Server Processing Capacity Research Articles

Pedestrian Abnormal Behavior Detection System Using Edge–Server Architecture for Large–Scale CCTVEnvironments

As the deployment of CCTV cameras for safety continues to increase, the monitoring workload has significantly exceeded the capacity of the current workforce. To overcome this problem, intelligent CCTV technologies and server-efficient deep learning analysis models are being developed. However, real-world applications exhibit performance degradation due to environmental changes and limited server processing capacity for multiple CCTVs. This study proposes a real-time pedestrian anomaly detection system with an edge–server structure that ensures efficiency and scalability. In the proposed system, the pedestrian abnormal behavior detection model analyzed by the edge uses a rule-based mechanism that can detect anomalies frequently, albeit less accurately, with high recall. The server uses a deep learning-based model with high precision because it analyzes only the sections detected by the edge. The proposed system was applied to an experimental environment using 20 video streams, 18 edge devices, and 3 servers equipped with 2 GPUs as a substitute for real CCTV. Pedestrian abnormal behavior was included in each video stream to conduct experiments in real-time processing and compare the abnormal behavior detection performance between the case with the edge and server alone and that with the edge and server in combination. Through these experiments, we verified that 20 video streams can be processed with 18 edges and 3 GPU servers, which confirms the scalability of the proposed system according to the number of events per hour and the event duration. We also demonstrate that the pedestrian anomaly detection model with the edge and server is more efficient and scalable than the models with these components alone. The linkage of the edge and server can reduce the false detection rate and provide a more accurate analysis. This research contributes to the development of control systems in urban safety and public security by proposing an efficient and scalable analysis system for large-scale CCTV environments.

Joint Replica Server Placement, Content Caching, and Request Load Assignment in Content Delivery Networks

With the explosive growth of information and communication technology and its services, some popular Websites currently generate an enormous amount of Internet traffic. A content delivery network (CDN) would then become imperative for supporting such services efficiently. In this paper, we propose joint optimizing approaches for replica server placement, content caching in selected servers, and content request load assignment among the servers, aiming to minimize the ratio of unserved content request load when the network resources and server capacity are both limited. For this, we develop a mixed integer linear programming (MILP) optimization model. To mitigate the computational complexity of the MILP model, we further decompose the optimization problem into three sub-problems, including: 1) choosing the replica server nodes optimally; 2) deciding the content items to be cached in the replica servers; and 3) allocating the content request loads from users onto different servers. For these sub-problems, we develop corresponding heuristic algorithms and show that the proposed approach is not only efficient but also performs very close to the MILP model. We also find that a number of system limitations, such as different numbers of replica servers placed, link capacity, server processing capacity, and server storage capacity, jointly affect the performance of the CDN. Saturation trends are observed on the performance, which indicate that as long as sufficient resources have already been provided, augmenting the resources further may only lead to marginal additional performance improvement.

Video Management and Resource Allocation for a Large-Scale VoD Cloud

We consider providing large-scale Netflix-like video-on-demand (VoD) service on a cloud platform, where cloud proxy servers are placed close to user pools. Videos may have heterogeneous popularity at different geo-locations. A repository provides video backup for the network, and the proxy servers collaboratively store and stream videos. To deploy the VoD cloud, the content provider rents resources consisting of link capacities among servers, server storage, and server processing capacity to handle remote requests. We study how to minimize the deployment cost by jointly optimizing video management (in terms of video placement and retrieval at servers) and resource allocation (in terms of link, storage, and processing capacities), subject to a certain user delay requirement on video access. We first formulate the joint optimization problem and show that it is NP-hard. To address it, we propose Resource allocation And Video management Optimization (RAVO), a novel and efficient algorithm based on linear programming with proven optimality gap. For a large video pool, we propose a video clustering algorithm to substantially reduce the run-time computational complexity without compromising performance. Using extensive simulation and trace-driven real data, we show that RAVO achieves close-to-optimal performance, outperforming other advanced schemes significantly (often by multiple times).

Server Resource Dimensioning and Routing of Service Function Chain in NFV Network Architectures

The Network Function Virtualization (NFV) technology aims at virtualizing the network service with the execution of the single service components in Virtual Machines activated on Commercial-off-the-shelf (COTS) servers. Any service is represented by the Service Function Chain (SFC) that is a set of VNFs to be executed according to a given order. The running of VNFs needs the instantiation of VNF instances (VNFI) that in general are software components executed on Virtual Machines. In this paper we cope with the routing and resource dimensioning problem in NFV architectures. We formulate the optimization problem and due to its NP-hard complexity, heuristics are proposed for both cases of offline and online traffic demand. We show how the heuristics works correctly by guaranteeing a uniform occupancy of the server processing capacity and the network link bandwidth. A consolidation algorithm for the power consumption minimization is also proposed. The application of the consolidation algorithm allows for a high power consumption saving that however is to be paid with an increase in SFC blocking probability.

Local SIP Overload Control: Controller Design and Optimization by Extremum Seeking

The session initiation protocol (SIP) is a signaling protocol for managing various types of real-time sessions between parties over an Internet Protocol (IP) network. An open issue is the control of overload situations that occur when the incoming flow of requests to a SIP server overcomes the server processing capacity. In particular, call establishment times increase due to overload, which triggers retransmissions and causes a further increase of the total incoming flow of requests. This paper proposes an overload control system for regulating both the queue length and the CPU load of the SIP server. The proposed control system is made of two proportional-integral controllers tuned by minimizing a proper cost function using the Extremum Seeking algorithm. A real working implementation of the SIP overload controller has been made in the open-source SIP server Kamailio. A performance evaluation and comparison of the proposed controller with the main proposals existing in the literature have been carried out. The results show that the proposed control system counteracts overload situations and provides a goodput close to the optimal while maintaining low call establishment delays and retransmission ratios.

