A Throughput-Driven Scheduling Algorithm of Differentiated Service for Web Cluster

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.

Modeling and Performance Analysis of Dispatching Policies in Dispatcher-Based Web Server Cluster

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.

DYNAMIC REQUEST DISPATCHING ALGORITHM FOR WEB SERVER CLUSTER

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.

Reliability analysis of web server cluster systems based on proportional hazards model

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.

A NEW DYNAMIC QUOTA-BASED ADMISSION CONTROL WITH SUB-NEGOTIATION FOR SOFTSWITCH-BASED CLUSTERED MEDIA SERVER

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.

Highly scalable metadata distribution algorithm in mass storage system

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.

An Efficient Load Balancing Approach for N-Hierarchical Web Server Cluster

Aiming at the load imbalance and poor scalability in single-tier Web server clusters, an efficient load balancing ap- proach is proposed for constructing an N-hierarchical （multi-tier） Web server cluster. In each layer, multiple load balancers are set to receive the user requests simultaneously, and different load bal- ancing algorithms are used to construct the high-scalable Web cluster system. At the same time, an improved load balancing al- gorithm is proposed, which can dynamically calculate weights according to the utilization of the server resources, and reasonably distribute the loads for each server according to the load status of the servers. The experimental results show that the proposed ap- proach can greatly decrease the load imbalance among the Web servers and reduce the response time of the entire Web cluster system.

HWM： a hybrid workload migration mechanism of metadata server cluster in data center

In data center, applications of big data analytics pose a big challenge to massive storage systems. It is signif- icant to achieve high availability, high performance and high scalability for PB-scale or EB-scale storage systems. Meta- data server （MDS） cluster architecture is one of the most effective solutions to meet the requirements of applications in data center. Workload migration can achieve load balance and energy saving of duster systems. In this paper, a hybrid workload migration mechanism of MDS cluster is proposed and named as HWM. In HWM, workload of MDS is classi- fied into two categories： metadata service and state service, and they can be migrated rapidly from a source MDS to a target MDS in different ways. Firstly, in metadata service migration, all the dirty metadata of one sub file system is flushed to a shared storage pool by the source MDS, and then is loaded by the target MDS. Secondly, in state service mi- gration, all the states of that sub file system are migrated from source MDS to target MDS through network at file granular- ity, and then all of the related structures of these states are reconstructed in target MDS. Thirdly, in the process of work- load migration, instead of blocking client requests, the source MDS can decide which MDS will respond to each request according to the operation type and the migration stage. The proposed mechanism is implemented in the Blue Whale MDS cluster. The performance measurements show that the HWM mechanism is efficient to migrate the workload of a MDS cluster system and provides low-latency access to metadata and states.

LinuxDirector: A Connection Director for Scalable Internet Services

LinuxDirector is a connection director that supports load balancing among multiple Internet servers, which can be used to build scalable Internet services based on clusters of servers. LinuxDirector extends the TCP/IP stack of Linux kernel to support three IP load balancing techniques, VS/NAT, VS/TUN and VS/DR. Four scheduling algorithms have been implemented to assign connections to different servers. Scalability is achieved by transparently adding or removing a node in the cluster. High availability is provided by detecting node or daemon failures and reconfiguring the system appropriately. This paper describes the design and implementation of LinuxDirector and presents several of its features including scalability, high availability and connection affinity.