M.G. Hayden. The ensemble system. PhD thesis, 1998.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....to improve the robustness of COTS micro kernels. By verifying certain predicates when a system call is performed, the wrapper detects errors due to corrupted parameters and may optionally perform some corrective actions to restore the system into a consistent state. Horus [12] and Ensemble [6] are two examples of modular group communication systems where protocol layers can be stacked on top of each other in a variety of ways. Exception handling mechanisms in several languages have been studied in [2] Some of them focus on language design. For example, both Mesa [8] and exceptional C ....

Mark Hayden. The Ensemble System. PhD thesis, January 1998. 11

....can send UDP IP multicast messages. PMNetwork uses the PM portable communication library [7] which bypasses the TCP IP stack to take advantage of low latency cluster interconnection networks, like Myrinet. Finally, EnsembleNetwork integrates the Ensemble group communication framework [8, 9] into Neko to perform reliable IP multicasts. Neko is focused on constructing prototypes. Nevertheless, we performed some measurements to evaluate Neko s performance. We compared Neko s performance with the performance of Java sockets, using both TCP and UDP. According to [10] the performance of ....

M. Hayden, "The Ensemble system," Technical Report TR98-1662, Department of Computer Science, Cornell University, 1998.

....scalability, and performance. Notable work includes the Delta4 project [Pow91] Aurora [AMW] and Chameleon [Bag98] among others. Providing fault tolerance at the process level through the use of the group communication paradigm has also been studied. Work in this area includes Ensemble [Hay98], Cactus [Bha97] and Rampart [Rei95] There have also been many attempts to provide fault tolerance to distributed CORBA applications by using group communication. Those projects include Maestro [Vay98] OpenDREAMS [Fel96] and Eternal [Nar99] The Object Management Group (OMG) which designed ....

....and system resources are used in a cost effective manner. An object factory that resides on each host is used to create and kill objects. The replicated objects form a group. Messages communicated among different objects are sent through groups. A group communication system (Maestro Ensemble [Vay98, Hay98]) is used to provide reliable message multicast, totally ordered message delivery, and group membership maintenance. In AQuA, we developed a unique proxy approach that uses AQuA gateways to give a client object the perception that it is talking to a single remote object. The AQuA gateway has a ....

M. G. Hayden, "The Ensemble System," Ph.D. thesis, Cornell University, 1998.

....under varying operating conditions. We use the Ensemble GCS as the basis for our experiments. Ensemble [27] is a popular group communication system developed at Cornell University. It was written in the OCAML dialect of the ML language so that it would be amenable to automated proof checking [16]. Ensemble communication stacks have a modular structure and are built up of several layers of micro protocols stacked on each other. Ensemble allows applications to specify which micro protocol layers to use in their stacks, thus allowing applications to choose group properties. Layers implement ....

....protocol itself, to help the reader better understand the results. 3. 1 Methodology for State Machine Construction Construction of a state machine to represent the temporal behavior of a highly layered, event driven stack like Ensemble is di#cult due to a lack of order between layer executions [16] and the simultaneous existence of multiple event flows in the stack. Existing specifications have either not considered the interaction between layers [14] or not captured the temporal behavior of the protocol [17] One approach would be to model each layer in isolation, and compose the ....

M. Hayden. The Ensemble System. PhD thesis, Comp. Sci., Cornell Univ., 1997.

....of the service are not group members and thus need not have access to JBora clients may interact with the service through TCP, for example. This feature contrasts to most other models for group communication, in which clients must run locally part of the group communication protocols (e.g. [Hayden 1998, Keidar et al. 1999] We use the terms process and group member interchangeably. Each process receives from JBora either messages or view changes. A view is a set of process identifiers, corresponding to those processes that can mutually communicate. We say that a process p delivers a view ....

....A prototype of JBora is being implemented in Java on top of JavaGroups, an open source group communication platform available at http: javagroups.sourceforge.net . JavaGroups is written in Java and its internals are based on a layered architecture like Horus [Van Renesse et al. 1996] and Ensemble [Hayden 1998]. We had to implement a JavaGroups layer for supporting partitionable membership, as JavaGroups provided only primary partition membership 3. Throughput measurements for PUTotMcast(m) can be found in [Bartoli et al. 2001] along with more details about its implementation and a correctness sketch. ....

