• KIPS Transactions on Computer and Communication Systems
• /
• v.11 no.7
• /
• pp.205-216
• /
• 2022

One of the many tools used for distributed deep learning training is Kubeflow, which runs on Kubernetes, a container orchestration tool. TensorFlow jobs can be managed using the existing operator provided by Kubeflow. However, when considering the distributed deep learning training jobs based on the parameter server architecture, the scheduling policy used by the existing operator does not consider the task affinity of the distributed training job and does not provide the ability to dynamically allocate or release resources. This can lead to long job completion time and low resource utilization rate. Therefore, in this paper we proposes a new operator that efficiently schedules distributed deep learning training jobs to minimize the job completion time and increase resource utilization rate. We implemented the new operator by modifying the existing operator and conducted experiments to evaluate its performance. The experiment results showed that our scheduling policy improved the average job completion time reduction rate of up to 84% and average CPU utilization increase rate of up to 92%.

• Journal of the Korea Society of Computer and Information
• /
• v.27 no.10
• /
• pp.95-103
• /
• 2022

As 5G mobile communication services gradually expand in P2P (peer-to-peer) or D2D (device-to-device) applications, traffic-oriented stream control such as YouTube streaming is emerging as an important technology. In D2D communication, the type of data stream most frequently transmitted by users is a video stream, which has the characteristics of a large-capacity transport stream. In a D2D communication environment, this type of stream not only provides a cause of traffic congestion, but also degrades the quality of service between D2D User Equipments (DUEs). In this paper, we propose a Traffic-Oriented Stream Scheduling (TOSS) scheme to minimize the interruption of dynamic media streams such as video streams and to optimize streaming service quality. The proposed scheme schedules the media stream by analyzing the characteristics of the media stream and the traffic type in the bandwidth of 3.5 GHz and 28 GHz under the 5G gNB environment. We examine the performance of the proposed scheme through simulation, and the simulation results show that the proposed scheme has better performance than other comparative methods.

• 제어로봇시스템학회:학술대회논문집
• /
• 2004.08a
• /
• pp.620-626
• /
• 2004

For general nonlinear processes, it is difficult to control with a linear model-based control method and nonlinear controls are considered. Among the numerous approaches suggested, the most rigorous approach is to use dynamic optimization. Many general engineering problems like control, scheduling, planning etc. are expressed by functional optimization problem and most of them can be changed into dynamic programming (DP) problems. However the DP problems are used in just few cases because as the size of the problem grows, the dynamic programming approach is suffered from the burden of calculation which is called as 'curse of dimensionality'. In order to avoid this problem, the Neuro-Dynamic Programming (NDP) approach is proposed by Bertsekas and Tsitsiklis (1996). To get the solution of seriously nonlinear process control, the interest in NDP approach is enlarged and NDP algorithm is applied to diverse areas such as retailing, finance, inventory management, communication networks, etc. and it has been extended to chemical engineering parts. In the NDP approach, we select the optimal control input policy to minimize the value of cost which is calculated by the sum of current stage cost and future stages cost starting from the next state. The cost value is related with a weight square sum of error and input movement. During the calculation of optimal input policy, if the approximate cost function by using simulation data is utilized with Bellman iteration, the burden of calculation can be relieved and the curse of dimensionality problem of DP can be overcome. It is very important issue how to construct the cost-to-go function which has a good approximate performance. The neural network is one of the eager learning methods and it works as a global approximator to cost-to-go function. In this algorithm, the training of neural network is important and difficult part, and it gives significant effect on the performance of control. To avoid the difficulty in neural network training, the lazy learning method like k-nearest neighbor method can be exploited. The training is unnecessary for this method but requires more computation time and greater data storage. The pH neutralization process has long been taken as a representative benchmark problem of nonlin ar chemical process control due to its nonlinearity and time-varying nature. In this study, the NDP algorithm was applied to pH neutralization process. At first, the pH neutralization process control to use NDP algorithm was performed through simulations with various approximators. The global and local approximators are used for NDP calculation. After that, the verification of NDP in real system was made by pH neutralization experiment. The control results by NDP algorithm was compared with those by the PI controller which is traditionally used, in both simulations and experiments. From the comparison of results, the control by NDP algorithm showed faster and better control performance than PI controller. In addition to that, the control by NDP algorithm showed the good results when it applied to the cases with disturbances and multiple set point changes.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.24 no.9B
• /
• pp.1643-1651
• /
• 1999

Video-On-Demand (VOD) is expected to provide the user with interactive operations such as VCR functions. In particular, fast scan functions like “Fast Forward” of “Fast Backward” for a certain speedup playback are required. Since they require a significant amount of system resources, schemes to reduce bandwidth requirements for the network or disk are needed. In MPEG standard, Group-of-Pictures (GoP) is a random access unit which can be decoded independently. Since storing and transmitting a video stream based on GoP is efficient, it is practical to support fast scan functions based on GoP. In this paper, we present a dynamic transmission scheduling scheme to support fast scan functions with m-times normal playback rate for a stored video. The proposed scheme writes a transmission schedule whenever user requests a fast scan function. That is, the scheme constructs the data set to be smoothed by skipping GoPs according to a given speedup factor, and then writes the transmission schedule by applying a bandwidth smoothing. Finally, the scheme restarts the transmission of video data to a client according to the new schedule. The proposed scheme results in speeding up the playback rate by utilizing “GoP skipping”, and then reduces the computational overhead by applying a bandwidth smoothing based on GoP.

