An approach for web server cluster(WSC)reliability and degradation process analysis is proposed.The reliability process is modeled as a non-homogeneous Markov process(NHMH)composed of several non-homogeneous Poisson p...An approach for web server cluster(WSC)reliability and degradation process analysis is proposed.The reliability process is modeled as a non-homogeneous Markov process(NHMH)composed of several non-homogeneous Poisson processes(NHPPs).The arrival rate of each NHPP corresponds to the system software failure rate which is expressed using Cox s proportional hazards model(PHM)in terms of the cumulative and instantaneous load of the software.The cumulative load refers to software cumulative execution time,and the instantaneous load denotes the rate that the users requests arrive at a server.The result of reliability analysis is a time-varying reliability and degradation process over the WSC lifetime.Finally,the evaluation experiment shows the effectiveness of the proposed approach.展开更多
There are two kinds of dispatching policies in content-aware web server cluster; segregation dispatching policy and mixture dispatching policy. Traditional scheduling algorithms all adopt mixture dispatching policy. T...There are two kinds of dispatching policies in content-aware web server cluster; segregation dispatching policy and mixture dispatching policy. Traditional scheduling algorithms all adopt mixture dispatching policy. They do not consider that dynamic requests' serving has the tendency to slow down static requests' serving, and that different requests have different resource demands, so they can not use duster's resource reasonably and effectively. This paper uses stochastic reward net (SRN) to model and analyze the two dispatching policies, and uses stochastic Petri net package (SPNP) to simulate the models. The simulation results and practical tests both show that segregation dispatching policy is better than mixture dispatching policy. The principle of segregation dispatching policy can guide us to design efficient scheduling algorithm.展开更多
Distributed architectures support increased load on popular web sites by dispatching client requests transparently among multiple servers in a cluster. Packet Single-Rewriting technology and client address hashing alg...Distributed architectures support increased load on popular web sites by dispatching client requests transparently among multiple servers in a cluster. Packet Single-Rewriting technology and client address hashing algorithm in ONE-IP technology which can ensure application-session-keep have been analyzed, an improved request dispatching algorithm which is simple, effective and supports dynamic load balance has been proposed. In this algorithm, dispatcher evaluates which server node will process request by applying a hash function to the client IP address and comparing the result with its assigned identifier subset; it adjusts the size of the subset according to the performance and current load of each server, so as to utilize all servers' resource effectively. Simulation shows that the improved algorithm has better performance than the original one.展开更多
Based on the demand of the admission control of softswitch-based clustered media server, this pa- per proposed a new dynamic quota-based admission control algorithm that has a sub-negotiation process. The strongpoint ...Based on the demand of the admission control of softswitch-based clustered media server, this pa- per proposed a new dynamic quota-based admission control algorithm that has a sub-negotiation process. The strongpoint of quota-based algorithm had been inherited in the algorithm and at the same time some new ideas had also been introduced into it. Simulations of the algorithm had been conducted on the Petri net model and the results show that this algorithm has excellent performance. In order to find the optimal resource quota set- ting in real time, the paper proposed two approximation analysis methods. It can be seen from analysis results that these two methods can be used to get sub-optimal quota values quickly and effectively. These two ap- proximation analysis methods will play important roles in implementation of the algorithm in system.展开更多
In this paper, we explored a load-balancing algorithm in a cluster file system contains two levels of metadata-server, primary-level server quickly distributestasks to second-level servers depending on the closest loa...In this paper, we explored a load-balancing algorithm in a cluster file system contains two levels of metadata-server, primary-level server quickly distributestasks to second-level servers depending on the closest load-balancing information. At the same time, we explored a method which accurately reflect I/O traffic and storage of storage-node: computing the heat-value of file, according to which we realized a more logical storage allocation. According to the experiment result, we conclude that this new algorithm shortens the executing time of tasks and improves the system performance compared with other load algorithm.展开更多
