Enhanced Dynamic Beamwidth Selection-based THz MAC Protocol for Wireless Data Center Networks⋆ Muhammad Absaruddina,∗ , Saim Ghafoora and Mubashir Husain Rehmanib a Department of Computing, Atlantic Technological University, Donegal, Ireland
arXiv:2607.17996v1 [cs.NI] 20 Jul 2026
b Department of Computer Science, Munster Technological University, Cork, Ireland
ARTICLE INFO
ABSTRACT
Keywords: Beamwidth Dynamic beamwidth Medium Access Control (MAC) protocol Terahertz Wireless Data Center
Terahertz (THz) wireless communication offers a promising alternative to traditional wired links in data centres (DCs), enabling ultra-high data rates, low latency, and greater scalability. However, THz signals suffer from high path loss, necessitating the use of directional antennas (DAs). While DAs enhance signal strength, they introduce challenges such as deafness and synchronisation, typically addressed through receiver-initiated MAC protocols. Most existing THz MAC protocols use fixed beamwidths, which results in a key performance trade-off: narrow beams improve gain for long-range links but reduce throughput for short distances due to increased alignment overhead, while wide beams benefit short links but degrade performance over longer distances. To overcome this limitation, we propose DBS-ADAPT, a dynamic beamwidth selection-based MAC protocol that adjusts the antenna beamwidth according to the distance between nodes, maximising throughput without compromising link range. We also introduce an enhanced version, EDBS-ADAPT, which further reduces beamwidth switching and control overhead while preserving throughput gains. Both protocols are evaluated using the NS-3 THz module. Simulation results show that DBS-ADAPT improves average throughput by up to 22% and reduces delay up to 10% compared to the baseline ADAPT-3 protocol. EDBS-ADAPT further cuts beamwidth switching overhead by 95%, making it more efficient for scalable and highperformance wireless DC environments.
1. Introduction Data centre (DC) traffic has been growing exponentially, driving an urgent need for higher bandwidth and more efficient data transmission [2]. Traditional wired infrastructure and network topologies struggle to keep pace with these demands due to their high deployment and maintenance costs, physical space requirements, and limited flexibility in reconfiguration. In contrast, wireless communication technologies offer a more flexible, scalable, and cost-effective alternative. They also enable dynamic resource allocation and improved adaptability compared to fixed wired links [3], [4]. Several wireless technologies have been explored for DC applications, including millimeter-wave (mmWave) [5], optical wireless communication (OWC)[6], and Terahertz (THz) communication [3]]. Among these, THz communication stands out as a highly promising candidate due to its ability to support ultra-high data rates and wide bandwidth, while offering enhanced physical-layer security compared to mmWave. Unlike OWC, THz communication experiences lower beam spreading and penetration losses, and requires less precise alignment, making it more robust to beam misalignment and tracking errors, which are key considerations in dynamic DC environments. Moreover, OWC systems are ⋆
A preliminary version of this paper has been published in 20th International Conference on Wireless and Mobile Computing, Networking and Communications (WiMob 2024) Paris, France, October 2024 [1]. This research was supported by Presidents Research Bursary Scholarship Award offered by Atlantic Technological University, Donegal, Ireland. ∗ Corresponding author [email protected] (M. Absaruddin) ORCID (s): 0000-0001-6569-6430 (M. Absaruddin)
M. Absaruddin et al.: Preprint submitted to Elsevier
more vulnerable to environmental challenges such as atmospheric absorption, turbulence, scintillation, and ambient light interference, which can significantly impair performance in practical settings [7], [8], [9], [10]. Terahertz wireless links in data centres enable terabitper-second (Tbps) data rates, ultra-high throughput, and low latency [9], [11]. They also support flexible, energy-efficient topologies that reduce deployment costs [12], [13], [14]. In DC environments, THz links can be used for both inter-rack (between Top-of-Rack switches) and intra-rack (within a rack) communication [15]. However, THz bands suffer from severe path losses, up to 82 dB at 300 GHz over just 1 meter, with losses increasing at higher frequencies and distances [9]. These losses can be mitigated using highly directional antennas (DAs) at both transmitter and receiver ends [8], [16], [17]. Directional antennas are essential in THz communication due to their ability to provide high antenna gain, improve signal-to-noise ratio, and reduce interference through spatially confined beams [3]. These benefits enhance link reliability, spectral efficiency, and energy efficiency while supporting secure and high-capacity communication. While essential for overcoming high path loss, DAs introduce several challenges. These include beam misalignment, directional deafness, and high overhead for beam training and tracking in static as well as dynamic and mobile scenarios [18], [13], [9]. Directional THz links are also highly susceptible to blockage and require precise synchronisation between nodes, complicating medium access control (MAC) and increasing protocol complexity in dense networks. These issues necessitate redesigning link-layer protocols to ensure robust and low-latency communication. Although rotating directional antennas can help address Page 1 of 17
Enhanced Dynamic Beamwidth Selection-based THz MAC Protocol for Wireless Data Center Networks
some of the challenges while increasing the benefits and usage. They introduced challenges such as frequent beam misalignment, increased overhead of beam tracking and switching, synchronisation difficulties, link stability, beam adjustment for coverage, and beam parameters optimisation. These trade-offs require careful protocol design to balance improved coverage and flexibility against increased complexity and energy consumption [19], [20], [9], [13]. Traditionally, MAC protocol design has concentrated on managing contention among nodes competing for channel access. However, in terahertz communications, the emphasis shifts toward MAC protocols that incorporate scheduling and coordination [13], [14]. In THz communication, we mainly have a transmitter (Tx) and a receiver (Rx) initiated MAC protocols [13]. Tx-initiated THz MAC protocols depend on the sender to start communication, which can cause higher collision rates in highly directional environments. In contrast, receiver-initiated protocols improve coordination and reduce unnecessary transmissions by having the receiver manage link establishment, helping address beam alignment and deafness challenges [13]. The existing receiver-initiated THz MAC protocols [21], [18], [22], [23], [24] predominantly use fixed beamwidth antennas, without accounting for beamwidth dynamics. However, beamwidth selection presents important tradeoffs: wider beams offer higher throughput and lower delay but at the expense of reduced range, whereas narrower beams provide longer range but result in lower throughput and higher delay [25]. Using a fixed beamwidth limits the ability of nodes to adapt to varying distances, leading to suboptimal performance, particularly in terms of throughput and latency. To communicate efficiently, nodes should dynamically adjust their beamwidth based on distance; yet, current receiver-initiated protocols do not incorporate distanceaware beamwidth adaptation. A balanced strategy is needed for beamwidth selection during the initial (discovery) and transmission stage [25]. To enable dynamic beamwidth selection, we previously proposed a protocol called Dynamic Beamwidth Selectionbased THz MAC protocol (DBS-ADAPT) [1], designed to enhance the performance of the existing receiver-initiated THz MAC protocol, ADAPT. DBS-ADAPT demonstrated improved average throughput and reduced average delay by allowing both client and Access Point (AP) nodes to use wider beamwidths during the data transmission phase [1]. Despite its advantages, DBS-ADAPT has certain limitations: it incurs significant overhead due to frequent antenna beamwidth switching and control packets transmission and involves repeated beamwidth selection processes. In this paper, we present an enhanced version of DBSADAPT, called EDBS-ADAPT, which significantly reduces beamwidth switching and control overhead while maintaining the throughput and delay benefits of the original protocol. In our earlier work, a bit error rate (BER) of 10−6 was assumed, which negatively impacted the overall reliability and Quality of Service (QoS) for data centres. Furthermore, the protocol was based on physical layer parameters from the M. Absaruddin et al.: Preprint submitted to Elsevier
now-outdated IEEE 802.15.3e-2017 THz standard, and its performance was not evaluated under realistic, bursty traffic patterns typically observed in DC environments.
1.1. Contributions The main contributions of our work are summarised as follows: 1. Dynamic Beamwidth Selection protocol: In this work, we propose a dynamic beamwidth selection protocol for terahertz (THz)-based data center networks, referred to as DBS-ADAPT. This protocol is designed to enhance link performance by adaptively adjusting the beamwidth according to the distance between nodes. The foundational version of this protocol was initially introduced in [1], where its core mechanisms and preliminary results were discussed. 2. Proposal of EDBS-ADAPT for Improved Efficiency: To address the limitations identified in DBS-ADAPT, we propose an Enhanced Dynamic Beamwidth Selectionbased THz MAC protocol (EDBS-ADAPT). Designed specifically for DC environments, EDBS-ADAPT significantly reduces antenna beamwidth switching and control overhead by eliminating the repetition of the beamwidth selection process, while preserving the throughput and delay advantages of the original protocol. 3. Extension of DBS-ADAPT with Realistic Parameters: We extend our previously proposed Dynamic Beamwidth Selection-based THz MAC protocol [1] by incorporating more realistic system-level parameters. Specifically, we adopt a BER of 10−12 , aligning with ITU technical specifications, ETSI recommendations, and IEEE standards for THz data centre requirements [26], [27], to ensure ultra-reliable and efficient communication where data integrity is critical. 4. Adoption of the Latest Standard and Modulation Schemes: The performance of DBS-ADAPT is further evaluated using physical layer parameters based on the most recent IEEE 802.15.3-2023 THz standard. This includes the integration of higher-order modulation schemes such as 16-APSK and 32-APSK, allowing for improved data rates and spectral efficiency. 5. Scalability Analysis: The scalability analysis is conducted for the proposed protocols by evaluating with a larger data-center topology of twenty five nodes. 6. Evaluation with Realistic Traffic Models: The protocol is assessed under a Poisson traffic model as well as a realistic data centre traffic pattern characterised by bursty traffic. This ensures a more accurate representation of actual DC network conditions and validates protocol performance under practical load variations. The structure of the paper is as follows: Section 2 reviews related work on THz MAC protocols. Section 3 introduces the system preliminaries. Section 4 describes the DBS-ADAPT protocol and its operational details. Section 5 presents the EDBS-ADAPT protocol, including its algorithmic implementation and design. Section 6 outlines Page 2 of 17
Enhanced Dynamic Beamwidth Selection-based THz MAC Protocol for Wireless Data Center Networks Ref Tx/Rx ini- Beamwidth DistanceTechnique/Features tiated Manageaware ment
Channel Ac- Strength cess Method
Limitations
[28] Tx initiated
Adaptive
No
Beam switching, En- TDMA ergy control, TDMA scheduling
Energy-efficient, beam Not suitable for bursty DC traffic flexibility due to fixed TDMA slots, synchronisation issues Assisted beamforming; Dual radio Improved beam align- Increased hardware cost, dual-radio coordination complexity, high optimal beamwidth ment, throughput antennas switching delays, large calculated from signal message overhead led to lower quality and distance Distance-based power Dual Radio Optimises power con- transmission speed and an increase in overall latency. Present chip allocation and CSMA sumption architecture and the standards Beam parameters ad- Dual radio Enhanced throughput, does not support coordination justed based on ex- and CSMA dynamic adaptability between the two radios changed beamwidth information
[29] Tx initiated
Adaptive
Yes
[30] Tx initiated [31] Tx initiated
Fixed Yes beamwidth Adaptive No beamwidth
[21] Rx initiated
Fixed No beamwidth
Receiver initiated align- CSMA ment, synchronization improvements
Overcomes deafness Fixed beamwidth: Wider beams and synchronization reduce range; narrow beams limit issues effectively throughput.