M. Hayden, "The Ensemble System", Ph.D. Thesis, Department of Computer Science, Cornell University, 1998.

....group involves agreement of the members of that group, which involves cooperation of multiple managers and subordinates. The local action merely initiates that process. In order to switch effectively between tolerance of crash and Byzantine failures, the ITUA project is extending the Ensemble [Hay98] group communication system. 5. Group Communication System Ensemble is a group communication system that tolerates crash failures. It consists of several protocol layers; their combination leads to various protocol stacks. In order to be used in the context of the ITUA project, Ensemble needs ....

M.G. Hayden, "The Ensemble System," Ph.D. thesis, Cornell University, 1998.

....has a high level method (using Quality Objects (QuO) Zin97, Loy98] that applications use to specify their desired dependability levels, and has a dependability manager that attempts to reconfigure a system at runtime so that dependability requests are satisfied. AQuA uses the Maestro Ensemble [Hay98, Vay98] group communication system to provide reliable multicast and total ordering. Previous papers on AQuA have addressed the overall architecture of AQuA [Cuk98, Ren99] and the algorithms used to tolerate crash failures [Sab99] In this paper, we focus on the issue of simultaneously tolerating value ....

....algorithms. Several related research projects are described and compared to AQuA in Section 6. Section 7 concludes the paper. 2. AQUA REVIEW Figure 1 shows the various components of the AQuA architecture and their interactions. The AQuA system uses the Ensemble group communication system [Hay98, Bir96] to ensure reliable communication between groups of processes. Ensemble assumes that the process failures are fail silent (i.e. fail only by crashing) and detects process failures through the use of I am alive messages. The AQuA architecture uses this detection mechanism to detect crash ....

M. G. Hayden, "The Ensemble System," Ph.D. thesis, Cornell University, 1998.

....of fault tolerance. Therefore, the AQuA system aims to adapt to changes in the system configuration while providing fault tolerance. The AQuA architecture is layered. The four layers, from top to bottom, are the CORBA application itself, Quality Objects (QuO) 4, 5] Proteus, and Maestro Ensemble [6, 7]. Each layer only communicates with the layer immediately above or below it. The QuO layer provides the CORBA application with an interface to specify its fault tolerance requirements at a high level. Proteus aims to translate the high level requirements of the application into a system ....

....for dynamically specifying dependability requirements. Proteus provides fault tolerance to the CORBA application. Proteus is composed of a dependability manager, an object factory, and a gateway. Figure 1 shows the AQuA architecture. Figure 1. AQuA Architecture 2.1. Maestro Ensemble Ensemble [6] is a group communication system developed at Cornell University. It provides services to an application developer for reliable distributed communication. Ensemble focuses on communication of processes within groups. In each process group, there is a set of members. An Ensemble group member is a ....

M. G. Hayden, "The Ensemble system," Ph.D. dissertation, Cornell University, Ithaca, NY, 1998.

....value, or time faults) the styles of replication to use (active or passive) the types of voting to use ( pass first, leader only, or majority) the degrees of replication to use, and the location of the replicas on the different hosts, among other factors. Proteus uses Maestro Ensemble [Hay98, Vay98] as the underlying group communication system that provides reliable multicast, total ordering, and virtual synchrony. Communication between all architecture components is done using gateways, which translate CORBA object invocations into messages that are transmitted via Ensemble, and contain ....

M. G. Hayden, "The Ensemble System," Ph.D. thesis, Cornell University, 1998.

....the node group facility provided by Loki that have the potential to cause problems in an experiment. The first concept is that membership lists are imprecise, and the second is that no global message ordering is provided. Although algorithms exist in group communication systems (such as Ensemble [11]) for maintaining a consistent membership view as well as global ordering of messages, these algorithms are not desirable in a system such as Loki since these algorithms have overhead that would cause Loki to be more intrusive to the system. 26 When designing an experiment using Loki, it is ....

M. Hayden, "The ensemble system," Ph.D. dissertation, Cornell University, Ithaca, NY, 1998.

....The work on Protocol Boosters [30] focuses on improving the performance of existing protocols by adding booster components that can be combined by nesting, resulting in a linear order of booster components. Finally, hierarchical composition models are used in the Horus [31] and Ensemble [32] [33] systems to construct configurable group communication services as stacks of protocol components. The interface between protocol components in these systems is richer than in systems such as the x kernel to accommodate information transfer such as membership changes. These project have also worked ....

M. Hayden, "The Ensemble system," Tech. Rep. TR98-1662, