X. Defago, A. Schiper, and N. Sergent. Semi-passive replication. In Proceedings of the 17th IEEE Symposium on Reliable Distributed Systems (SRDS), 1998.  Home/Search   Document Details and Download   Summary   Related Articles   Check

.... to 2 (line 15) Note that a notification of no leadership imposes RE DM J( and WRITEM J( to fail (i.e. to return 2 ) Consequently if r was serving a request and executing statement 11, it sets primary to 2 (line 13) Note that the proposed implementation adopts an optimistic approach[4]: it allows internal inconsistencies among the sequencer replica states as it requires only a majority of replicas to be updated at the end of each definitive assignment. In other words the implementation sacrifices update atomicity to achieve better performances in failure free runs. The price to ....

X. Dfago, A. Schiper, and N. Sergent. Semi-passive replication. In Proceedings of the 17th IEEE Symposium on Reliable Distributed Systems (SRDS), pages 43-50, West Lafayette, IN, USA, October 1998.

....invoke multiple replicas. When dealing with the replication of a stateful service, system designers face the problem of guaranteeing a given degree of consistency among the local states of the replicas despite concurrent 5 client requests and failures. Active [1] passive [2] and semi passive [3] replication are wellknown approaches to increase the availability of a stateful service while maintaining strong replica consistency. In these approaches each replica embeds the service implementation and uses group communication primitives and services such as total order multicast, view ....

....is executed either by all or none of the server replicas; ordering means that each server replica executes state updates in the same order. 2.3. 2T replication techniques A large number of replication techniques enforcing strong replica consistency have been proposed 15 in the literature, e.g. [1 3,16]. These techniques are based on 2T architectures in which clients directly interact with server replicas embedding group communication toolkits and the actual service implementation. As examples, we summarize here the active and passive replication techniques as they will be used in the next ....

Defago X, Schiper A, Sergent N. Semi-passive replication. Proceedings 17th IEEE Symposium on Reliable Distributed 30 Systems (SRDS), West Lafayette, IN, October 1998; 43--50.

....Interoperable Replication Logic (IRL) architecture which is a Fault Tolerant CORBA compliant infrastructure exploiting a threetier approach to replicate CORBA objects. We finally present an extensive performance study of an IRL prototype. 1 Introduction Active [24] passive [5] semi passive [7] and quorum replication [18] are well known approaches to build fault tolerant services in distributed systems by using software replication. These approaches are inherently two tiers in the sense that independent clients (client tier) interact with a set of server replicas (the end tier) which ....

....respect to classical (two tiers) replication. The price to pay is one additional hop in the client server request invocation path with respect to classical two tier replication. Let us final remark that three tier replication is not an alternative to classical two tiers software replication, e.g. [5, 7, 10, 13, 18, 24] . It is rather an architectural solution when facing software replication in asynchronous distributed systems. Three tier replication actually needs two tier software replication to make the replication logic resilient to failures. The result is an architecture in which (i) asynchronous aspects ....

[Article contains additional citation context not shown here]

X. Defago, A. Schiper, and N. Sergent, Semi-passive replication, Proceedings of the 17th IEEE Symposium on Reliable Distributed Systems (SRDS) (West Lafayette, IN, USA), October 1998, pp. 43--50.

....of a mid tier to free replicas (end tier) and clients (client tier) from the need of underlying partial synchrony assumptions. Then we propose how to specialize the mid tier in order to manage active replication of server replicas. 1 Introduction Active [27] passive [5] semi passive [12] and quorum replication [19, 25] are well known approaches used to keep consistent the internal states of a set of server replicas connected by a communication network. A replicated stateful service should indeed appear to clients as implemented by a single logical entity. Informally, ....

X. Defago, A. Schiper, and N. Sergent. Semi-passive replication. In Proceedings of the 17th IEEE Symposium on Reliable Distributed Systems (SRDS), pages 43--50, West Lafayette, IN, USA, October 1998.

....dependence at the middle tier [10] Our e Transaction protocol uses the very same replication scheme, both as a highly available storage for the transaction processing state , and as an e#ective way to retry transactions behind the scenes. In contrast to most replication schemes we know about [11, 12, 13, 14], we assume stateless servers that interact with third party databases replication schemes have usually been designed in a client server context: servers are stateful but do not interact with third party entities. Another 2 A wo register can also be viewed as a distributed form of software ....

X. Defago, A. Schiper, and N. Sergent, "Semi-passive replication," in Proceedings of the IEEE Symposium on Reliable Distributed Systems, October 1998. 15

No context found.

X. Defago, A. Schiper, and N. Sergent. Semi-passive replication. In Proc. of the 17th Symposium on Reliable Distributed Systems (SRDS'98), pages 43--50, West Lafayette, Indiana, Oct. 1998.