[22] Rx initiated
Fixed No beamwidth
Enhanced node rotation CSMA capability, extended link discovery
Increased flexibility, Still limited by fixed beamwidth isimproved client node sues (range-throughput trade-off) alignment
[24] Rx initiated
Fixed No beamwidth
Limited data center CSMA/TDMAQoS support specific optimizations
Insufficiently detailed MAC attributes, fixed beamwidth limitations remain
Table 1 THz MAC protocols related Works with techniques, strengths and limitations.
the simulation setup, including parameters, implementation details, and the performance metrics used for evaluation. Section 7 discusses the simulation results, comparing DBSADAPT and EDBS-ADAPT with the ADAPT-3 THz MAC protocol. Finally, Section 8 concludes the paper and presents the future work.
2. Related Works In this section, we review related work on THz MAC protocols, with a particular focus on dynamic beamwidth selection strategies for data centre networks. Existing THz MAC protocols are primarily categorised into Tx-initiated and Rx-initiated approaches [13]. In Tx-initiated protocols, the transmitter node initiates communication either by directly sending the data packet in a 0-way handshake or by first sending a control packet (e.g., Request to Send, RTS) in a 2-way handshake. Conversely, Rx-initiated protocols involve the receiver node initiating the communication by broadcasting a control packet indicating its readiness to receive data. The transmitter then responds by sending the data packet. This mechanism is particularly relevant in centralised network architectures typical of data centres, where the receiver acts as an Access Point (AP) and the transmitters are client nodes [13]. Several transmitter-initiated THz MAC protocols have been proposed that employ dual-radio designs or assisted beamforming to address challenges such as beam alignment and deafness in directional communication. TAB-MAC [29] M. Absaruddin et al.: Preprint submitted to Elsevier
introduces an assisted beamforming MAC protocol that uses a lower-frequency control plane to aid beam alignment in the THz band. While it improves beamforming accuracy and reduces setup delay, it relies on dual-radio hardware, which is not supported in current chip architectures and adds complexity [32]. MBPA-MAC [30] proposes a multi-beam, on-demand power allocation MAC protocol for MIMObased THz networks. It enhances spatial reuse and energy efficiency through adaptive power control across beams, but it inherits similar limitations due to its dependence on dual radios. DRA-MAC [31] presents a Dual-Radio-Assisted MAC protocol tailored for distributed THz networks, aiming to support robust MAC operations via cross-band communication. However, hardware cost, coordination challenges between radios, and antenna switching delays remain major concerns, especially for high-performance data centre environments [13]. RS-MAC [33] enhances link reliability using an intelligent reflecting surface (IRS)-based relay selection mechanism in a dual-channel MAC design, but it incurs high control overhead and added latency due to reliance on a separate low-frequency channel, limiting its scalability. These protocols offer innovative solutions to key MAC-layer issues but face limitations such as hardware incompatibility, increased cost, and higher latency, making them less suitable for ultra-fast, low-latency data centre applications. Several Rx-initiated THz MAC protocols ([21], [18], [22], [23], [24]) have been proposed for centralised networks, where the receiver (AP) starts communication by Page 3 of 17
Enhanced Dynamic Beamwidth Selection-based THz MAC Protocol for Wireless Data Center Networks
broadcasting a control packet, prompting the client transmitter to send data. In [18], a receiver-initiated MAC and linklayer synchronisation protocol for THz networks to mitigate deafness and improve timing coordination is proposed. The protocol assumes fixed beamwidths and ideal beam alignment, lacking adaptability for varying link distances and spatial conditions in data centres. In [21], the ADAPT protocol is introduced, which is an adaptive directional antenna MAC protocol that dynamically manages directional scheduling in THz links. ADAPT does not support dynamic beamwidth adjustment, which can limit performance in dense environments where adaptive spatial coverage is critical. In [22], an extension of the NS-3 simulator with THz link discovery capabilities, enabling simulation of Rx-initiated MAC behaviour, is presented. Focused on simulation infrastructure; does not propose or evaluate new MAC strategies or beam adaptation mechanisms. [23] further enhances ns-3’s Terasim module to enable TCP communication over THz links, addressing transport-layer challenges. Concentrates on transport-layer support without incorporating adaptive MAC-layer techniques or dynamic beamwidth considerations. In [24], efficient synchronous MAC protocols for THz wireless data centres are proposed, aiming to reduce latency and improve throughput using coordinated scheduling. While tailored for data centres, the approach does not explicitly address dynamic beamwidth control or physicallayer variations such as real-time BER adaptation. ADAPT-1 and ADAPT-3 are receiver-initiated THz MAC protocols designed for short-range applications like WPAN and WLAN to achieve link-layer synchronisation [21]. ADAPT-1 uses a 1-way handshake and is an extension of the LLS-MAC protocol [18], while ADAPT-3 uses a 3way handshake and has been enhanced for node discovery [22] and downlink communication [23]. Both protocols use fixed beamwidths, where wider beams reduce range and narrower beams limit throughput. Previous studies [14] compared these protocols for wireless data centres, and further analysis [25] highlighted the trade-offs involved in antenna beamwidth selection. In [28], a PLBS-MAC, which is a pulse-level beam switching scheme with adaptive beamwidth, is presented. However, its contention-free TDMA approach struggles to handle the bursty traffic common in data centres. The THz MAC protocols proposed for other applications include TAN-MAC for airborne networks [34], which is for long-range covering a 500-meter distance, and CSMA-IIOT for industrial internet of things [35]. Recent studies ([18], [21]) show that receiver-initiated protocols achieve higher throughput and better synchronisation than transmitter-initiated ones. This is mainly due to improved coordination, which is crucial when using directional antennas at both ends to address high path loss in THz communication. The limitations of these works include reliance on fixed beamwidths and a lack of dynamic beam adaptation, which reduces efficiency in varying link conditions typical of data centres. Additionally, some focus primarily on simulation or transport-layer aspects without
M. Absaruddin et al.: Preprint submitted to Elsevier
Figure 1: Centralized wireless DC topology with AP and client nodes as ToR THz nodes.
fully addressing MAC-layer challenges like real-time coordination, control overhead, and bursty traffic handling. The shortcomings found in these MAC protocols are mentioned in Table 1. This work proposes a novel THz MAC protocol for data centres that dynamically selects antenna beamwidth based on the distance between nodes, enhancing communication reliability and efficiency. It introduces the first distanceaware, dynamic beamwidth selection algorithm within a receiver-initiated THz MAC protocol, ensuring all nodes are discovered and communication links are quickly established. By using a single THz radio, the protocol achieves ultra-high throughput and reduced delay, making it well-suited for nextgeneration data centre environments.
