Abstract To share limited, large-capacity resources, the high-performance computing field provides services by allocating available resources to jobs through batch job schedulers. Therefore, it is natural that a queue waiting time occurs until the resources are available if resources are not sufficient. The prediction of queue waiting time is very useful to improve overall resource utilization. However, the queue waiting time is very difficult to predict because it is significantly affected by the many factors such as applied scheduling algorithm and characteristics of the executed job. In this study, a method of predicting queue waiting time using only the historical log data created by the batch job scheduler is examined. Specifically, a method of predicting queue waiting time based on a hidden Markov model is proposed. It has the following three stages. First, outliers are removed by applying the outlier detection algorithm using a statistics-based parametric method. Second, the parameters of the hidden state are estimated using the observed queue waiting time sequence based on the historical job log. Third, the queue waiting interval at time $$t+1$$ t+1 is provided using the estimated parameters at timet. Comparing the prediction accuracy with those of the other prediction methods, experimental results show that the proposed algorithm improves the prediction accuracy by up to 60%.

As demand for high-performance computing power is increasing, operation management technologies like check-pointing, failure-aware task scheduling, and system simulations are becoming more important for the stable operation of the system. To maintain and manage a stable system, a detailed analysis of failed tasks is necessary. For this, this paper intends to analyze the characteristics of failed jobs in high performance computing system. Our contributions can be viewed in three ways. Firstly, it offers detailed analysis results of failed jobs based on the job logs of a currently operating supercomputer. Secondly, it offers not only an overall statistical analysis result but also identifies the distribution of the failed job submission inter-arrival time. Thirdly, it analyzes the occurrence probability of the event using hazard rate.

This nowadays, along with popularity of application and use of network technologies, the structure of network is being complex and enormous to support emerging applications. A NMS (Network Management System) is required to support emerging services through the complex network structure. In this paper, we present an on-going work called SOLA (Smart One-view Look and Analysis) to integrate the existing NMSs. SOLA has three remarkable features. First, it can reduce the CAPEX/OPEX by sharing a common platform. Second, since SOLA is a web application complied with HTML 5, it can be run on emerging smart devices such as smart phone and table PC as well. Lastly, it can provide an end-to-end analysis since the event messages are collected from an entire network.

It is very difficult to identify the location of bottlenecks when there is deterioration of Quality of Service (QoS) because current Internet service is the complicate structure which is provided through a number of ISP. This paper proposes bottleneck detection algorithm which can detect the bottleneck point easily and efficiently even though there is no prior knowledge of the network connection status. The execution procedure of the proposed algorithm can be largely divided into two phases. In the first phase, the proposed algorithm finds the choke point which is suspected to be the bottleneck point using Recursive Packet Train (RPT). In second phase, it identifies the bottleneck point by monitoring packet process time of the detected choke point. For this, this paper proposes two methods. The first method monitors packet process time of the choke point utilizing timestamp option. This method has the advantage which can monitor the bottleneck point without increasing network load because it uses passive monitoring approach using the IP header part of the data packet of the actual users. But agent should be run in all of servers and clients. The second method monitors packet process time of the choke point utilizing ICMP packet. This method is possible to detect the bottleneck point by running agent solely in client without running agent of servers because ICMP packet which is sent to the senders to report the problems found while processing IP packet in routers.

Parallel download protocols that establish multiple TCP connections to distributed replica servers have the potential to reduce file download time and to achieve a server-side load balancing. Unfortunately, parallel download protocols are also inherently unfair towards single-flow downloads and may even starve them. This paper presents TCP-ROME, a parallel download protocol that allows a dynamic mitigation of throughput and fairness. The key novelty is a receiver-based framework that allows a dynamic adjustment of the congestion and rate control of each subconnection. TCP-ROME offers two usage modes: a binary mode where the congestion control of each subconnection can be switched between a TCP-fair rate (high priority) and at a TCP-LP fair rate (low priority), and a more complex range mode where the aggregated throughput aims at meeting a specified target rate. Apart from the protocol specification, this paper develops novel analytical throughput models for TCP-LP and for TCP-ROME. The models are validated via simulations. Extensive simulation scenarios show the flexibility of TCP-ROME in mitigating performance for fairness.

Live streaming of multimedia content over the Internet is rapidly increasing, e.g. with the Web 2.0 and via peer-to-peer systems. While many homes today are equipped with sufficient bandwidth (DSL) to support high quality live streaming, the quality of experience of a user is dismal: the video resolution and size is small and the displaying is often interrupted. One of the key problems arises because TCP is used as the underlying transport layer protocol: packet loss, retransmissions and timeouts prevent TCP from delivering a timely, steady flow of packets and thus inhibits the provision of online multimedia content. This paper presents TCP-ROME, a novel transport-layer framework that allows an establishment of multiple TCP connections between one client with possibly multiple home addresses and multiple co-located or distributed servers. The download over multiple TCP connections increases the total throughput by aggregating the resources of multiple TCP connections. Moreover, TCP-ROME provides the means to overcome bandwidth fluctuations and network bottlenecks by dynamically coordinating the downloaded content from the different servers and by coordinating and adapting the streaming rate of the different connections to meet the bandwidth requirements of the video.