No context found.

X. Defago, A. Schiper, and N. Sergent. Semi-passive replication. In 17th Symposium on Reliable Distributed Systems (SRDS'98), pages 43--50, October 1998. 162

No context found.

X. D efago, A. Schiper, and N. Sergent. Semi-passive replication. In Proc. 17th IEEE Intl. Symp. on Reliable Distributed Systems (SRDS-17), pages 43--50, West Lafayette, IN, USA, Oct. 1998.

No context found.

X. D efago, A. Schiper, and N. Sergent. Semi-passive replication. In Proc. of the 17th IEEE Symposium on Reliable Distributed Systems (SRDS'98), pages 43--50, West Lafayette, Indiana, October 1998.

....be made on a per invocation basis. algorithm is used by the replicated objects to agree on a set of messages to deliver and their respective ordering, thus ensuring that the copies of the replicated object remain consistent. A passive replication algorithm based on consensus has been proposed in [4]. This algorithm is based on a variant of consensus, called consensus (Deferred Initial Values consensus) In contrast with the traditional consensus problem, participants are not required to have an initial value when starting the consensus algorithm. Instead, the consensus algorithm asks ....

....backups in passive replication. This behavior is depicted in Figure 5, in which is the primary and the coordinator of the consensus algorithm, while and are the backups. A more detailed description of the consensus algorithm and its application to passive replication can be found in [4]. We have used the consensus algorithm presented in [4] for providing a generic consensus service that lets a set of CORBA objects agree on application specific data. Basing agreement protocols on a generic consensus service bears many advantages. It decouples the algorithmic aspects from the ....

[Article contains additional citation context not shown here]

X. D efago, A. Schiper, and N. Sergent. Semi-passive replication. In Proceedings of the 17th Symposium on Reliable Distributed Systems (SRDS-17), West Lafayette, Indiana, USA, October 1998.

....1 Atomic Broadcast (request) processing Active Replication client replica 1 = primary replica 2 send (request) processing state update response response Passive Replication ack Fig. 1. Principle of active vs passive replication 3. 2 Semi passive Replication Semi passive replication [13,12] is a variant of passive replication: it retains its major characteristics (e.g. allows for non deterministic processing) The main difference between passive and semi passive replication is the selection of the primary. In semipassive replication the selection of the primary is based on the ....

X. Defago, A. Schiper, and N. Sergent. Semi-passive Replication. In 17th IEEE Symp. on Reliable Distributed Systems (SRDS-17), pages 43--58, West Lafayette, USA, October 1998.

....which every process in # # that crashes is permanently suspected by all correct processes in # # . EVENTUAL WEAK ACCURACY) There is a time after which some correct process in # # is never suspected by any correct process in # # . An earlier version of this algorithm was called ###consensus [13]. Note that ###consensus used to designate an algorithm, whereas Lazy Consensus now designates a problem. round 1 decide ack # S Figure 6: Lazy Consensus with ## (no crash, no suspicion) 6.2 Lazy Consensus Algorithm using The algorithm for Lazy Consensus is adapted from the consensus ....

X. Defago, A. Schiper, and N. Sergent. Semi-passive replication. In Proceedings of the 17th IEEE Symposium on Reliable Distributed Systems (SRDS), pages 43--50, West Lafayette, IN, USA, October 1998.

....presented in the previous section aim at improving the response time in the case where no failure occur. In this section, we present a modification that dramatically reduces the time needed to react to the crash of a replica. The concepts on which this modification is based were first presented in [4]. It has been implemented successfully in the context of OGS. 5.1. Two time out group membership A slow reaction to failure is a common problem of group communication platforms (e.g. Isis [2] Horus [2, 16] Consul [12] Transis [1] Phoenix [10] for asynchronous systems. In such platforms, ....

X. Defago, A. Schiper, and N. Sergent. Semi-passive replication. In Proceedings of the 17th IEEE Symposium on Reliable Distributed Systems (SRDS), pages 43--50, West Lafayette, IN, USA, Oct. 1998.

....stage operation phase, then (2) solves an agreement problem with all replica agents, and (3) finally a i 1 ; M i 1 is sent to the next stage. Items (1) and (2) are done together as part of a variant of the consensus problem, called Deferred Initial Value Consensus (DIV Consensus for short) [3]. DIV Consensus is the first building block of our FATOMAS system. In the consensus problem, each process needs an initial value at the beginning of consensus [1] In our problem, the initial value at stage S i for place p j i is obtained by executing agent a i . Executing a i on all the places ....

X. Defago, A. Schiper, and N. Sergent. Semi-passive replication. In Proc. of the 17th IEEE Symposium on Reliable Distributed Systems (SRDS), pages 43--50, West Lafayette, IN, Oct. 1998.