3. System Preliminaries In this section, the system model with the channel model is presented.
3.1. System Model We consider a wireless data centre network composed of a single access point (AP) and 14 client nodes, each equipped with a single THz transceiver and a steerable directional antenna. As illustrated in Figure 1,, the topology comprises of three rows of racks with five racks per row, totaling 15 racks. Each rack is equipped with a single THz transceiver and a steerable directional antenna (ToR), and the focus of this simulation is on inter-rack communication. The horizontal distance between adjacent racks within a row is 1.2 meters, while each rack measures 0.58 meters in width and 0.67 meters in depth. In this topology, a THz access point is centrally located at the ToR of one rack, with the assumption that no client node resides in the same rack as the AP. Nodes are deployed in a top-of-rack configuration and are static. Each node performs a 360-degree sweep to discover and align with intended communication partners [36]. Page 4 of 17
Enhanced Dynamic Beamwidth Selection-based THz MAC Protocol for Wireless Data Center Networks
3.1.1. Assumptions To design and evaluate the proposed DBS-ADAPT and EDBS-ADAPT protocols, we base our model on the following realistic and practical assumptions: • Directional Antenna Deployment: Both the access point and all client nodes are equipped with steerable directional antennas mounted at the top-of-rack level. These antennas are capable of beam rotation and dynamic beamwidth adjustment based on node distance. • Receiver-Initiated Medium Access: The protocol employs a receiver-initiated three-way handshake using CTA (Call-to-Access), RTS (Request to Send), and CTS (Clear to Send) frames. This handshake mechanism ensures that only one node is granted transmission access per time slot, eliminating contention and avoiding simultaneous transmissions from multiple clients. • Exclusive Time Slot Allocation: After the AP sends a CTS to a selected client, all other client nodes are restricted from transmitting until the time slot ends. This guarantees exclusive use of the wireless channel, thereby preventing beam overlap and co-channel interference. • Distance-Aware Beamwidth Control: Beamwidth is dynamically adapted based on the estimated distance between communicating nodes. This limits unnecessary spatial coverage and further minimises the probability of unintended signal reception. • Static Node Placement: All nodes are assumed to be static, which is typical in data centre environments. This assumption simplifies beam tracking and alignment, as node positions do not change over time. It is assumed that the nodes are already aware of their perfect coordinates. • Line-of-Sight (LoS) Availability: The propagation environment is assumed to provide clear LoS paths between racks, which is a reasonable assumption given the structured and obstruction-free layout of modern data centres. • Interference Negligibility: The combination of directional antennas, exclusive time slot access, and dynamic beamwidth control ensures that interference is minimal, and spatial reuse is efficiently managed. • Single-Radio Architecture: Each node, including the AP, operates with a single THz radio, eliminating the need for additional hardware and maintaining low system complexity. • Synchronised Network Operation: All nodes are timesynchronised to a global clock, allowing precise coordination of access slots and handshake operations.
Figure 2: Mechanism of DBS-ADAPT in which both AP and client nodes changes beamwidth during data transmission.
3.2. THz Channel Model In this work, we adopt the THz channel model presented in [37], which accounts for two key propagation phenomena: free-space spreading loss and molecular absorption loss [[37], [9], [8]]. The received power is calculated as [37]: 𝑃 𝑜𝑤𝑒𝑟𝑟 (𝑑) =
∫𝐵
𝑃 𝐷𝑡 (𝑓 )|𝐿𝑐 (𝑓 , 𝑑)|2 𝐺𝑎𝑖𝑛𝑡 (𝑓 )𝐺𝑎𝑖𝑛𝑟 (𝑓 )𝑑𝑓 , (1)
where 𝐿𝑐 (𝑓 , 𝑑), the channel frequency response, is determined based on the losses, 𝐺𝑎𝑖𝑛𝑡 , 𝐺𝑎𝑖𝑛𝑟 represent the transmitter and receiver antenna gains, respectively, and 𝑃 𝐷𝑡 denotes the one-sided power spectral density of the transmitted signal.
4. Dynamic Beamwidth Selection-based THz MAC Protocol (DBS-ADAPT) The DBS-ADAPT [1] THz MAC protocol is an enhancement of the ADAPT-3 protocol, specifically designed for wireless data centre environments. It is a receiverinitiated MAC protocol that overcomes the limitations of fixed beamwidth antennas by dynamically adjusting the antenna beamwidth at both the Access Point and client nodes during the data transmission phase. Initially, narrow beams are employed during link discovery to maximise node detection and alignment accuracy. After successful link establishment and control packet exchange, the AP calculates the distance to the client node and instructs both nodes to adapt their beamwidths dynamically, widening the beam for data transmission if the distance is within range constraints. This distance-aware beamwidth adaptation balances communication range and throughput, enabling efficient, highspeed data transfer without compromising link reliability.1
4.1. Protocol description and flow In DBS-ADAPT, each client node transmits its positional coordinates to the AP during the handshake phase. Using 1 We presented a preliminary version of DBS-ADAPT in WiMob 2024 [1], this article introduces a fully extended and refined version.
M. Absaruddin et al.: Preprint submitted to Elsevier
Page 5 of 17
Enhanced Dynamic Beamwidth Selection-based THz MAC Protocol for Wireless Data Center Networks
Algorithm 1: Algorithm of DBS-ADAPT [1] Input: Δ𝜃 (Beamwidth), 𝑅(𝜙) (Range), Φ - Different beamwidths, 3◦ ,6◦ ,9◦ ,12◦ ,15◦ ,18◦ ,21◦ . . . dynamic 2 Output: Δ𝜃 1: Let Δ𝜃 = {Φ1 , Φ2 , … , Φ𝑛 } 2: Δ𝜃narrow ← Narrower beamwidth 3: Δ𝜃wider ← Wider beamwidth 4: (1) CN and AP operations using Δ𝜃narrow 5: Set initial Δ𝜃 = min{Φ} for CN and AP during discovery 6: AP sending the broadcast CTA packet for each sector in AP coverage area do 1
Figure 3: Antenna beamwidth inversely proportional to virtual antenna sectors and gain.
this information, the AP calculates the distance, and both the AP and client dynamically adjust their beamwidths based on that distance. Designed for static wireless data centre topologies, DBS-ADAPT accommodates clients positioned at varying distances from the AP, as shown Table 2. The detailed procedure is outlined in Algorithm 1. The DBS-ADAPT protocol operates in four distinct stages, each described below. An overview of the DBSADAPT mechanism is illustrated in Figure 2. 1. Stage 1: Discovery Stage and Initial Link Establishment Communication begins with the Access Point initiating the discovery process by broadcasting a Call-to-Access (CTA) packet while sweeping its directional antenna. To maximise communication range, the AP initially uses a narrow beamwidth of 6◦ , which provides higher antenna gain. At this stage, client nodes are unaware of the AP’s location. As the AP sweeps, any client node that receives the CTA packet aligns its antenna toward the direction of the AP using the same fixed narrow beamwidth of 6◦ for accurate and long-range detection. This ensures reliable initial link establishment between the AP and client nodes. A wider beamwidth cannot reach distant nodes effectively due to lower antenna gain, while using an extremely narrow beamwidth, such as 3◦ , significantly increases the time required for a full 360◦ sweep. Therefore, a 6◦ beamwidth is chosen as an optimal balance, narrow enough to maintain sufficient gain for reaching the farthest node in the network, yet wide enough to complete discovery efficiently without missing any client nodes. This effect is also illustrated in Figure 3 and Table 2. 2. Stage 2: Exchange of Positional Information and Distance Calculation Once a client node aligns its antenna toward the AP and has data to transmit, it initiates the handshake process by sending a Request to Send (RTS) packet using a narrow 6◦ beamwidth. The RTS packet includes the client’s positional coordinates. Upon receiving the RTS, the AP extracts the coordinates and computes the Euclidean distance to
M. Absaruddin et al.: Preprint submitted to Elsevier
7:
Broadcast CTA → Δ𝜃narrow
8: end for 9: CN Operation of discovering the AP
for each CN do 10:
if R ECEIVE(CTA) then 11:
A LIGNA NTENNA(towards AP) 12: end if 13: end for 14: CN sending RTS to AP by including its position
for each RTS sent by CN do 15:
Include position in RTS packet ← (𝑥𝑛 , 𝑦𝑛 )
16: end for 17: (2) AP operation, it calculates distance based on position
of CN, and mentions in CTS packet for each RTS received by AP node do 18:
Calculate 𝑑 between AP and CN.
19: AP coordinates ← (𝑥𝐴𝑃 , 𝑦𝐴𝑃 ) 20: X ← 𝑥𝐴𝑃 − 𝑥𝑛 , Y ← 𝑦𝐴𝑃 − 𝑦𝑛
√
21: 𝑑 (distance) ← X2 + Y2 22: Select Δ𝜃 dynamic = max{Φ} while ensuring 𝑑 ≤ 𝑅(𝜙) 23: Set AP Δ𝜃 = Δ𝜃 dynamic 24: end for
for each CTS sent by AP node do 25:
IncludeΔ𝜃 dynamic in CTS. 26: end for 27: (3) CN Δ𝜃 → Δ𝜃wider as mentioned by AP
for each CTS received by client node do 28:
Extract CTS 29: Set CN Δ𝜃 = Δ𝜃 dynamic for data transmission 30: end for
the client node using Equation (5), as shown in Line 21 of Algorithm 1. 𝑑=
√
(𝑥AP − 𝑥𝑛 )2 + (𝑦AP − 𝑦𝑛 )2
(2)
where 𝑑 is the Distance in meters (m), (𝑥𝐴𝑃 , 𝑦𝐴𝑃 ) are AP coordinates, and (𝑥𝑛 , 𝑦𝑛 ) are Client node coordinates. This distance is then used to determine the appropriate beamwidth for efficient data transmission. 3. Stage 3: Dynamic Beamwidth Selection for AP and Client Nodes Based on the calculated distance, Page 6 of 17
Enhanced Dynamic Beamwidth Selection-based THz MAC Protocol for Wireless Data Center Networks Client node ID
Node Node Node Node Node Node Node Node Node Node Node Node Node Node 1 2 3 4 5 6 7 8 9 10 11 12 13 14
Distance to AP
2.2𝑚
1.95𝑚 1.86𝑚 1.95𝑚 2.2𝑚
1.16𝑚 0.58𝑚 0.58𝑚 1.16𝑚 2.2𝑚
1.95𝑚 1.86𝑚 1.95𝑚 2.2𝑚
BPSK selected Δ𝜃
24◦
24◦
24◦
24◦
24◦
30◦
48◦
48◦
30◦
24◦
24◦
24◦
24◦
24◦
QPSK selected Δ𝜃
18◦
18◦
18◦
18◦
18◦
24◦
42◦
42◦
24◦
18◦
18◦
18◦
18◦
18◦
8PSK selected Δ𝜃
◦
12
12
◦
◦
12
◦
12
12
◦
◦
18
◦
30
30
◦
18
◦
◦
12
◦
12
12
◦
◦
12
12◦
16QAM selected Δ𝜃
12◦
12◦
12◦
12◦
12◦
18◦
24◦
24◦
18◦
12◦
12◦
12◦
12◦
12◦
16APSK selected Δ𝜃
◦
12
12
◦
◦
12
◦
12
12
◦
◦
18
◦
24
24
◦
18
◦
◦
12
◦
12
12
◦
◦
12
12◦
32APSK selected Δ𝜃
◦
12
◦
◦
◦
12
◦
◦
◦
18
◦
12
◦
◦
◦
12
◦
◦
12◦
12
12
12
12
18
12
12
12
Table 2 Details of beamwidth Δ𝜃 selected by each of the client nodes concerning their distance and different modulation and coding schemes for DBS-ADAPT and EDBS-ADAPT THz MAC protocol.
