The replication scheme of a distributed database determines how many replicas of each object are created, and to which processors these replicas are allocated. This scheme critically affects the performance of a distributed system, since reading an object locally is less costly than reading it from a remote processor. Therefore in a read-intensive network a widely distributed replication is mandated. On the other hand, an update of an object is usually written to all, or a majority of the replicas, and therefore in a write-intensive network a narrowly distributed replication is mandated. In other words, the optimal replication scheme depends on the read-write pattern for each object. The authors propose a practical algorithm, called dynamic-data-allocation (DDA), that changes the replication scheme of an object (i.e. the processors which store a replica of the object) dynamically as the read-write pattern of the object changes in the network. They assume that the changes in the read-write pattern are not known a priori.<>
{"title":"An algorithm for dynamic data distribution","authors":"O. Wolfson, S. Jajodia","doi":"10.1109/MRD.1992.242616","DOIUrl":"https://doi.org/10.1109/MRD.1992.242616","url":null,"abstract":"The replication scheme of a distributed database determines how many replicas of each object are created, and to which processors these replicas are allocated. This scheme critically affects the performance of a distributed system, since reading an object locally is less costly than reading it from a remote processor. Therefore in a read-intensive network a widely distributed replication is mandated. On the other hand, an update of an object is usually written to all, or a majority of the replicas, and therefore in a write-intensive network a narrowly distributed replication is mandated. In other words, the optimal replication scheme depends on the read-write pattern for each object. The authors propose a practical algorithm, called dynamic-data-allocation (DDA), that changes the replication scheme of an object (i.e. the processors which store a replica of the object) dynamically as the read-write pattern of the object changes in the network. They assume that the changes in the read-write pattern are not known a priori.<<ETX>>","PeriodicalId":314844,"journal":{"name":"[1992 Proceedings] Second Workshop on the Management of Replicated Data","volume":"24 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1992-11-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"125210310","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Describes Manetho, an experimental protocol system, whose goal is to explore the extent to which transparent fault tolerance can be added to long-running distributed applications. Transparent techniques are attractive because they can automatically add fault tolerance to existing applications that were written without consideration for reliability. Previous techniques for providing transparent fault-tolerance relied on rollback-recovery. However, rollback recovery is not appropriate for server processes where the lack of service during rollback is intolerable. Furthermore, rollback-recovery assumes that a process can be restarted on any available host. As a result, extended downtime cannot be tolerated for example in file servers, which have to run on the host where the disks reside. Manetho solves these problems with an integrated approach by using process replication for server processes and rollback-recovery for client processes.<>
{"title":"An integrated approach to fault tolerance","authors":"E. Elnozahy, W. Zwaenepoel","doi":"10.1109/MRD.1992.242611","DOIUrl":"https://doi.org/10.1109/MRD.1992.242611","url":null,"abstract":"Describes Manetho, an experimental protocol system, whose goal is to explore the extent to which transparent fault tolerance can be added to long-running distributed applications. Transparent techniques are attractive because they can automatically add fault tolerance to existing applications that were written without consideration for reliability. Previous techniques for providing transparent fault-tolerance relied on rollback-recovery. However, rollback recovery is not appropriate for server processes where the lack of service during rollback is intolerable. Furthermore, rollback-recovery assumes that a process can be restarted on any available host. As a result, extended downtime cannot be tolerated for example in file servers, which have to run on the host where the disks reside. Manetho solves these problems with an integrated approach by using process replication for server processes and rollback-recovery for client processes.<<ETX>>","PeriodicalId":314844,"journal":{"name":"[1992 Proceedings] Second Workshop on the Management of Replicated Data","volume":"41 7","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1992-11-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"120907301","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
The authors describe the implementation and an initial performance study of Frolic, a cluster-based dynamic file replication system. This system specifically targets large scale commercial environments with extensive file sharing, although it is expected to also perform well in systems with limited file sharing as well. Instead of keeping copies of a widely shared file at each client workstation, these files are dynamically replicated onto the cluster file servers, so that they become locally available. The costs of maintaining consistency among the limited number of server copies should be much less than that among a large number of clients. The problems of network and server congestion as well as that of high access latency may be alleviated. Simulation studies using a statistical workload have shown that cluster-based file replication can significantly reduce file access times for a wide range of workload parameters.<>
