• Formulate the inconsistent JCT problem using gSLA for the first time. • Design a new JCT increase prediction model and job scheduler for GPU sharing. • Achieve up to 47.3× better gSLA satisfaction and 50× lower gSLA excess ratio. • Improve JCT and GPU efficiency by ∼ 60% and ∼ 44% over existing methods. • Demonstrate TensorShare’s effectiveness in improving gSLA and JCT for unseen jobs. GPU sharing aims to enhance the efficiency of GPU utilization by running distributed deep learning training jobs concurrently. However, GPU sharing poses a significant challenge: the increase in job completion time (JCT) caused by interference between jobs is inconsistent, complicating job scheduling. Our experiments reveal that the degree of JCT increase varies by as much as ∼ 3.7 × . While previous studies have analyzed this JCT inconsistency problem, none of them have been able to minimize the inconsistency. We propose TensorShare, a proactive GPU sharing technique that leverages a deep learning model to predict the extent of JCT increase. This study defines a new metric, called GPU SLA, which represents the upper threshold of JCT increase. TensorShare then introduces a novel scheduler that proactively identifies which jobs meet GPU SLA while minimizing the JCT increase. Our evaluation shows that TensorShare improves GPU SLA satisfaction rates by 26.1 × –47.3 × and reduces the JCT increase by 37%–60%. Furthermore, we evaluate TensorShare with large language models that are not included in training TensorShare’s prediction model, achieving ∼ 7 × and ∼ 10.3 × improvements in GPU SLA satisfaction and JCT inconsistency, respectively.

As more distributed deep learning (DDL) jobs run in public clouds, their effective scheduling becomes a major challenge. Current studies prioritize the execution of jobs with less remaining time, which is known to be the best in reducing average job completion time (JCT). However, we observe that this approach does not work when the preemption for pausing and loading jobs weighs in; sometimes, the preemption overheads of DDL jobs take up to hundreds of seconds. This results in very ineffective scheduling, so in some cases, the first-in-first-out policy performs much better. This paper proposes a new scheduling framework called Xion that takes into account the preemption overheads and only preempts DDL jobs when it is beneficial. Our evaluation results demonstrate that Xion effectively reduces the average JCT by 19% and improves the waiting time by 1.64×.

Performance isolation is an essential property that network virtualization must provide for clouds. This study addresses the performance isolation of the control plane in virtualized software-defined networking (SDN), which we call control channel isolation. First, we report that the control channel isolation is seriously broken in the existing network hypervisor in that the end-to-end control latency grows by up to 15 x as the number of virtual switches increases. This jeopardizes the key network operations, such as routing, in datacenters. To address this issue, we take a machine learning approach that learns from the past control traffic as time-series data. We propose a new network hypervisor, Meteor, that designs an LSTM autoencoder to predict the control traffic per virtual switch. Our evaluation results show that Meteor improves the processing latency per control message by up to 12.7x. Furthermore, Meteor reduces the end-to-end control latency by up to 73.7%, which makes it comparable to the non-virtualized SDN.

We introduce TeaVisor that provides bandwidth isolation guarantee for network virtualization (NV) based on software-defined networking (SDN). SDN-based NV (SDN-NV) offers many benefits to clouds, such as topology and address virtualization while allowing flexible resource provisioning, control, and monitoring on virtual networks. In SDN-NV, however, routing is done by tenants independently; thus, existing studies have difficulties in bandwidth isolation guarantee due to the overloaded link problem. Bandwidth isolation guarantee is essential for providing stable and reliable throughput on network services in SDN-NV. Without bandwidth isolation guarantee, tenants suffer degraded service qualities and significant loss in revenue. To address this problem, we design and implement TeaVisor in three components: path virtualization, bandwidth reservation, and path establishment. Through extensive experiments, TeaVisor shows that bandwidth isolation is guaranteed with near-zero errors, which is three orders of magnitude better than existing studies. In addition, TeaVisor guarantees the minimum and maximum bandwidth at the same time. We also present an overhead analysis of TeaVisor in control traffic and memory consumption.

This article presentsElixir, an automated prediction model formulation framework for control traffic using machine learning. Control traffic is vital in software-defined networking (SDN) systems because it determines the reliability and scalability of the entire system. Various studies have sought to design control traffic prediction models for the proper provisioning and planning of SDN systems. However, previously proposed models are based on descriptive modeling, well-suited for only specific SDN system instances. Furthermore, these models exhibit poor accuracy (errors of up to 85%) because of the heterogeneity of SDN systems. Because descriptive modeling requires a significant amount of human contemplation, it is impossible to formulate adequate prediction models for countless SDN system instances.Elixiraddresses this problem by applying machine learning.Elixirstarts the model formulation through self-generated datasets. Then,Elixirsearches prediction models to fit the accuracy for respective SDN systems. Also,Elixirpicks robust models that exhibit reasonable accuracy even in a network topology that differs from the topology used for model training. We evaluate theElixirframework on nine heterogeneous SDN systems. As a key outcome,Elixirsignificantly reduces prediction errors, achieving up to 10.6× improvement compared to the previous model for control traffic throughput of OpenDayLight controller.

Deep learning models have a wide spectrum of GPU execution time and memory size. When running distributed training jobs, however, their GPU execution time and memory size have not been taken into account, which leads to the high variance of job completion time (JCT). Moreover, the jobs often run into the GPU out-of-memory (OoM) problem so that the unlucky job has to restart all over. To address the problems, we propose Xonar to profile the deep learning jobs and order them in the queue. The experiments show that Xonar with TensorFlow v1.6 reduces the tail JCT by 44% with the OoM problem eliminated.

Network virtualization (NV) becomes an essential technology in cloud computing that isolates network flows for tenants. However, because existing NV technologies like overlay do not enable tenants to directly program (i.e., provision, control, and monitor) network resources, software-defined networking (SDN)-based NV (SDN-NV) has been proposed. Despite its great benefits, SDN-NV has been believed to bring considerable overheads due to the network hypervisor (NH). However, to date, there is no definite performance evaluation that proves the overheads of SDN-NV. To this end, this paper comprehensively investigates the performance and overheads of SDN-NV. Our experiment results reveal that SDN-NV provides the data plane performance comparable to or even better (up to 10.5× better TCP throughput) than the existing NV technologies. Also, the results on NH show that its overheads remain mostly constant, even when the number of switches, virtual networks, or network flows increases. In short, our evaluation indicates that the overhead of SDN-NV should not deter its practical use in datacenters.

In the presence of nonlinearity, we analyze the impact of the selected mapping (SLM) technique on bit-error-rate (BER) performance of orthogonal frequency division multiplexing (OFDM) systems in an additive white Gaussian noise channel. The peak-to-average-ratio (PAR) reduction gain of SLM can be increased by improving the PAR statistics at the cost of complexity in the OFDM transmitter, thereby helping to decrease the required power amplifier (PA) output backoff (OBO). However, since the PAR statistics focus only on the statistical distribution of the highest peak in an OFDM symbol, the statistics cannot be used to quantify BER performance degradation in the presence of nonlinearity such as that caused by a PA or digital-to-analog converter (DAC). We first derive a closed-form expression for the envelope power distribution in an OFDM system with SLM. Then, using this derived envelope power distribution, we investigate the BER performance and the total degradation (TD) of OFDM systems with SLM under the existence of nonlinearity. We discuss peak backoff (PBO) and the clipping ratio, which determine the operating point of the PA and dynamic range of the DAC, respectively. Lastly, we consider the total degradation (TD), which indicates the tradeoff between the OBO and the E_b/N_o penalty due to nonlinearity, and numerically compute the PBO and clipping ratio that minimize the TD. The TD-minimizing PBO and clipping ratio are given as functions of the number of candidate signals in SLM.

The mitigation of intercell interference is an importance issue for current and next-generation wireless cellular networks where frequencies are aggressively reused and hierarchical cellular structures may heavily overlap. The paper examines the benefit of coordinating transmission strategies and resource allocation schemes across multiple base-stations for interference mitigation. Two different wireless cellular architectures are studied: a multicell network where base-stations coordinate in their transmission strategies, and a mixed macrocell and femtocell/picocell deployment with coordination among macro and femto/pico base-stations. For both scenarios, this paper proposes a heuristic joint proportionally fair scheduling, spatial multiplexing, and power spectrum adaptation algorithm that coordinates multiple base-stations with an objective of optimizing the overall network utility. The proposed scheme optimizes the user schedule, transmit and receive beamforming vectors, and transmit power spectra jointly, while taking into consideration both the intercell and intracell interference and the fairness among the users. System-level simulation results show that coordination at the transmission strategy and resource allocation level can already significantly improve the overall network throughput as compared to a conventional network design with fixed transmit power and per-cell zero-forcing beamforming.

Heterogeneous cellular networks which consist of macrocells and small cells can offer significant capacity gain by utilizing the resources of the small cells. However, to achieve this, the interference between the macrocells and the small cells must be carefully managed. In this work, we propose an uplink intercell interference control (ICIC) scheme which is a unified ICIC approach of handover based interference control and rate-split based interference control. The handover based interference control scheme is a win-win strategy which enhances both interfering user's rate and interfered user's rate. On the other hand, the rate-split based interference control scheme is a yield-win strategy where an interfering user sacrifices his rate to save the interfered user's rate. In this paper, we assume that users have their target QoS such as minimum rate when they send their data. The proposed uplink ICIC scheme reduces the interference as best as possible while it guarantees the minimum QoS. Simulation results shows that the proposed ICIC scheme offers enhanced rate or fairness than legacy ICIC schemes which are not considering user QoS. The proposed ICIC scheme works in distributed manner with low-complexity so that it can be applied to self-organizing network and mobile ad-hoc networks as well as heterogeneous cellular networks.

In heterogeneous cellular networks (HCN) which consists of macrocells and numerous picocells, efficient interference management schemes between macrocells and picocells are so crucial to the overall system performance. We propose a dynamic cooperative silencing control (DCS) scheme for intercell interference control (ICIC). It is a low-complex, low-feedback and distributed algorithm using only strongly interfered neighboring users' information. The system simulation shows that the system performance and in particular the cell-edge throughput is significantly increased with the proposed silencing scheme. It offers 420% and 190% higher average spectral efficiency and edge-user spectral efficiency in compared to macrocell only case, respectively.

The mitigation of intercell interference is a central issue for future-generation wireless cellular networks where frequencies are reused aggressively and where hierarchical cellular structures may heavily overlap. The paper examines the benefit of coordinating transmission strategies and resource allocation schemes across multiple cells for interference mitigation. For a multicell network serving multiple users per cell sectors and where both the base-stations and the remote users are equipped with multiple antennas, this paper proposes a joint proportionally fair scheduling, spatial multiplexing, and power spectrum adaptation method that coordinates multiple base-stations with an objective of optimizing the overall network utility. The proposed scheme optimizes the user schedule, transmit and receive beamforming vectors, and transmit power spectra jointly, while taking into consideration both the intercell and intracell interference and the fairness among the users. The proposed system is shown to significantly improve the overall network throughput while maintaining fairness as compared to a conventional network with per-cell zero-forcing beamforming and with fixed transmit power spectrum. The proposed system goes toward the vision of a fully coordinated multicell network, whereby transmission strategies and resource allocation schemes (rather than transmit signals) are coordinated across the base-stations as a first step.