the AP selects an optimal beamwidth that guaran5. Enhanced Dynamic Beamwidth tees the communication range exceeds the required Selection-based THz MAC Protocol distance, as implemented in Line 22 of Algorithm (EDBS-ADAPT) 1. This ensures reliable connectivity while allowWe propose an enhancement to DBS-ADAPT called Ening faster antenna rotation, which contributes to imhanced DBS-ADAPT (EDBS-ADAPT), a receiver-initiated proved throughput. The AP then includes the selected THz MAC protocol specifically designed for wireless data beamwidth in the Clear to Send (CTS) packet sent to centres. EDBS-ADAPT follows the same core mechanism the client node. Upon receiving the CTS, the client as DBS-ADAPT, with one key difference: in DBS-ADAPT, adjusts its antenna beamwidth to match the value beamwidth calculation is included in every control packet specified by the AP. As a result, both the AP and exchange, causing it to be repeated for each data packet client operate using the same dynamically selected transmission. This results in frequent beamwidth switching beamwidth. The specific beamwidth values correand increased overhead. In contrast, EDBS-ADAPT selects sponding to various distances and Modulation and the beamwidth only once at the start of communication, Coding Schemes (MCS) are detailed in Table 2. 4. Stage 4: Data Transmission with Adjusted Beamwidth maintaining it throughout the session. This significantly reduces beamwidth switching overhead. Figure 2, also repFor data transmission, the client node uses the newly resents the EDBS-ADAPT mechanism the only difference is adjusted beamwidth determined during the handshake that both client and AP nodes selects the beamwidth only phase. Once the AP successfully receives the data once, and maintains it throughout the session. packet, it responds with an acknowledgement (ACK) packet. By using dynamically selected beamwidths, 5.1. Protocol description and flow both the AP and client can rotate their antennas more To illustrate the decision-making process of the EDBSquickly, leading to faster link alignment, increased ADAPT THz MAC protocol, Algorithm 2 outlines the decommunication opportunities, reduced transmission tailed steps. The protocol’s operation is divided into several delay, and overall improved throughput. stages, as described below. The protocol begins with a narrow beamwidth during 1. Stage 1: Initial link establishment with narrow link discovery to ensure all nodes, regardless of distance, beams The access point initiates communication can establish connections while maximising communication using a fixed beamwidth of 6◦ , broadcasting CTA range. Then, during control packet exchange, the AP selects packets while rotating through all sectors to signal its an optimal beamwidth based on the client’s distance, and the readiness to receive data. Since client nodes are iniclient adjusts accordingly, repeating this process for every tially unaware of the AP’s location, each client enters data transmission to maintain efficiency. a discovery phase. During this phase, the client uses DBS-ADAPT improves performance primarily through the same fixed beamwidth and performs exhaustive dynamic beamwidth adjustment. In this protocol, the access scanning by rotating its antenna across all sectors, point determines the appropriate beamwidth based on the pausing in each one to listen for a CTA packet. When distance to the target client node. The client then adjusts its a client detects a CTA packet in a particular sector, it beamwidth according to the value received from the AP. This stops scanning, aligns its antenna in that direction, and adjustment occurs for every data packet transmission, leadtransitions to the communication phase to begin data ing to frequent beamwidth switching and increased control exchange with the AP. overhead. While DBS-ADAPT achieves better throughput 2. Stage 2: RTS transmission and position exchange and delay performance, it does so at the cost of higher The client node generates an RTS packet that includes overhead. To mitigate this issue, an improved version called its position as coordinates (𝑥𝐴𝑃 , 𝑦𝐴𝑃 ). Upon receiving Enhanced DBS-ADAPT is proposed. M. Absaruddin et al.: Preprint submitted to Elsevier
Page 7 of 17
Enhanced Dynamic Beamwidth Selection-based THz MAC Protocol for Wireless Data Center Networks
Algorithm 2: Algorithm of EDBS-ADAPT 1 2
Input: Δ𝜃 (Beamwidth), 𝑅(𝜙) (Range), Φ - Different Δ𝜃 Output: Δ𝜃 dynamic 1: Let Δ𝜃 = {Φ1 , Φ2 , … , Φ𝑛 } 2: Δ𝜃narrow ← Narrower beamwidth, Δ𝜃wider ← Wider beamwidth 3: (1) CN and AP operations using Δ𝜃narrow 4: Set initial Δ𝜃 = min{Φ} for CN and AP during discovery 5: AP sending the broadcast CTA packet for each sector in AP coverage area do 6:
Broadcast CTA → Δ𝜃narrow
for each CN do 9:
if R ECEIVE(CTA) then 10:
A LIGNA NTENNA(towards AP) 11: end if 12: end for 13: CN sending RTS to AP by including its position
for each RTS sent by CN do Include position in RTS packet ← (𝑥𝑛 , 𝑦𝑛 )
15: end for 16: (2) Beamwidth Learning and One-Time Switching (at AP)
for each RTS received by the AP node do 17:
Calculate 𝑑 between AP and CN.
18: AP coordinates ← (𝑥𝐴𝑃 , 𝑦𝐴𝑃 )
• If not previously learned: The AP selects an appropriate beamwidth based on the client’s distance and link quality, then stores this information in the mapping table. • If already learned: The AP retrieves the previously stored beamwidth value for the client.
7: end for 8: CN Operation of discovering the AP
14:
the RTS packet, the AP extracts the client’s coordinates and calculates the distance to the client node. 3. Stage 3: AP learning and beamwidth assignment The AP consults its beamwidth mapping table to determine whether an optimal beamwidth for the client has already been established:
√
19: X ← 𝑥𝐴𝑃 − 𝑥𝑛 , Y ← 𝑦𝐴𝑃 − 𝑦𝑛 𝑑 (distance) ← X2 + Y2 20: AP checks if a beamwidth is already stored for this Client: 21: If no beamwidth is stored for this Client:
The AP then sends a Clear-To-Send packet to the client, including the assigned beamwidth value. 4. Stage 4: One-time beamwidth adjustment Upon receiving the CTS packet, the client extracts the assigned dynamic beamwidth and immediately adjusts its antenna to this value for all future data transmissions to the AP. It also sets a beamwidth adjustment flag to prevent further changes during the session. Similarly, before receiving the first DATA packet from the client, the AP switches its antenna to the selected dynamic beamwidth for that specific client and sets a corresponding per-client adjustment flag. The dynamic beamwidth values selected based on distance and various modulation and coding schemes are detailed in Table 2.
6. Simulation details
In this section, we describe our simulation setup, parameters, implementation details, and performance metrics. The simulations are conducted using Terasim, an open-source • Set AP Δ𝜃 = Δ𝜃 dynamic THz network simulator built as an extension of NS3 [38]. ∗ • AP stores Δ𝜃AP,Client ← Δ𝜃 ∗ for this client in a table. The specific simulation parameters are listed in Table 3. Packet arrivals are modelled using a Poisson process [21], 22: end for 23: If beamwidth already stored for this Client: chosen for its effectiveness in representing random traffic ∗ patterns. A log-normal traffic model is also used for bursty • AP retrieves Δ𝜃AP,Client from the table. traffic. All simulation results presented in Section 7 are ∗ 24: AP includes Δ𝜃AP,Client in the CTS reply to Client. AP switches beamwidth for this Client only once at first reported with a 95% confidence interval. • AP selects Δ𝜃 dynamic = max{Φ} while ensuring 𝑑 ≤ 𝑅(𝜙)
data exchange; subsequent transmissions reuse stored value. 25: (3) Beamwidth Switching at Client (One-Time)
6.1. Simulation Parameters
The performance evaluation uses various modulation and coding schemes defined in the recent THz standard IEEE • If the Client has not switched beamwidth before: 802.15.3-2023. These include Binary Phase Shift Keying ∗ (BPSK), Quadrature Phase Shift Keying (QPSK), 8-Phase – Client switches its antenna to Δ𝜃Client . – Client marks "beamwidth switched" (flag); future Shift Keying (8-PSK), 16-Quadrature Amplitude Modulation (16-QAM), 16-Amplitude and Phase Shift Keying (16data exchanges use the same beamwidth. APSK), and 32-Amplitude and Phase Shift Keying (32• If already switched, continue using stored beamwidth. APSK) [39]. For data centre applications, a bit error rate 26: (4) Data Transmission (No Further Switching) (BER) of 10−12 is adopted, as recommended by the IEEE • Data and ACK packets are exchanged using the Δ𝜃 = standard, ITU technical parameters, and ETSI guidelines [27], [26] [13] [40]. Additionally, the performance is evalΔ𝜃 dynamic optimal beamwidth. • No further beamwidth changes are performed for either uated by varying interarrival times to analyse how the protocol behaves under different traffic loads, reflecting the high the AP or the Client node variability typical of data centre traffic [41]. The interarrival time is calculated as described in Equation 6. ∗ • Upon receiving CTS, reads the Δ𝜃Client (from CTS).