{"title":"Implementation and performance of cluster-based file replication in large-scale distributed systems","authors":"J.Y.C. Pang, D. S. Gill, Songnian Zhou","doi":"10.1109/MRD.1992.242607","DOIUrl":"https://doi.org/10.1109/MRD.1992.242607","url":null,"abstract":"The authors describe the implementation and an initial performance study of Frolic, a cluster-based dynamic file replication system. This system specifically targets large scale commercial environments with extensive file sharing, although it is expected to also perform well in systems with limited file sharing as well. Instead of keeping copies of a widely shared file at each client workstation, these files are dynamically replicated onto the cluster file servers, so that they become locally available. The costs of maintaining consistency among the limited number of server copies should be much less than that among a large number of clients. The problems of network and server congestion as well as that of high access latency may be alleviated. Simulation studies using a statistical workload have shown that cluster-based file replication can significantly reduce file access times for a wide range of workload parameters.<<ETX>>","PeriodicalId":314844,"journal":{"name":"[1992 Proceedings] Second Workshop on the Management of Replicated Data","volume":"243 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1992-11-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"124683498","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
The author mainly concentrates on transactional distributed systems. Most previous research on replication in such environments has concentrated in employing replication to achieve high availability. The position is that high availability along is not enough. First, it is important to consider the cost of providing high availability through replication. Second, one must exploit the potential of replication as a means of improving performance. Then performance issues (in addition to availability) in which one is mostly interested are: transaction latency, bottlenecks and throughput, and scalability (in particular as it affects the former issues). The author briefly outlines his related research efforts which can be classified in the following areas: replication-control protocols, recovery strategies, and studying availability in large-scale distributed systems.<>
{"title":"High availability is not enough (distributed systems)","authors":"P. Triantafillou","doi":"10.1109/MRD.1992.242621","DOIUrl":"https://doi.org/10.1109/MRD.1992.242621","url":null,"abstract":"The author mainly concentrates on transactional distributed systems. Most previous research on replication in such environments has concentrated in employing replication to achieve high availability. The position is that high availability along is not enough. First, it is important to consider the cost of providing high availability through replication. Second, one must exploit the potential of replication as a means of improving performance. Then performance issues (in addition to availability) in which one is mostly interested are: transaction latency, bottlenecks and throughput, and scalability (in particular as it affects the former issues). The author briefly outlines his related research efforts which can be classified in the following areas: replication-control protocols, recovery strategies, and studying availability in large-scale distributed systems.<<ETX>>","PeriodicalId":314844,"journal":{"name":"[1992 Proceedings] Second Workshop on the Management of Replicated Data","volume":"37 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1992-11-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"116308093","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
The author compares two replication schemes designed to provide high availability in an efficient manner: HA-NFS and ARM. Both schemes use the primary copy method for replica control. Both schemes were designed with the goal of minimizing the overheads during failure-free operation. In a primary copy scheme these overheads primarily consist of updating the secondary replicas. The two schemes were designed for different applications; ARM for providing high availability in a Shared Nothing database system, HA-NFS for providing high availability in an NFS file server environment. They also differ in that the HA-NFS scheme uses dual-ported disks to provide high availability, the ARM scheme uses replication over a network. In spite of the seemingly major differences, the schemes have the same key conceptual idea viz. propagating updates asynchronously to remote replicas. In addition to this idea, HA-NFS uses an unusual hardware arrangement in the form of dual-ported disks to further lower the overhead of updating secondary replicas.<>