The interference alignment (IA) is a promising technique to effectively mitigate interferences in wireless communication systems. To show the potential benefits of such an IA scheme, this letter focuses on a two-cell multiple-input multiple-output (MIMO) Gaussian interfering broadcast channels (MIMO-IFBC) with M transmit antennas and N receive antennas. It corresponds to a downlink scenario for cellular networks with two base stations (BSs) with M transmit antennas per BS, and two users with N receive antennas per user, on the cell-boundary of each BS. In this scenario, we propose a novel IA technique jointly designing transmit and receive beamforming vectors in a closed-form expression without iterative computation. It is also analytically shown that the proposed IA algorithm achieves the optimal degrees of freedom (DoF) of 2N in the case of [¾N] ≤ M <; 2N. The simulations demonstrate that not only the analytical results are valid, but the sum-rate of our proposed scheme also outperforms those of conventional techniques, especially in the high signal-to-noise ratio (SNR) regime.

In heterogeneous cellular networks (HTN) which consists of macro-cells and numerous femto-cells, efficient interference management schemes between macro-cells and femto-cells are so crucial to the overall system performance. To mitigate inter-cell interference in the HTN, we propose a new rate-split transmission scheme which has following characteristics. First, it guarantees serving user''s QoS by deciding common message power for an interfered user. Second, it is a low complex scheme using only ISNR (Interference to Signal and Noise Ratio) feedback between a macro-base station and a femto-base station. Third, it operates in a distributed manner. The performance evaluation shows that the proposed algorithm significantly reduces the interference for severely interfered users while guaranteeing serving user''s QoS.

Relay-assisted space-time block code (STBC) and space-frequency block code (SFBC) for single carrier frequency-domain equalization (SC-FDE) were presented. They achieve spatial diversity over fading relay channels under the assumption of perfect channel state information (CSI). In this paper, we propose a pilot position selection/detection technique for channel estimation of those systems. Unlike the conventional block-type channel estimation techniques, the proposed scheme superimposes pilots on data-carrying tones whose positions are selected to minimize the distortion of original signals. Without additional pilot overhead, the proposed technique can track the CSI even when the mobile equipment speed is high. The corresponding destination structure and frequency domain equalization are also presented, where the pilot positions are blindly detected and the distorted data symbols are iteratively reconstructed. Simulation results show that the proposed method gives better BER performance than the block-type channel estimation for the distributed SFBC (D-SFBC) SC-FDE over fast fading relay channels, without the loss of spectral efficiency.

We find that placing each pilot tone in an equally spaced manner according to the conventional placement scheme is not suitable for doubly selective channel estimation. In this paper, we propose an efficient pilot tone placement scheme enabling accurate channel estimation in OFDM systems regardless of time variations of a channel. Since the number of channel impulse response taps to be estimated is typically much greater than the number of pilot tones, linear minimum mean square error (LMMSE) estimation schemes for time-invariant channels cannot be straightforwardly extended to doubly selective channel estimation. To overcome this problem, we propose an accurate LMMSE channel estimator that exploits a small number of pilot tones located according to the derived pilot placement. To achieve performance close to the LMMSE estimator but with lower complexity, an approximate LMMSE (ALMMSE) channel estimator is also proposed. Finally, we propose a novel iterative ALMMSE channel estimator that achieves better performance than the LMMSE and ALMMSE estimators, while having complexity in between the two.

Multiple-input multiple-output (MIMO) orthogonal frequency division multiplexing (OFDM) systems are attractive because of their large capacity and high data rate. The paper presents a blind channel estimation method for MIMO-OFDM systems with and without a cyclic prefix (CP) by using the presence of virtual carriers. The proposed method is based on the noise subspace method. In addition, we establish a sufficient condition for channels to be identifiable. The proposed algorithm can be applied to MIMO-OFDM systems with insufficient CP or no CP, thereby potentially increasing channel utilization. Furthermore, by using reduced CP or no CP, the proposed method can achieve lower computational complexity. We present simulation results demonstrating the performance of the proposed method via numerical experiments.