M. Absaruddin et al.: Preprint submitted to Elsevier
Page 8 of 17
Enhanced Dynamic Beamwidth Selection-based THz MAC Protocol for Wireless Data Center Networks Description Simulator Technology Frequency Range Central Frequency Channel bandwidth Transmission power Bit error rate (BER) Simulation time for each run Simulation Runs Noise figure Data Packet Size Retry limit Traffic Model Modulation and Coding Schemes
Value Terasim - NS3 [38] IEEE 802.15.3-2023 [39] 252.72 321.84 GHz [39] 287.28 GHz [39] 69.12 GHz [39] 10dBm [27] 10−12 [26], [27], [13] 0.01 seconds 20 7dB 65 kB 5 Poisson [21], Log Normal BPSK, QPSK, 8PSK, 16QAM, 16APSK, and 32APSK [39].
Table 3 Simulation Parameters
𝑃 =
𝑇 𝐼
(3)
where 𝑃 is number of data packets, 𝑇 is the simulation duration (in seconds), and 𝐼 is the inter-arrival time (in seconds). Another important parameter in the evaluation is the antenna turning speed, which depends on the beamwidth and the time required to cover a sector. The turning speed refers to how quickly the antenna completes a full 360◦ rotation. This speed influences how fast the antenna can sweep its coverage area, directly impacting throughput and delay. The antenna turning speed increases proportionally with the beamwidth. In other words, antennas with wider beamwidths rotate faster than those with narrower beamwidths, allowing nodes to complete rotations more quickly. The antenna turning speed is calculated using the following expression: 𝜔=
1 𝑇circle
(4)
where 𝜔 is the Antenna turning speed measured in rotations per second (degrees/sec), 𝑇circle is the total time for one complete antenna rotation, measured in seconds. The total rotation time 𝑇circle can be calculated as: 𝑇circle = 𝑇sector ⋅ 𝑁sector
(5)
where 𝑇sector is calculated as the sum of control packets duration, propagation time, SIFS, maximum backoff time, 360 . data transmission time, and 𝑁sector is calculated as beamwidth
6.2. Implementation Details For DBS-ADAPT, the Terasim simulator was modified to include client node position information in the RTS control packet header. Specifically, additional fields were added M. Absaruddin et al.: Preprint submitted to Elsevier
Figure 4: Centralized wireless DC topology with AP and client nodes as ToR THz nodes.
to the RTS packet to store the client’s 𝑥 − 𝑦 coordinates. This allows the AP to calculate the distance to the client node. Similarly, the CTS packet was updated with a new field to carry the beamwidth information, which is sent back to the client. A new mechanism was implemented at the MAC layer to enable dynamic beamwidth selection during data transmission. Additionally, the antenna module was modified to update the beamwidth, recalculate antenna gain whenever the Tx or Rx changes beamwidth, and adjust the antenna turning speed accordingly. For EDBS-ADAPT, we modified the NS3 THz MAC modules for both AP and client nodes to implement a lowoverhead beamwidth switching mechanism. Unlike previous approaches that adjusted beamwidth frequently, our design ensures that both AP and clients switch their antenna beamwidth only once per communication session. Specifically, after receiving the CTS packet with the optimal beamwidth, the client sets its antenna accordingly and maintains this setting throughout the session. This onetime adjustment is enforced in the client’s MAC module, significantly reducing protocol overhead. On the AP side, the protocol is optimised by maintaining a mapping between each client’s address and its learned optimal beamwidth. When an RTS packet is received, the AP either retrieves the stored beamwidth for that client or calculates a new one based on the client’s position and distance. This beamwidth is sent to the client via the CTS packet. After the first data transmission, the AP saves the beamwidth for future use, avoiding repeated adjustments. A per-client flag ensures that the AP switches beamwidth only once per communication session.
6.3. Performance Metrics The performance of our proposed DBS-ADAPT and EDBS-ADAPT THz MAC protocols is evaluated using three key metrics: average throughput, average delay, and beamwidth switching overhead, as described below. 1. Average throughput: This metric measures the effectiveness of data delivery. It represents the rate at Page 9 of 17
Enhanced Dynamic Beamwidth Selection-based THz MAC Protocol for Wireless Data Center Networks Interarrival 10 Time (𝜇s) No of 1000 data packets
100
200
300
400
500
600
800
1000
100
50
33
25
20
16
12
10
Table 4 Number of packets sent against each selected inter-arrival time.
which data packets are successfully transmitted by each client node. Average throughput is calculated by averaging the throughput of all individual client nodes, reflecting the overall efficiency of data transmission. It is typically measured in bits per second (bps) [25]. The average throughput is calculated as follows [25]:
(a) Average throughput with BPSK, QPSK, and 8PSK MCS.
𝑁
𝑇avg =
1 ∑ 𝐿𝑛 𝑁 𝑛=1 𝑡𝑛
(6)
Where 𝑇avg is the average throughput (in bps or Gbps), 𝑁 is the total number of client nodes, 𝐿𝑛 is the total size of packets transmitted by node 𝑛 (in bits), and 𝑡𝑛 is the total transmission time for node 𝑛 (in seconds). 2. Average delay: This metric is measured from the moment a packet is queued for transmission until it is successfully received by the destination node. It accounts for transmission time, queuing delays, MAC protocol processing time, and propagation delay. Average delay is a key indicator of the protocol’s efficiency and QoS in THz communication systems [42]. The average delay is calculated as follows [42]: 𝑁
𝐷avg =
) 1 ∑( 𝑇end,𝑖 − 𝑇start,𝑖 𝑁 𝑖=1
(7)
where 𝐷avg is the average delay measured in milliseconds (ms), 𝑇end,𝑖 is the time when the transmission of the 𝑖-th packet is completed. 3. Beamwidth switching overhead: This metric represents the total number of times a node changes its beamwidth during communication.
7. Simulation Results In this section, we present the performance results of DBS-ADAPT and EDBS-ADAPT compared to the baseline THz MAC protocol, ADAPT-3 [21], under both Poisson and bursty traffic models.
7.1. Simulation Results with Poisson Traffic Model DBS-ADAPT is evaluated against ADAPT-3 [21] using modulation and coding schemes (Section 5.3) and a BER of 10−12 , as specified in the latest IEEE 802.15.3-2023 standard. This comparison demonstrates the effectiveness of DBS-ADAPT under realistic THz communication parameters for data centres. EDBS-ADAPT is further compared M. Absaruddin et al.: Preprint submitted to Elsevier
(b) Average delay with BPSK, QPSK, and 8PSK MCS.
Figure 5: EDBS-ADAPT 14 nodes case with varying interarrival time.
against both DBS-ADAPT and ADAPT-3. The proposed protocols are analysed based on varying interarrival times, number of nodes, and modulation schemes. In ADAPT-3, a fixed beamwidth of 6 degrees is used for both the discovery and data transmission phases. This beamwidth was chosen based on evaluations within a data centre topology, offering a balance between range and antenna gain to mitigate high path loss and ensure stable performance.
7.1.1. EDBS-ADAPT, and DBS-ADAPT with varying inter-arrival times and modulation schemes EDBS-ADAPT and DBS-ADAPT are evaluated under varying interarrival times and modulation schemes to assess their performance under different traffic loads. Interarrival times, ranging from 10 to 1000 𝜇s, follow an exponential distribution and represent the time between the arrivals of two consecutive packets at a node. This variation helps capture the random and highly variable nature of data centre traffic [41]. For analysis, traffic is categorised into high and low traffic scenarios: high traffic corresponds to interarrival times between 1-299 𝜇s, while low traffic ranges from 3001000 𝜇s. Table 4 presents the number of data packets sent by each client based on these interarrival times.
Page 10 of 17
Enhanced Dynamic Beamwidth Selection-based THz MAC Protocol for Wireless Data Center Networks
In addition to traffic variability, performance is evaluated across six modulation and coding schemes (MCS): BPSK, QPSK, 8-PSK, 16-QAM, 16-APSK, and 32-APSK. These MCSs are essential for enabling efficient and flexible communication in THz networks. Their varying encoding capabilities directly influence key performance metrics such as average throughput and delay. Higher-order modulation schemes, such as 16-QAM, 16-APSK, and 32-APSK, are particularly important for achieving the ultra-high throughput and low latency required in data centre environments.