• Journal of KIISE:Computer Systems and Theory
• /
• v.34 no.10
• /
• pp.529-538
• /
• 2007

Fair Queueing algorithms based on Generalized Processor Sharing (GPS) not only guarantee sessions with service rate and delay, but also provide sessions with instantaneous fair sharing. This fair sharing distributes server capacity to currently backlogged sessions in proportion to their weights without regard to the amount of service that the sessions received in the past. From a long-term perspective, the instantaneous fair sharing leads to a different quality of service in terms of delay and bandwidth to sessions with the same weight depending on their traffic pattern. To minimize such long-term unfairness, we propose a delay-bandwidth normalization model that defines the concept of value of service (VoS) from the aspect of both delay and bandwidth. A model and a packet-by-packet scheduling algorithm are proposed to realize the VoS concept. Performance comparisons between the proposed algorithm and algorithms based on fair queueing and service curve show that the proposed algorithm provides better long-term fairness among sessions and that is more adaptive to dynamic traffic characteristics without compromising its service rate and delay guarantees.

• Journal of KIISE:Information Networking
• /
• v.33 no.3
• /
• pp.247-256
• /
• 2006

Among the resource provisioning mechanisms for the hose based Virtual Private Network (VPN) Quality of Service (QoS ), VPN-specific state provisioning allows the service provider to obtain highest resource multiplexing gains. However, dynamic and automatic resource reservation for the VPN-specific state provisioning is difficult due to the lack of appropriate resource reservation protocol. Furthermore, users of a VPN may experience unfair usage of resources among themselves since the reserved resources of a VPN are shared by the VPN users in a similar way that the traditional LAN bandwidth is shared by the attached hosts. In this paper, we propose a resource reservation protocol and a traffic service mechanism, which not only enable dynamic and automatic resource reservation according to the VPN-specific state provisioning algorithm, but also enforce the fair usage of reserved resources among the users of a VPN in case of congestion.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.38B no.1
• /
• pp.46-53
• /
• 2013

Recently, intensive research has been performed for reducing video decoder energy consumption, especially based on DVFS (Dynamic Voltage and Frequency Scaling) technique. Our previous work [1] has proposed the optimal DVFS algorithm for energy reduction in video decoders. In spite of the mathematical optimality of the algorithm, the precondition of known frame decoding cycle/complexity limits its application to some realistic scenarios. This paper overcomes this limitation by frame data size-based estimation of frame decoding complexity. The proposed decoding complexity estimation method shows over 90% accuracy. And with this estimation method and buffer underflow margin of around 20% of frame size, almost same power consumption reduction performance as the optimal algorithm can be achieved.

• Journal of the Institute of Electronics Engineers of Korea CI
• /
• v.45 no.6
• /
• pp.18-26
• /
• 2008

In this paper, dynamic GTS scheduling method based on IEEE 802.15.4 is proposed for wireless control system considering reliability and real-time property. The proposed methods can guarantee a transmission of real-time periodic and sporadic data within the limited time frame in factory environment. The superframe of IEEE 802.15.4 is used for the dynamic transmission method of real-time mixed traffic (periodic data, sporadic data, and non real-time message). By separating CFP and CAP properly, the periodic, sporadic, and non real-time messages are transmitted effectively and guarantee real-time transmission within a deadline. The simulation results show the improvement of real-time performance of periodic and sporadic data at the same time.

• Journal of Korea Multimedia Society
• /
• v.7 no.7
• /
• pp.886-895
• /
• 2004

In this paper, we propose the dynamic allocation scheme of the CPU bandwidth to efficiently integrate and schedule these tasks in the same system, where multimedia tasks and hard real-time tasks can coexist simultaneously. Hard real-time tasks are guaranteed based on worst case execution times, whereas multimedia tasks modeled as soft real-time tasks are served based on mean parameters. This paper describes a server-based allocation scheme for assigning the CPU resource to two types of tasks. Especially for MPEG video streams, we show how to dynamically control the fraction of the CPU bandwidth allocated to each multimedia task. The primary purpose of the proposed method is to minimize the mean tardiness of multimedia tasks while satisfying the timing constraints of hard real-time tasks present in the system. We showed through simulations that the tardiness experienced by multimedia tasks under the proposed allocation scheme is much smaller than that experienced by using other scheme.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.29 no.7A
• /
• pp.844-851
• /
• 2004

Various types of error are caused due to many factors of various environment in air interface channel of wireless communications. In this case, the reliability of the channel is much lower than that of wired case. IEEE 802.15.3 high-rate WPAN, which operates in an ad hoc networking environment, is more susceptible to such errors. The problem has been investigated for wireless LANs, for example, as follows. If the queue size of a certain node is longer than that of other nodes, the node estimates that its channel state is bad and the resource of the node is decreased. However this method has a disadvantage that a central controller must always monitor the status. To avoid this disadvantage, in this paper, a new MAC protocol that the throughput of overall piconet is increased by LDS (Link-status Dependent Scheduling) is proposed.