QCWS: an implementation of QoS-capable multimedia web services

QoS, that defines service quality such as latency, availability, timeliness and reliability, is important for Web applications that provide real-time information, multimedia content, or time-critical services. Many Web applications require a guaranteed server processing capacity. We study the QoS control issues using the current Web services standards. We propose a QoS Web service architecture, QCWS, by deploying a QoS broker between Web service clients and Web service providers. In the QCWS architecture, a QoS broker collects the QoS information about service providers that may offer qualified Web services to a client, makes selection decisions based on some cost criteria, and negotiates with servers to secure QoS service commitment from one of them. We have implemented a QCWS prototype using IBM WSDK, enhanced with simple QoS capabilities. We have also measured the performance running under different service priorities.

A least flow-time first load sharing approach for distributed server farm

The most critical property exhibited by a heavy-tailed workload distribution (found in many WWW workloads) is that a very small fraction of tasks make up a large fraction of the workload, making the load very difficult to distribute in a distributed system. Load balancing and load sharing are the two predominant load distribution strategies used in such systems. Load sharing generally has better response time than load balancing because the latter can exhibit excessive overheads in selecting servers and partitioning tasks. We therefore further explored the least-loaded-first (LLF) load sharing approach and found two important limitations: (a) LLF does not consider the order of processing, and (b) when it assigns a task, LLF does not consider the processing capacity of servers. The high task size variation that exists in heavy-tailed workloads often causes smaller tasks to be severely delayed by large tasks. This paper proposes a size-based approach, called the least flow-time first (LFF-SIZE), which reduces the delay caused by size variation while maintaining a balanced load in the system. LFF-SIZE takes the relative processing time of a task into account and dynamically assigns a task to the fittest server with a lighter load and higher processing capacity. LFF-SIZE also uses a multi-section queue to separate larger tasks from smaller ones. This arrangement effectively reduces the delay of smaller tasks by larger ones as small tasks are given a higher priority to be processed. The performance results performed on the LFF-SIZE implementation shows a substantial improvement over existing load sharing and static size-based approaches under realistic heavy-tailed workloads.

Continuous-Review Tracking Policies for Dynamic Control of Stochastic Networks

This paper is concerned with dynamic control of stochastic processing networks. Specifically, it follows the so called “heavy traffic approach,” where a Brownian approximating model is formulated, an associated Brownian optimal control problem is solved, the solution of which is then used to define an implementable policy for the original system. A major challenge is the step of policy translation from the Brownian to the discrete network. This paper addresses this problem by defining a general and easily implementable family of continuous-review tracking policies. Each such policy has the following structure: at each point in time t, the controller observes the current vector of queue lengths q and chooses (i) a target position z(q) of where the system should be at some point in the near future, say at time t+l, and (ii) an allocation vector v(q) that describes how to split the server's processing capacity amongst job classes in order to steer the state from q to z(q). Implementation of such policies involves the enforcement of small safety stocks. In the context of the “heavy traffic” approach, the solution of the approximating Brownian control problem is used in selecting the target state z(q). The proposed tracking policy is shown to be asymptotically optimal in the heavy traffic limiting regime, where the Brownian model approximation becomes valid, for multiclass queueing networks that admit orthant Brownian optimal controls; this is a form of pathwise, or greedy, optimality. Several extensions are discussed.

Online smoothing of variable-bit-rate streaming video

Bandwidth smoothing techniques for stored video perform end to end workahead transmission of frames into the client playback buffer, in advance of their display times. Such techniques are very effective in reducing the burstiness of the bandwidth requirements for transmitting compressed, stored video. This paper addresses online bandwidth smoothing for a growing number of streaming video applications such as newscasts, sportscasts, and distance learning, where many clients may be willing to tolerate a playback delay of a few seconds in exchange for a smaller bandwidth requirement. The smoothing can be performed at either the source of the videocast or at special smoothing server(s) (e.g., proxies or gateways) within the network. In contrast to previous work on stored video, the online smoothing server has limited knowledge of frame sizes and access to only a segment of the video at a time. This is either because the feed is live or because it is streaming past the server. We formulate an online smoothing model which incorporates playback delay, client and server buffer sizes, server processing capacity, and frame size prediction techniques. Our model can accommodate an arbitrary arrival process. Using techniques for smoothing stored video at the source as a starting point, we develop an online, window-based smoothing algorithm for delay tolerant applications. Extensive experiments with MPEG-1 and M-JPEG video traces demonstrate that online smoothing significantly reduces the peak rate, coefficient of variation, and effective bandwidth of variable-bit-rate video streams. These reductions can be achieved with modest playback delays of a few seconds to a few tens of seconds and moderate client buffer sizes, and closely approximate the performance of optimal offline smoothing of stored video. In addition, we show that frame size prediction can offer further reduction in resource requirements, though prediction becomes relatively less important for longer playback delays. However, the ability to predict future frame sizes affects the appropriate division of buffer space between the server and client sites. Our experiments show that the optimal buffer allocation shifts to placing more memory at the server as the server has progressively less information about future frame sizes.