Average Throughput and delay with varying interarrival times and modulation schemes: Figures 5(a), 5(b), 6(a), and 6(b) present the average throughput and delay results for EDBS-ADAPT, DBS-ADAPT, and ADAPT3 under a fixed 14-node setup with varying interarrival times. In scenarios using the low modulation scheme BPSK, EDBS-ADAPT and DBS-ADAPT consistently outperforms ADAPT-3 in terms of average throughput. For instance, under low traffic conditions with an interarrival time of 1000 𝜇s, EDBS-ADAPT and DBS-ADAPT achieves an average throughput of 32 Gbps, compared to 28 Gbps for ADAPT3, an improvement of approximately 15%. When the interarrival time is reduced to 500 𝜇s, throughput for EDBSADAPT and DBS-ADAPT is 26 Gbps versus 23 Gbps for ADAPT-3, marking a 13% gain. Even under high traffic conditions, where the interarrival time is as low as 10 𝜇s, EDBS-ADAPT and DBSADAPT maintains better performance. In this worst-case scenario, EDBS-ADAPT and DBS-ADAPT achieves an average throughput of 0.65 Gbps, compared to 0.63 Gbps for ADAPT-3, representing a 3% improvement. These results demonstrates EDBS-ADAPT and DBS-ADAPT superior ability to handle varying traffic loads, particularly in maintaining higher throughput under both light and heavy conditions. The average delay results for EDBS-ADAPT and DBSADAPT, is shown in Figure 5(b), indicate a consistent reduction compared to the ADAPT-3 THz MAC protocol when using the BPSK modulation scheme. Under low traffic conditions with an interarrival time of 1000 𝜇s, EDBSADAPT and DBS-ADAPT achieves an average delay of 26 ms, compared to 29 ms for ADAPT-3, a reduction of approximately 10%. With an interarrival time of 500 𝜇s, EDBSADAPT and DBS-ADAPT records a delay of 29 ms, while ADAPT-3 has 32 ms, showing a 10% improvement. Even under high traffic conditions (interarrival time of 10 𝜇s), EDBS-ADAPT and DBS-ADAPT reduces average delay by 9% compared to ADAPT-3. This trend continues across higher modulation schemes, as illustrated in Figures 6(a) and 6(b). In both high and low traffic scenarios, EDBS-ADAPT and DBS-ADAPT consistently outperforms ADAPT-3 in terms of average throughput and delay. These results demonstrate EDBS-ADAPT and DBS-ADAPT effectiveness in improving communication efficiency, particularly in reducing latency and maintaining
M. Absaruddin et al.: Preprint submitted to Elsevier
(a) Average throughput with 16QAM, 16APSK, and 32APSK MCS.
(b) Average delay OF DBS-ADAPT with 16QAM, 16APSK, and 32APSK MCS.
Figure 6: EDBS-ADAPT 14 nodes case with varying interarrival time.
performance across a range of traffic loads and modulation settings. EDBS-ADAPT and DBS-ADAPT consistently outperforms ADAPT-3 across all evaluated modulation schemes, BPSK, QPSK, 8-PSK, 16-QAM, 16-APSK, and 32-APSK, showing both higher average throughput and lower average delay. This performance holds across varying traffic loads, from low to high. The consistent improvement demonstrates that EDBS-ADAPT and DBS-ADAPT effectively addresses the unique challenges of different modulation levels: providing robustness for lower-order schemes and maintaining precision and efficiency for higher-order schemes. This adaptability makes EDBS-ADAPT and DBS-ADAPT well-suited for diverse communication requirements in THz-based data centre environments.
Beamwidth switching overhead with varying interarrival times and modulation schemes: Table 5 presents the beamwidth switching overhead for EDBS-ADAPT, DBSADAPT, and ADAPT-3 under a fixed 14-node setup with varying interarrival times. ADAPT-3 overhead is 0 for all the scenarios as it uses the fixed beamwidth. While, EDBSADAPT beamwidth switching overhead is minimised by Page 11 of 17
Enhanced Dynamic Beamwidth Selection-based THz MAC Protocol for Wireless Data Center Networks
(a) Average throughput with BPSK, QPSK and 8PSK MCS
(a) Average throughput with 16QAM, 16APSK, and 32APSK MCS.
(b) Average delay with BPSK, QPSK and 8PSK MCS
Figure 7: EDBS-ADAPT with varying nodes case and interarrival time 500 𝜇s
(b) Average delay with 16QAM, 16APSK, and 32APSK MCS.
Figure 8: EDBS-ADAPT with varying nodes case and interarrival time 500 𝜇s.
upto 95% for both the AP and client nodes for all the case as compared to DBS-ADAPT which have high beamwidth switching overhead.
7.1.2. EDBS-ADAPT, and DBS-ADAPT with varying number of nodes and modulation schemes The performance of the EDBS-ADAPT and DBS-ADAPT THz MAC protocols is evaluated by varying the number of nodes and applying different modulation and coding schemes, as detailed in Table 3. The number of nodes ranges from 2 to 14, while six MCS configurations are considered. For this evaluation, the interarrival time is fixed at 500 𝜇s, corresponding to a traffic load of 20 packets transmitted by each client node. The resulting average throughput and average delay for each protocol, across different node counts, are presented in Figures 7 and 8.
Average Throughput and delay with varying nodes and modulation schemes In the 2-node scenario using the lowest modulation scheme (BPSK), both EDBSADAPT and DBS-ADAPT achieved an average throughput of approximately 34 Gbps, while ADAPT-3 reached only 28 Gbps, as illustrated in Fig. 7(a). This represents a 21% increase in throughput for EDBS-ADAPT and DBS-ADAPT M. Absaruddin et al.: Preprint submitted to Elsevier
compared to ADAPT-3. While the average delay reduced by 6% compared to ADAPT-3 as shown in Fig.7(b). Similarly for the 10 nodes scenario, Additionally, these protocols reduced the average delay by about 5% as shown in Fig. 7(b). Similarly, in the 10-node scenario, EDBS-ADAPT and DBS-ADAPT demonstrated a 15% improvement in average throughput and a 7% reduction in average delay compared to ADAPT-3. In the 2-node scenario using the highest modulation scheme (32-APSK), EDBS-ADAPT and DBS-ADAPT achieved an average throughput of approximately 148 Gbps, while ADAPT-3 reached 137 Gbps, as shown in Fig. 8(a). This corresponds to a 8% cincrease in throughput for EDBS-ADAPT and DBS-ADAPT compared to ADAPT-3. Additionally, the average delay was reduced by about 4% as illustrated in Fig.8(b). A similar trend was observed in the 10-node scenario, where EDBS-ADAPT and DBS-ADAPT maintained a 5% higher average throughput and a 4% lower average delay compared to ADAPT-3.
Page 12 of 17
Enhanced Dynamic Beamwidth Selection-based THz MAC Protocol for Wireless Data Center Networks Table 5 Beamwidth-Switching Overhead 14 nodes case with varying inter-arrival time Interarrival Time 𝜇s
DBS-ADAPT client node overhead
EDBS-ADAPT client node overhead per node
EDBS-ADAPT node overhead
10 100 200 300 400 500 600 800 1000
210 90 47 31 24 19 16 12 10
1 1 1 1 1 1 1 1 1
210 90 47 31 24 19 16 12 10
AP
EDBS-ADAPT AP node overhead per node 1 1 1 1 1 1 1 1 1
Note: • DBS-ADAPT: DBS-ADAPT overhead is for both AP and client node and it is the average adjustment per node. • EDBS-ADAPT: Overhead is for per client node. • ADAPT-3: AP overhead = 0, Client overhead = 0 (constant for all interarrival times).
Table 6 Beamwidth-Switching Overhead with varying nodes case and interarrival time 500 𝜇s Name
AP node overhead
Client node overhead
Details
EDBS-ADAPT
1
1 (Per Client Node)
DBS-ADAPT
19 (Avg adjustment per node)
19 (Avg adjustment per node)
ADAPT-3 [21]
0
0
Each client node and the AP node adjust their beamwidth only once per node. Both the AP and each client node adjusted their beamwidth multiple times. Here 19 is the average number of adjustments per node. Both AP and client nodes use a fixed beamwidth.
Beamwidth switching overhead for varying nodes and modulation schemes: Table 6 compares the beamwidth switching overhead of EDBS-ADAPT, DBS-ADAPT, and ADAPT-3 in a fixed 14-node setup with varying interarrival times. ADAPT-3 incurs no overhead due to its fixed beamwidth, while EDBS-ADAPT reduces switching overhead by up to 94.74% for both AP and client nodes compared to the higher overhead of DBS-ADAPT. Figures 7 and 8 show that EDBS-ADAPT and DBSADAPT consistently achieve higher average throughput and lower average delay than ADAPT-3 across all tested modulation schemes (BPSK to 32-APSK) and node counts from 2 to 14. Figure 9: Topology for 25-node scalability scenario.
7.1.3. Scalability Analysis The performance of the EDBS-ADAPT and DBS-ADAPT THz MAC protocols is evaluated by increasing the number of nodes to 25 as shown in Fig. 9. In this scenario, there are 24 client nodes and 1 AP node. In this evaluation, the interarrival time is fixed at 500 𝜇s. This evaluation is conducted to determine if the expected throughput gains are sustained at a larger scale. The resulting average throughput and average delay for each protocol are presented in Figures 10(a) and 10(b). In M. Absaruddin et al.: Preprint submitted to Elsevier
this scenario, EDBS-ADAPT and DBS-ADAPT maintained a higher average throughput and a lower average delay compared to ADAPT-3. The results shows the advantages of the proposed protocol even in the case of scalability when the number of node increases.
Page 13 of 17
Enhanced Dynamic Beamwidth Selection-based THz MAC Protocol for Wireless Data Center Networks 70
60
50
40
30
20
10
0
100
200
300
400
500
600
700
800
900
1000
(a) Average throughput with 16QAM, 16APSK, and 32APSK MCS.
10 3
ADAPT-3 achieves lower average throughput compared to DBS-ADAPT and EDBS-ADAPT due to its use of a fixed beamwidth. Fixed beamwidth prevents nodes from adapting to varying distances, wider beams are better for nearby nodes to reduce delay, while narrower beams are ideal for distant nodes to maintain connectivity. This lack of adaptability leads to reduced throughput and increased delay, which is the primary cause of ADAPT-3’s performance decline in the THz MAC protocol. Overall, our detailed evaluations under varying interarrival time and varying nodes clearly shows that EDBSADAPT achieves higher throughput and less delay while EDBS-ADAPT achieves identical throughput and delay to DBS-ADAPT as it follows the same core mechanism as DBS-ADAPT. However, the advantage of our EDBSADAPT lies in substantially reduced antenna beamwidth switching overhead, while preserving the throughput and delay advantages of the original protocol.