....agreement. No two processes in decide differently. Uniform validity. If a process in decides some value v, then v was proposed by some process in . Reference [3] solves the consensus problem with the unreliable failure detector 3S and a majority of correct processes. DIV consensus 10 [5] modifies the consensus problem such that all processes need not have an initial value. The initial value is computed during the execution of the consensus algorithm, whenever needed. Specifically, in the absence of failures, only one process computes the initial value. For this purpose, the ....

X. Defago, A. Schiper, and N. Sergent. Semi-passive replication. In Proceedings of the 17th IEEE Symposium on Reliable Distributed Systems (SRDS), pages 43--50, West Lafayette, IN, USA, Oct. 1998.

....the stage operation phase, then (2) solves an agreement problem with all replica agents, and (3) finally ## ### ## ### # is sent to the next stage. Items (1) and (2) are done together as part of a variant of the consensus problem, called Deferred Initial Value Consensus (DIV Consensus for short) [3]. DIV Consensus is the first building block of our FATOMAS system. In the consensus problem, each process needs an initial value at the beginning of consensus [1] In our problem, the initial value at stage # # for place # # # is obtained by executing agent # # . Executing # # on all the places ....

X. Defago, A. Schiper, and N. Sergent. Semi-passive replication. In Proc. of the 17th IEEE Symposium on Reliable Distributed Systems (SRDS), pages 43--50, West Lafayette, IN, Oct. 1998.

....No two processes in# decide differently. # Uniform validity. If a process in# decides some value #, then # was proposed by some process in# . Reference [3] solves the consensus problem with the unreliable failure detector ## and a majority of correct processes. DIV consensus 10 [5] modifies the consensus problem such that all processes need not have an initial value. The initial value is computed during the execution of the consensus algorithm, whenever needed. Specifically, in the absence of failures, only one process computes the initial value. For this purpose, the ....

X. Defago, A. Schiper, and N. Sergent. Semi-passive replication. In Proceedings of the 17th IEEE Symposium on Reliable Distributed Systems (SRDS), pages 43--50, West Lafayette, IN, USA, Oct. 1998.

....10 3. All replicas execute the request in the order they are delivered. 4. In case of a non deterministic choice, the leader informs the followers using the View Synchronous Broadcast. 5. The servers sends back the response to the client. 3. 5 Semi Passive Replication Semi passive replication [DSS98] is a variant of passive replication which can be implemented in the asynchronous model without requiring the view synchronous communication mechanism, i.e. without requiring the notion of views. The main advantage over passive replication is two allow for aggressive time outs value to suspect ....

X. Dfago, A. Schiper, and N. Sergent. Semi-passive replication. In Proceedings of the 17th IEEE Symposium on Reliable Distributed Systems (SRDS), pages 43--50, West Lafayette, IN, USA, October 1998.

....3. All replicas execute the request in the order they are delivered. 4. In case of a non deterministic choice, the leader informs the followers using the View Synchronous Broadcast. 5. The servers send back the response to the client. 3.5. Semi Passive Replication Semi passive replication [10] is a variant of passive replication which can be implemented in the asynchronous model without requiring any notion of views. The main advantage over passive replication is to allow for aggressive time outs values and suspecting crashed processes without incurring too high a cost for incorrect ....

X. Dfago, A. Schiper, and N. Sergent. Semi-passive replication. In Proceedings of the 17th IEEE Symposium on Reliable Distributed Systems (SRDS), pages 43--50, West Lafayette, IN, USA, Oct. 1998.

No context found.

X. Defago, A. Schiper, and N. Sergent. Semi-passive replication. In Proceedings of the 17th IEEE Symposium on Reliable Distributed Systems (SRDS), 1998.

No context found.

X. Defago, A. Schiper, and N. Sergent. Semi-passive replication. In Symposium on Reliable Distributed Systems, pages 43--50, West Lafayette, IN, October 1998.

No context found.

X. Defago, A. Schiper, and N. Sergent, "Semi-passive replication," in Proceedings of the 17th IEEE Symposium on Reliable Distributed Systems, 1998, pp. 43-50.

No context found.

X. Defago, A. Schiper, and N. Sergent, "Semi-Passive Replication," Proceedings of Symposium on Reliable Distributed Systems, pp. 43--50, 1998.

No context found.

X. D efago, A. Schiper, and N. Sergent. Semi-Passive Replication. In Proceedings of the 17th Symposium on Reliable Distributed Systems (SRDS-17), pages 43--50, October 1998.

No context found.

X. Defago, A. Schiper, and N. Sergent. Semi-passive replication. In Proceedings of the IEEE Symposium on Reliable Distributed Systems, October 1998.

First 50 documents

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC