads to different challenges like enormous power consumption, underutilization of installed equipment especially physical server machines, CO2 emission causing global warming, and so on. In this article, we highlight the data center issues in the context of Pakistan where the data center industry is facing huge power deficits and shortcomings to fulfill the power demands to provide data and operational services to business enterprises. The research investigates these challenges and provides solutions to reduce the number of installed physical server machines and their related device equipment. In this article, we proposed server consolidation technique to increase the utilization of already existing server machines and their workloads by migrating them to virtual server machines to implement green energy-efficient cloud data centers. To achieve this objective, we also introduced a novel Virtualized Task Scheduling Algorithm to manage and properly distribute the physical server machine workloads onto virtual server machines. The results are generated from a case study performed in Pakistan where the proposed server consolidation technique and virtualized task scheduling algorithm are applied on a tier-level data center. The results obtained from the case study demonstrate that there are annual power savings of 23,600 W and overall cost savings of US$78,362. The results also highlight that the utilization ratio of already existing physical server machines has increased to 30% compared to 10%, whereas the number of server machines has reduced to 50% contributing enormously toward huge power savings.

Adaptive Preference-Aware Co-Location for Improving Resource Utilization of Power Constrained Datacenters

Large-scale datacenters often host latency-sensitive services that have stringent Quality-of-Service requirement and experience diurnal load pattern. Co-locating best-effort applications that have no QoS requirement with the latency-sensitive services has been widely used to improve the resource utilization of datacenters with careful shared resource management. However, existing co-location techniques tend to result in the power overload problem on power constrained servers due to the ignorance of the power consumption. To this end, we propose Sturgeon, a runtime system proactively manages resources between co-located applications in a power constrained environment, to ensure the QoS of latency-sensitive services while maximizing the throughput of best-effort applications. Our investigation shows that, at a given load, there are multiple feasible resource configurations to meet both QoS requirement and power budget, while one of them yields the maximum throughput of best-effort applications. To find such a configuration, we establish models to accurately predict the performance and power consumption of the co-located applications. Sturgeon monitors the QoS of the services periodically, in order to eliminate the potential QoS violation caused by the unpredictable interference. Besides, when the datacenter hosts different types of applications to perform co-location, Sturgeon places applications with their preferable candidates to improve the overall throughput. The experimental results show that at server level Sturgeon improves the throughput of the best-effort application by 25.43 percent compared to the state-of-the-art technique, while guaranteeing the 95%-ile latency within the QoS target; at cluster level, Sturgeon improves the overall throughput of best-effort applications by 13.74 percent compared to the baseline.

The Power Control Method of Data Center Based on Cloud- Edge Collaboration

Data centers have the characteristics of huge energy consumption and strong power fluctuations. To solve the problem of unstable supply and demand such as frequent power fluctuations in the micro-grid of the data center using renewable energies, this paper establishes a controllable load model of server clusters based on the task migration mechanism by analyzing the various tasks and characteristics of large cloud data centers, as well as edge computing nodes. Through the collaborative optimization control of edge and cloud computing tasks, and dynamically adjusting and migrating the server cluster load, a new data center tie-line power control method is proposed. The results of calculation examples show that the method can achieve power smoothing of large-scale cloud data center tie lines through cloud-edge collaborative control.

Management of cloud resources and social change in a multi-tier environment: A novel finite automata using ant colony optimization with spanning tree

The enormous demand for computational resources due to rapid Cloud growth has led to the creation of large-scale data centers (DC). Consequently, a tremendous amount of energy is consumed with high carbon dioxide emissions, which necessitates that cloud service providers (CSP) develop high quality of service (QoS) strategies to address such challenges. Proper management and utilization of resources ensure a well-balanced load distribution that makes energy consumption sufficient. Developing new algorithms and exploring efficient methods and techniques is highly desired for the management of virtualized DCs. This research proposes Finite Automata using Ant Colony Optimization with Spanning Tree (FAACOST), a machine-learning concept for managing cloud resources in a multi-tier environment. The overall objective is to minimize energy consumption through data placement leverage and virtual machines (VM) consolidation. The proposed technique's efficiency was benchmarked on four performance metrics (machine learning, dynamic nature, scalability, and QoS). Based on the extensive experiments conducted, FAACOST recorded an energy consumption of 132 w with 20 tasks compared to the benchmarked techniques that consumed 363 w. The experimental results show that FAACOST achieves the optimal number of physical machines (PMs) and is more energy-efficient.

Ernie: Scalable Load-Balanced Multicast Source Routing for Cloud Data Centers

Nowadays, most applications hosted on public cloud data centers (DCs) disseminate data from a single source to a group of receivers for service deployment, data replication, software upgrade, etc. For such one-to-many data communication paradigm, multicast routing is the natural choice as it reduces network traffic and improves application throughput. Unfortunately, recent approaches adopting IP multicast routing suffer from scalability and load balancing issues, and do not scale well with the number of supported multicast groups when used for cloud DC networks. Furthermore, IP multicast does not exploit the topological properties of DCs, such as the presence of multiple parallel paths between end hosts. Despite the recent efforts aimed at addressing these challenges, there is still a need for multicast routing protocol designs that are both scalable and load-balancing aware. This paper proposes Ernie, a scalable load-balanced multicast source routing for large-scale DCs. At its heart, Ernie further exploits DC network structural properties and switch programmability capabilities to encode and organize multicast group information inside packets in a way that minimizes downstream header sizes significantly, thereby reducing overall network traffic. Additionally, Ernie introduces an efficient load balancing strategy, where multicast traffic is adequately distributed at downstream layers. To study the effectiveness of Ernie, we extensively evaluate Ernie’s scalability behavior (i.e., switch memory, packet size overheads, and CPU overheads), and load balancing ability through a mix of simulation and analysis of its performances. For example, experiments of large-scale DCs with 27k+ servers show that Ernie requires a downstream header sizes that are 10× smaller than those needed under state-of-the-art schemes while keeping end-host overheads at low levels. Our simulation results also indicate that at highly congested links, Ernie can achieve a better multicast load balancing than other existing schemes.

Clustered Multicast Source Routing for Large-Scale Cloud Data Centers

The multi-tenancy concept in cloud data center (DC) networks paves the way towards advancements and innovation in the underlying infrastructure such as network virtualization. Multicast routing is essential in leveraging multi-tenancy to its full potential. However, traditional IP multicast routing is not suitable for DC networks due to the need to support a massive amount of multicast groups and hosts. State-of-the-art DC multicast routing approaches aim to overcome these scalability issues by, for instance, taking advantage of the symmetry of DC topologies and the programmability of DC switches to compactly encode multicast group information inside packets, thereby reducing the overhead resulting from the need to store the states of flows at the network switches. Although these approaches scale well with the number of multicast groups, they do not perform well with group sizes and, as a result, yield substantial traffic control overhead and network congestion. In this article, we present Bert, a scalable source-initiated DC multicast routing approach that scales well with both the number and size of multicast groups through the clustering of multicast group members where each cluster employs its own forwarding rules. Compared to the state-of-the-art approach, Bert yields much less traffic control overhead by significantly reducing packet header sizes and eliminating switch memory usage across the switches.

FlexChain: Bridging Parallelism and Placement for Service Function Chains

A Service Function Chain (SFC) is an ordered sequence of network functions (NFs). With the emerging Network Function Virtualization (NFV) paradigm, NFs can be deployed as software instances on commodity servers, leading to a more flexible provision of network service. However, the flexibility of NFV comes with considerable compromises since virtual network functions (VNFs) introduce significant processing latency overheads. In this article, we design a flexible SFC parallel system called FlexChain, enabling the parallelism among VNFs to reduce the processing latency of SFCs. To leverage the benefits of parallelism, we study the problem of joint optimization over SFC parallelism and placement with the objective of accepting as many requests as possible. Since the problem is proved to be NP-hard, we propose a parallelism-aware approximation placement algorithm with performance guarantees, and an efficient heuristic algorithm for large-scale data center networks. Our simulation results show that FlexChain combined with our placement algorithm can substantially improve the number of accepted flows in the latency-sensitive scenario, and significantly reduce the average latency of accepted flows at the same time.

Toward lower-diameter large-scale HPC and data center networks with co-packaged optics

We investigate the advantages of using co-packaged optics for building low-diameter, large-scale high-performance computing (HPC) and data center networks. The increased escape bandwidth offered by co-packaged optics can enable high-radix switch implementations of more than 150 switch ports, which can be combined with data rates of up to 400 Gb/s per port. From the network architecture perspective, the key benefits of using co-packaged optics in future fat-tree networks include (a) the ability to implement large-scale topologies of > --> 11 , 000 end points by eliminating the need for a third switching layer and (b) the ability to provide up to 4 × higher bisection bandwidth compared to existing solutions, reducing at the same time the number of required switch application-specific integrated circuits by > --> 80 % . From the network operation perspective, both reduced energy consumption and lower packet delays can be achieved since fewer hops are required; i.e., packets need to traverse fewer serializer/deserializer lanes and fewer switch buffers, which reduces the probability of contending with other packets and improves the tolerance of network congestion. The performance of the proposed architecture is evaluated via discrete-event simulations for a wide range of representative HPC synthetic-traffic cases that include both hotspot and non-hotspot scenarios. The simulation results suggest that co-packaged optics form a promising solution to keep up with bandwidth scaling in future networks, while the reduced number of switching layers can lead to significant mean packet delay improvements that start from 30% and reach up to 74% for high-load conditions.