7.2. Simulation Results with Bursty Traffic 10 2
0
100
200
300
400
500
600
700
800
900
1000
(b) Average delay with 16QAM, 16APSK, and 32APSK MCS.
Figure 10: EDBS-ADAPT, DBS-ADAPT and ADAPT-3 performance in the 25-node scalability scenario.
7.1.4. Discussion on Improved Performance The increased average throughput of EDBS-ADAPT and DBS-ADAPT is primarily due to its dynamic beamwidth selection during data transmission. By using a wider beamwidth, the number of sectors decreases, allowing the antenna to rotate faster and create more frequent communication opportunities between the AP and client nodes. This faster sector scanning reduces waiting times and delays, leading to higher throughput. The key factor driving this improvement is the antennas turning speed, which depends on the beamwidth and sector timing. Since turning speed is directly proportional to beamwidth (as shown in Fig. 4), wider beams enable quicker coverage of the communication area, positively impacting both throughput and delay. Higher-order modulation schemes achieve greater average throughput compared to lower-order schemes, mainly due to their higher physical layer data rates. For example, 32APSK yields the highest throughput, followed by 16-APSK, 16-QAM, 8-PSK, QPSK, and BPSK. This is because higherorder modulations encode more bits per symbol, allowing more data to be transmitted within the same time frame. Conversely, lower-order schemes like BPSK carry fewer bits, resulting in lower throughput despite their robustness. M. Absaruddin et al.: Preprint submitted to Elsevier
Data center traffic is often bursty, with periods of intense activity followed by idle intervals [43]. This burstiness impacts the performance of THz MAC protocols, which depend on precise timing and coordination between nodes. To simulate such traffic, the NS-3 Terasim simulator was modified to use a log-normal distribution in the traffic generator, accurately modeling the bursty behavior observed in data centers. The traffic generator class was specifically extended to produce packets based on this distribution [44]. In our simulation, the log-normal distribution parameters were set to (𝜇) = 5 and (𝜎) = 1, reflecting typical bursty traffic patterns observed in data center studies [43]. These values were chosen to capture moderate average interarrival times (𝜇) and high variability (𝜎), effectively modeling the unpredictable nature of data center traffic with bursts of high packet arrivals followed by idle periods [43]. Additionally, in the bursty traffic scenario we evaluated with 15 nodes scenario, in which there are 14 client nodes. Using the lowest modulation scheme (BPSK), EDBSADAPT and DBS-ADAPT consistently outperformed ADAPT3 in both throughput and delay. In the 2-node scenario, they achieved approximately 33 Gbps compared to ADAPT-3’s 27 Gbps, a 22% increase. For 10 nodes, the throughput gain was around 22% (28 Gbps vs. 23 Gbps). In terms of delay, EDBS-ADAPT and DBS-ADAPT reduced average latency by 9% in the 2-node case and by 10% in the 10-node case, as shown in Figures 11(a) and 11(b). At the highest modulation scheme (32-APSK), EDBSADAPT and DBS-ADAPT outperformed ADAPT-3 in both throughput and delay. In the 2-node scenario, they achieved approximately 144 Gbps, about 6% higher than ADAPT-3’s 136 Gbps, with an 4% reduction in average delay, as shown in Figures 12(a) and 12(b). For 10 nodes, throughput improved by 7%, and delay was reduced by 6% compared to ADAPT-3. In scenarios with varying numbers of nodes, average throughput remains stable due to the heavy-tailed nature of the log-normal distribution, which generates bursts of high Page 14 of 17
Enhanced Dynamic Beamwidth Selection-based THz MAC Protocol for Wireless Data Center Networks Description Traffic Model Simulation time for each run Simulation Runs Log-Normal (𝜇) Log-Normal (𝜎)
Value Log normal (bursty) [44] 0.01 seconds 20 5 1 [43]
Table 7 Summary of Log-normal bursty-traffic Parameters used in the simulation
(a) Average throughput with BPSK, QPSK and 8PSK MCS.
(a) Average throughput with 16QAM, 16APSK, and 32APSK MCS.
(b) Average delay with BPSK, QPSK and 8PSK MCS.
Figure 11: EDBS-ADAPT, DBS-ADAPT and ADAPT-3 with varying nodes case and bursty traffic.
(b) Average delay with 16QAM, 16APSK, and 32APSK MCS.
Figure 12: EDBS-ADAPT, DBS-ADAPT and ADAPT-3 with varying nodes case and bursty traffic.
activity followed by extended silent periods. This allows nodes to transmit packets quickly and then remain silent for a while. As the number of nodes increases, the likelihood of all nodes transmitting simultaneously stays low, reducing channel contention and maintaining throughput stability. Figures 11 and 12 show that EDBS-ADAPT and DBSADAPT consistently achieve higher throughput and lower delay than ADAPT-3 across all modulation schemes (BPSK to 32-APSK) and node counts from 2 to 14. This performance improvement is attributed to their dynamic beamwidth adjustment, unlike ADAPT-3’s fixed beamwidth approach.
M. Absaruddin et al.: Preprint submitted to Elsevier
8. Conclusion This paper proposes DBS-ADAPT, a dynamic beamwidth selection-based THz MAC protocol for wireless data centres, along with its enhanced version, EDBS-ADAPT. Both protocols overcome the limitations of fixed beamwidth by dynamically selecting the optimal beamwidth based on node distance, which improves throughput and reduces delay. By exchanging node position information, the protocols ensure that the chosen beamwidth maintains communication range without compromise, enhancing overall protocol efficiency.
Page 15 of 17
Enhanced Dynamic Beamwidth Selection-based THz MAC Protocol for Wireless Data Center Networks
The performance of DBS-ADAPT and EDBS-ADAPT was evaluated through simulations using the Terasim (NS3) THz simulator with realistic data centre parameters. The results demonstrate that both protocols achieve higher average throughput and lower delay compared to the baseline ADAPT protocol across various traffic loads, node counts, and modulation schemes. Additionally, EDBS-ADAPT reduces antenna overhead compared to DBS-ADAPT. Future work will focus on scaling the evaluation to larger data centre environments with multiple APs to test the protocols’ scalability and robustness. Additionally, we plan to extend our approach by incorporating an adaptive, distance-aware power allocation strategy to improve energy efficiency.
References [1] M. Absaruddin, S. Ghafoor, M. H. Rehmani, Dynamic beamwidth selection-based thz mac protocol for wireless data center networks, in: 20th International Conference on Wireless and Mobile Computing, Networking and Communications (WiMob), 2024, pp. 526–531. doi: {https://doi.org/10.1109/WiMob61911.2024.10770372}. [2] K. Geresu, H. Gu, M. Fadhel, W. Wei, X. Yu, Odrad: An optical wireless dcn dynamic-bandwidth reconfiguration with awgr and deep reinforcement learning, Optical Switching and Networking 52 (2024) 100771. doi:https://doi.org/10.1016/j.osn.2024.100771. [3] J. M. Eckhardt, T. Doeker, T. Kürner, Channel measurements at 300 ghz for low terahertz links in a data center, IEEE Open Journal of Antennas and Propagation (2024). doi:https://doi.org/10.1109/ OJAP.2024.3391798. [4] C. Castro, R. Elschner, T. Merkle, F. Rodrigues, J. Machado, A. Teixeira, C. Schubert, Thz communications: Paving the way towards wireless tbps, THz Communications. Springer (2022) 417–421doi: https://doi.org/10.1007/978-3-030-73738-2. [5] C. Terzi, I. Korpeoglu, 60 ghz wireless data center networks: A survey, Computer Networks 185 (2021) 107730. doi:https://doi. org/10.1016/j.comnet.2020.107730. [6] S. Zhang, R. Kraemer, X. Xue, N. Tessema, H. Freire Santana, E. Tangdiongga, N. Calabretta, Photonic integrated multicast switchbased optical wireless data center network, Journal of Optical Communications and Networking 15 (7) (2023) C54–C62. doi:https: //doi.org/10.1364/JOCN.484675. [7] S. Aboagye, M. A. Saeidi, H. Tabassum, Y. Tayyar, E. Hossain, H.-C. Yang, M.-S. Alouini, Multi-band wireless communication networks: Fundamentals, challenges, and resource allocation, IEEE Transactions on Communications (2024). doi:https://doi.org/10.1109/ TCOMM.2024.3366816. [8] H. Elayan, O. Amin, B. Shihada, R. M. Shubair, M.-S. Alouini, Terahertz band: The last piece of rf spectrum puzzle for communication systems, IEEE Open Journal of the Communications Society 1 (2019) 1–32. doi:https://doi.org/10.1109/OJCOMS.2019.2953633. [9] I. F. Akyildiz, C. Han, Z. Hu, S. Nie, J. M. Jornet, Terahertz band communication: An old problem revisited and research directions for the next decade, IEEE Transactions on Communications 70 (6) (2022) 4250–4285. doi:https://doi.org/10.1109/TCOMM.2022.3171800. [10] A. J. Alqaraghuli, J. V. Siles, J. M. Jornet, The road to high data rates in space: Terahertz versus optical wireless communication, IEEE Aerospace and Electronic Systems Magazine 38 (6) (2023) 4–13. doi:https://doi.org/10.1109/MAES.2023.3243584. [11] J. M. Jornet, V. Petrov, H. Wang, Z. Popović, D. Shakya, J. V. Siles, T. S. Rappaport, The evolution of applications, hardware design, and channel modeling for terahertz (thz) band communications and sensing: Ready for 6g?, Proceedings of the IEEE (2024). doi:https: //doi.org/10.1109/JPROC.2024.3412828. [12] S. Ghafoor, M. H. Rehmani, A. Davy, Next Generation Wireless Terahertz Communication Networks, CRC press, 2021. doi:https:
M. Absaruddin et al.: Preprint submitted to Elsevier
//doi.org/10.1201/9781003001140. [13] S. Ghafoor, N. Boujnah, M. H. Rehmani, A. Davy, Mac protocols for terahertz communication: A comprehensive survey, IEEE Communications Surveys & Tutorials 22 (4) (2020) 2236–2282. doi:https: //doi.org/10.1109/COMST.2020.3017393. [14] M. Absaruddin, S. Ghafoor, M. H. Rehmani, Comparative analysis of terahertz mac protocols for wireless data center, in: International Wireless Internet Conference, Springer, 2023, pp. 125–138. doi: https://doi.org/10.1007/978-3-031-58053-6_9. [15] C.-L. Cheng, A. Zajić, Characterization of propagation phenomena relevant for 300 ghz wireless data center links, IEEE Transactions on Antennas and Propagation 68 (2) (2019) 1074–1087. doi:https: //doi.org/10.1109/TAP.2019.2949135. [16] A.-A. A. Boulogeorgos, J. M. Jornet, A. Alexiou, Directional terahertz communication systems for 6g: Fact check, IEEE Vehicular Technology Magazine 16 (4) (2021) 68–77. doi:https://doi.org/10.1109/ MVT.2021.3113883. [17] V. Petrov, H. Guerboukha, A. Singh, J. M. Jornet, Wavefront hopping for physical layer security in 6g and beyond near-field thz communications, IEEE Transactions on Communications (2024). doi:https: //doi.org/10.1109/TCOMM.2024.3484937. [18] Q. Xia, Z. Hossain, M. Medley, J. M. Jornet, A link-layer synchronization and medium access control protocol for terahertz-band communication networks, IEEE Transactions on Mobile Computing 20 (1) (2019) 2–18. doi:https://doi.org/10.1109/TMC.2019.2940441. [19] I. F. Akyildiz, J. M. Jornet, Realizing ultra-massive mimo (1024× 1024) communication in the (0.06–10) terahertz band, Nano Communication Networks 8 (2016) 46–54. [20] I. F. Akyildiz, J. M. Jornet, C. Han, Terahertz band: Next frontier for wireless communications, Physical Communication 12 (2014) 16–32. doi:https://doi.org/10.1016/j.phycom.2014.01.006. [21] D. Morales, J. M. Jornet, Adapt: An adaptive directional antenna protocol for medium access control in terahertz communication networks, Ad Hoc Networks 119 (2021) 102540. doi:https://doi.org/ 10.1016/j.adhoc.2021.102540. [22] F. Siddiqui, B. Mazumdar, Link discovery extension to ns-3s terahertz communication module, in: Proceedings of the 2023 Workshop on ns3, 2023, pp. 52–59. doi:https://doi.org/10.1145/3592149.3592153. [23] F. Siddiqui, Enabling tcp communication over terahertz links: An extension to ns-3’s terasim module, in: Proceedings of the 3rd ACM Workshop on 5G and Beyond Network Measurements, Modeling, and Use Cases, 2023, pp. 28–33. doi:https://doi.org/10.1145/3609382. 3610513. [24] T. Wang, X. Shi, J. Tao, X. Wang, B. Han, Efficient synchronous mac protocols for terahertz networking in wireless data center, in: IEEE Conference on Computer Communications Workshops (INFOCOM WKSHPS), IEEE, 2022, pp. 1–6. doi:https://doi.org/10.1109/ INFOCOMWKSHPS54753.2022.9798122. [25] M. Absaruddin, S. Ghafoor, M. H. Rehmani, Beamwidth tradeoffs analysis of thz mac protocols for wireless data center, in: 2024 International Wireless Communications and Mobile Computing (IWCMC), IEEE, 2024, pp. 1345–1350. doi:https://doi.org/10. 1109/IWCMC61514.2024.10592466. [26] Etsi gr thz 001 v1.1.11 (2024-01), identification of use cases for thz communication systems, https://www.etsi.org/deliver/etsi_ gr/THz/001_099/001/01.01.01_60/gr_THz001v010101p.pdf, accessed on 15.10.2025 (2024). [27] ITU, Itu-r m.2417-1: Technical and operational characteristics of land-mobile service applications in the frequency range 275-450 ghz (2022). URL https://www.itu.int/dms_pub/itu-r/opb/rep/R-REP-M. 2417-1-2022-PDF-E.pdf
[28] J. Lin, M. A. Weitnauer, Pulse-level beam-switching for terahertz networks, Wireless Networks 25 (6) (2019) 3047–3062. doi:https: //doi.org/10.1007/s11276-018-1702-7. [29] X.-W. Yao, J. M. Jornet, Tab-mac: Assisted beamforming mac protocol for terahertz communication networks, Nano Communication Networks 9 (2016) 36–42. doi:https://doi.org/10.1016/j.nancom.
Page 16 of 17
Enhanced Dynamic Beamwidth Selection-based THz MAC Protocol for Wireless Data Center Networks 2016.07.003. [30] X.-W. Yao, F.-Z. Ni, C.-C. Wang, Multi-beam on-demand power allocation mac protocol for mimo terahertz communication networks, in: IEEE/CIC International Conference on Communications Workshops in China (ICCC Workshops), IEEE, 2019, pp. 158–163. doi:https: //doi.org/10.1109/ICCChinaW.2019.8849973. [31] X. Zhang, C. Han, X. Wang, Dual-radio-assisted (dra) mac protocols for distributed terahertz networks, IEEE Open Journal of Vehicular Technology 2 (2021) 111–124. doi:https://doi.org/10.1109/OJVT. 2021.3049969. [32] C. DAndrea, J. M. Jornet, A. Singh, P. Sen, Z. Ghassemlooy, S. Zvanovec, S. R. Teli, A. Gholami, C. DAndrea, J. M. Jornet, et al., 6g wireless technologies, in: The Road towards 6G: Opportunities, Challenges, and Applications: A Comprehensive View of the Enabling Technologies, Springer, 2024, pp. 51–114. doi:https: //doi.org/10.1007/978-3-031-42567-7_3. [33] J. Zhai, N. Ma, S. Liu, S. Zheng, A dual channel mac protocol with irs relay selection mechanism for terahertz communication, in: 2023 9th International Conference on Computer and Communications (ICCC), IEEE, 2023, pp. 321–326. doi:https://doi.org/10.1109/ICCC59590. 2023.10507253. [34] L. He, F. Hu, Z. Chu, J. Zhao, Y. Sagduyu, N. Thawdar, S. Kumar, Intelligent terahertz medium access control (mac) for highly dynamic airborne networks, IEEE Transactions on Aerospace and Electronic Systems 59 (3) (2022) 2494–2512. doi:https://doi.org/10.1109/ TAES.2022.3215464. [35] S. Cavallero, C. Buratti, A. Tsarev, G. Cuozzo, E. Khayrov, Y. Gaidamaka, R. Verdone, Applying carrier sense multiple access to industrial iot at terahertz frequencies, IEEE Internet of Things Journal (2023). doi:https://doi.org/10.1109/JIOT.2023.3334514. [36] A. AlGhadhban, F4tele: Fso for data center network management and packet telemetry, Computer Networks 186 (2021) 107711. doi:https: //doi.org/10.1016/j.comnet.2020.107711. [37] J. M. Jornet, I. F. Akyildiz, Channel modeling and capacity analysis for electromagnetic wireless nanonetworks in the terahertz band, IEEE Transactions on Wireless Communications 10 (10) (2011) 3211–3221. doi:10.1109/TWC.2011.081011.100545. [38] Z. Hossain, Q. Xia, J. M. Jornet, Terasim: An ns-3 extension to simulate terahertz-band communication networks, Nano Communication Networks 17 (2018) 36–44. doi:https://doi.org/10.1016/j.nancom. 2018.08.001. [39] IEEE, Ieee standard for wireless multimedia networks, IEEE Std 802.15.3-2023 (Revision of IEEE Std 802.15.3-2016) (2024) 1– 684doi:10.1109/IEEESTD.2024.10443750. [40] M. Absaruddin, S. Ghafoor, M. H. Rehmani, Performance evaluation of THz physical layer parameters based on ieee 802.15.3-2023, in: Irish Signals and Systems Conference (ISSC), IEEE, 2026, pp. 1–6, (Accepted for publication). [41] J. Hu, C. Zeng, Z. Wang, J. Zhang, K. Guo, H. Xu, J. Huang, K. Chen, Load balancing with multi-level signals for lossless datacenter networks, IEEE/ACM Transactions on Networking (2024). doi:https: //doi.org/10.1109/TNET.2024.3366336. [42] N. Boujnah, S. Ghafoor, et al., Final dll layer model and simulator: Terahertz based ultra high bandwidth wireless access networks, deliverable ID: D5.4, Milestone: Final proposed (October 2020). URL https://www.terapod-project.eu/ [43] K. Zhao, P. Goyal, M. Alizadeh, T. E. Anderson, Scalable tail latency estimation for data center networks, in: 20th USENIX Symposium on Networked Systems Design and Implementation (NSDI 23), USENIX Association, Boston, MA, 2023, pp. 685–702. [44] ns-3 Consortium, ns-3 lognormalvariable class reference, https://www.nsnam.org/docs/release/3.17/doxygen/classns3_1_1_ log_normal_variable.html, accessed: 2025-05-7 (2013).
M. Absaruddin et al.: Preprint submitted to Elsevier
Page 17 of 17