The problem of state space explosion is still ran outstanding challenge in Markovian performance analysis for multiserver multiqueue (MSMQ) systems. The system behavior of a MSMQ system is described using stochastic h...The problem of state space explosion is still ran outstanding challenge in Markovian performance analysis for multiserver multiqueue (MSMQ) systems. The system behavior of a MSMQ system is described using stochastic high-level Petri net (SHLPN) models, and an approximate performance analysis technique is proposed based on decomposition and refinement methods as well as iteration technique. A real MSMQ system, Web-server cluster, is investigated. The performance of an integrated scheme of request dispatching and scheduling is analyzed with both Poisson and self-similar request arrivals. The study shows that the approximate analysis technique significantly reduces the complexity of the model solution and is also efficient for accuracy of numerical results.展开更多
Distribution of metadata in a metadata server cluster is important in mass storage system. A good distribution algorithm has a significant influence on the system performance, availability and scalability. Subtree par...Distribution of metadata in a metadata server cluster is important in mass storage system. A good distribution algorithm has a significant influence on the system performance, availability and scalability. Subtree partition and hash are two traditional metadata distribution algorithms used in distributed file systems. They both have a defect in system scalability. This paper proposes a new directory hash (DH) algorithm. By treating directory as the key value of hash function, implementing concentrated storage of metadata, pipelining operations and prefetching technology, DH algorithm can enhance the system scalability on the premise without sacrificing system performance.展开更多
Requests distribution is an key technology for Web cluster server. This paper presents a throughput-driven scheduling algorithm (TDSA). The algorithm adopts the throughput of cluster back-ends to evaluate their load...Requests distribution is an key technology for Web cluster server. This paper presents a throughput-driven scheduling algorithm (TDSA). The algorithm adopts the throughput of cluster back-ends to evaluate their load and employs the neural network model to predict the future load so that the scheduling system features a self-learning capability and good adaptability to the change of load. Moreover, it separates static requests from dynamic requests to make full use of the CPU resources and takes the locality of requests into account to improve the cache hit ratio. Experimental re suits from the testing tool of WebBench^TM show better per formance for Web cluster server with TDSA than that with traditional scheduling algorithms.展开更多
基金The National Natural Science Foundation of China(No.61402333,61402242)the National Science Foundation of Tianjin(No.15JCQNJC00400)
文摘An approach for web server cluster(WSC)reliability and degradation process analysis is proposed.The reliability process is modeled as a non-homogeneous Markov process(NHMH)composed of several non-homogeneous Poisson processes(NHPPs).The arrival rate of each NHPP corresponds to the system software failure rate which is expressed using Cox s proportional hazards model(PHM)in terms of the cumulative and instantaneous load of the software.The cumulative load refers to software cumulative execution time,and the instantaneous load denotes the rate that the users requests arrive at a server.The result of reliability analysis is a time-varying reliability and degradation process over the WSC lifetime.Finally,the evaluation experiment shows the effectiveness of the proposed approach.
基金Supported by the National Natural Science Foun-dation of China (90204008) the Science Council of Wuhan(20001001004)
文摘There are two kinds of dispatching policies in content-aware web server cluster; segregation dispatching policy and mixture dispatching policy. Traditional scheduling algorithms all adopt mixture dispatching policy. They do not consider that dynamic requests' serving has the tendency to slow down static requests' serving, and that different requests have different resource demands, so they can not use duster's resource reasonably and effectively. This paper uses stochastic reward net (SRN) to model and analyze the two dispatching policies, and uses stochastic Petri net package (SPNP) to simulate the models. The simulation results and practical tests both show that segregation dispatching policy is better than mixture dispatching policy. The principle of segregation dispatching policy can guide us to design efficient scheduling algorithm.
基金This work was supported by the National "863" program of China ( No.2003AA148010) and National Torch Project of China (No.2001EB001233) .
文摘Distributed architectures support increased load on popular web sites by dispatching client requests transparently among multiple servers in a cluster. Packet Single-Rewriting technology and client address hashing algorithm in ONE-IP technology which can ensure application-session-keep have been analyzed, an improved request dispatching algorithm which is simple, effective and supports dynamic load balance has been proposed. In this algorithm, dispatcher evaluates which server node will process request by applying a hash function to the client IP address and comparing the result with its assigned identifier subset; it adjusts the size of the subset according to the performance and current load of each server, so as to utilize all servers' resource effectively. Simulation shows that the improved algorithm has better performance than the original one.
基金(1) Specialized Research Fund for the Doctoral Program of Higher Education (No.20030013006) (2) National Specialized R&D Pro-ject for the Product of Mobile Communications (Devel-opment and Application of Next Generation Mobile In-telligent Network System) (3) Development Fund for Electronic and Information Industry (Value-added Ser-vice Platform and Application System for Mobile Communications).
文摘Based on the demand of the admission control of softswitch-based clustered media server, this pa- per proposed a new dynamic quota-based admission control algorithm that has a sub-negotiation process. The strongpoint of quota-based algorithm had been inherited in the algorithm and at the same time some new ideas had also been introduced into it. Simulations of the algorithm had been conducted on the Petri net model and the results show that this algorithm has excellent performance. In order to find the optimal resource quota set- ting in real time, the paper proposed two approximation analysis methods. It can be seen from analysis results that these two methods can be used to get sub-optimal quota values quickly and effectively. These two ap- proximation analysis methods will play important roles in implementation of the algorithm in system.
基金Supported by the Industrialized Foundation ofHebei Province(020501) the Natural Science Foundation of HebeiUniversity(2005Q04)
文摘In this paper, we explored a load-balancing algorithm in a cluster file system contains two levels of metadata-server, primary-level server quickly distributestasks to second-level servers depending on the closest load-balancing information. At the same time, we explored a method which accurately reflect I/O traffic and storage of storage-node: computing the heat-value of file, according to which we realized a more logical storage allocation. According to the experiment result, we conclude that this new algorithm shortens the executing time of tasks and improves the system performance compared with other load algorithm.
文摘The problem of state space explosion is still ran outstanding challenge in Markovian performance analysis for multiserver multiqueue (MSMQ) systems. The system behavior of a MSMQ system is described using stochastic high-level Petri net (SHLPN) models, and an approximate performance analysis technique is proposed based on decomposition and refinement methods as well as iteration technique. A real MSMQ system, Web-server cluster, is investigated. The performance of an integrated scheme of request dispatching and scheduling is analyzed with both Poisson and self-similar request arrivals. The study shows that the approximate analysis technique significantly reduces the complexity of the model solution and is also efficient for accuracy of numerical results.
基金Project supported by the National Grand Fundamental Research 973 Program of China (Grant No.2004CB318203), and the National Natural Science Foundation of China (Grant No.60603074)
文摘Distribution of metadata in a metadata server cluster is important in mass storage system. A good distribution algorithm has a significant influence on the system performance, availability and scalability. Subtree partition and hash are two traditional metadata distribution algorithms used in distributed file systems. They both have a defect in system scalability. This paper proposes a new directory hash (DH) algorithm. By treating directory as the key value of hash function, implementing concentrated storage of metadata, pipelining operations and prefetching technology, DH algorithm can enhance the system scalability on the premise without sacrificing system performance.
基金Supported by the National Natural Science Funda-tion of China (60175015)
文摘Requests distribution is an key technology for Web cluster server. This paper presents a throughput-driven scheduling algorithm (TDSA). The algorithm adopts the throughput of cluster back-ends to evaluate their load and employs the neural network model to predict the future load so that the scheduling system features a self-learning capability and good adaptability to the change of load. Moreover, it separates static requests from dynamic requests to make full use of the CPU resources and takes the locality of requests into account to improve the cache hit ratio. Experimental re suits from the testing tool of WebBench^TM show better per formance for Web cluster server with TDSA than that with traditional scheduling algorithms.