{"title":"Experiences with two high availability designs (replication techniques)","authors":"A. Bhide","doi":"10.1109/MRD.1992.242618","DOIUrl":"https://doi.org/10.1109/MRD.1992.242618","url":null,"abstract":"The author compares two replication schemes designed to provide high availability in an efficient manner: HA-NFS and ARM. Both schemes use the primary copy method for replica control. Both schemes were designed with the goal of minimizing the overheads during failure-free operation. In a primary copy scheme these overheads primarily consist of updating the secondary replicas. The two schemes were designed for different applications; ARM for providing high availability in a Shared Nothing database system, HA-NFS for providing high availability in an NFS file server environment. They also differ in that the HA-NFS scheme uses dual-ported disks to provide high availability, the ARM scheme uses replication over a network. In spite of the seemingly major differences, the schemes have the same key conceptual idea viz. propagating updates asynchronously to remote replicas. In addition to this idea, HA-NFS uses an unusual hardware arrangement in the form of dual-ported disks to further lower the overhead of updating secondary replicas.<<ETX>>","PeriodicalId":314844,"journal":{"name":"[1992 Proceedings] Second Workshop on the Management of Replicated Data","volume":"19 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1992-11-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"127179287","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
The author discusses some theoretical limitations on the potential benefits of replication. In particular, he investigates two fundamental questions: (1) does placing copies of data around a network increase the probability that the data will be available, and (2) does such a technique decrease the mean duration of unavailability of the data. Given that many applications require mutually exclusive access to the data, he shows that the potential benefits of replication are rather low with respect to both of these metrics. Although these results are not necessarily surprising, it is interesting that the proofs are protocol independent and, in the case of the availability measure, topology independent. They are useful, therefore, in focusing attention on the replica consistency model and performance measures and away from any particular protocol or network configuration.<>
{"title":"Bounds on the effects of replication on availability","authors":"L. Raab","doi":"10.1109/MRD.1992.242620","DOIUrl":"https://doi.org/10.1109/MRD.1992.242620","url":null,"abstract":"The author discusses some theoretical limitations on the potential benefits of replication. In particular, he investigates two fundamental questions: (1) does placing copies of data around a network increase the probability that the data will be available, and (2) does such a technique decrease the mean duration of unavailability of the data. Given that many applications require mutually exclusive access to the data, he shows that the potential benefits of replication are rather low with respect to both of these metrics. Although these results are not necessarily surprising, it is interesting that the proofs are protocol independent and, in the case of the availability measure, topology independent. They are useful, therefore, in focusing attention on the replica consistency model and performance measures and away from any particular protocol or network configuration.<<ETX>>","PeriodicalId":314844,"journal":{"name":"[1992 Proceedings] Second Workshop on the Management of Replicated Data","volume":"11 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1992-11-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"131141944","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Early replica control work focused on maintaining the availability of an object despite the unavailability of its copies. More recent work has focused on improving the response time, autonomy, and scalability of the earlier work while preserving their availability characteristics. The authors analyze the several ways the new algorithms make such improvements.<>
{"title":"An analysis of replica control","authors":"Shu-Wie F. Chen, C. Pu","doi":"10.1109/MRD.1992.242625","DOIUrl":"https://doi.org/10.1109/MRD.1992.242625","url":null,"abstract":"Early replica control work focused on maintaining the availability of an object despite the unavailability of its copies. More recent work has focused on improving the response time, autonomy, and scalability of the earlier work while preserving their availability characteristics. The authors analyze the several ways the new algorithms make such improvements.<<ETX>>","PeriodicalId":314844,"journal":{"name":"[1992 Proceedings] Second Workshop on the Management of Replicated Data","volume":"275 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1992-11-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"125187663","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
The authors derive lower bounds and the corresponding optimal protocols for three parameters for synchronous primary-backup systems. They compare their results with similar results for active replication in order to determine whether the common folklore on the virtues of the two approaches can be shown formally. They also extend some of their results to asynchronous primary-backup systems. They implement an important subclass of primary-backup protocols that they call 0-blocking. These protocols are interesting because they introduce no additional protocol related delay into a failure-free service request. Through implementing these protocols the authors hope to determine the appropriateness of their theoretical system model and uncover other practical advantages or limitations of the primary-backup approach.<>
{"title":"Highly-available services using the primary-backup approach","authors":"Navin Budhiraja, K. Marzullo","doi":"10.1109/MRD.1992.242619","DOIUrl":"https://doi.org/10.1109/MRD.1992.242619","url":null,"abstract":"The authors derive lower bounds and the corresponding optimal protocols for three parameters for synchronous primary-backup systems. They compare their results with similar results for active replication in order to determine whether the common folklore on the virtues of the two approaches can be shown formally. They also extend some of their results to asynchronous primary-backup systems. They implement an important subclass of primary-backup protocols that they call 0-blocking. These protocols are interesting because they introduce no additional protocol related delay into a failure-free service request. Through implementing these protocols the authors hope to determine the appropriateness of their theoretical system model and uncover other practical advantages or limitations of the primary-backup approach.<<ETX>>","PeriodicalId":314844,"journal":{"name":"[1992 Proceedings] Second Workshop on the Management of Replicated Data","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1992-11-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"131044633","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
In the Tapestry system developed at Xerox PARC, users provide queries to filter incoming streams of documents. These queries run continuously over a growing database of electronic mail messages, news articles, and other textual documents. In the current implementation, all filter queries for all users run on a single database server. Those documents that match a user's filter query (or queries) are queued up for the user and can be retrieved directly via an RPC interface or, as is typically the case, can be sent to the user via electronic mail. This paper shows that the design of a distributed information filtering service involves challenges not faced in other distributed applications. Replication is needed, not for fault-tolerance or performance but simply for scalability. Of course, once replication is provided, it can be used to increase the fault-tolerance of the system (with some additional work). A new technique called filter-based replication is proposed for deciding what to replicate and where.<>
{"title":"Replication in an information filtering system","authors":"D. Terry","doi":"10.1109/MRD.1992.242615","DOIUrl":"https://doi.org/10.1109/MRD.1992.242615","url":null,"abstract":"In the Tapestry system developed at Xerox PARC, users provide queries to filter incoming streams of documents. These queries run continuously over a growing database of electronic mail messages, news articles, and other textual documents. In the current implementation, all filter queries for all users run on a single database server. Those documents that match a user's filter query (or queries) are queued up for the user and can be retrieved directly via an RPC interface or, as is typically the case, can be sent to the user via electronic mail. This paper shows that the design of a distributed information filtering service involves challenges not faced in other distributed applications. Replication is needed, not for fault-tolerance or performance but simply for scalability. Of course, once replication is provided, it can be used to increase the fault-tolerance of the system (with some additional work). A new technique called filter-based replication is proposed for deciding what to replicate and where.<<ETX>>","PeriodicalId":314844,"journal":{"name":"[1992 Proceedings] Second Workshop on the Management of Replicated Data","volume":"57 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1992-11-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"132312619","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
The authors recommend replicating the procedure call abstraction as a method for constructing highly available distributed programs. In order to make highly available systems much more widespread than they are today, one must make it easier for application developers to incorporate replication into one's systems. Given that remote procedure call has proven to be a useful abstraction in building distributed programs, replicated procedure call seems to be an appropriate method for introducing high availability while hiding the complexities of replication. The authors argue that the simplicity and familiarity of the procedure call mechanism makes it an excellent model for introducing replication, and they discuss different choices one can make in designing a replicated procedure call system.<>
{"title":"Replicating the procedure call abstraction","authors":"R. Ladin, M. Mazer, A. Wolman","doi":"10.1109/MRD.1992.242610","DOIUrl":"https://doi.org/10.1109/MRD.1992.242610","url":null,"abstract":"The authors recommend replicating the procedure call abstraction as a method for constructing highly available distributed programs. In order to make highly available systems much more widespread than they are today, one must make it easier for application developers to incorporate replication into one's systems. Given that remote procedure call has proven to be a useful abstraction in building distributed programs, replicated procedure call seems to be an appropriate method for introducing high availability while hiding the complexities of replication. The authors argue that the simplicity and familiarity of the procedure call mechanism makes it an excellent model for introducing replication, and they discuss different choices one can make in designing a replicated procedure call system.<<ETX>>","PeriodicalId":314844,"journal":{"name":"[1992 Proceedings] Second Workshop on the Management of Replicated Data","volume":"11 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1992-11-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"125768272","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：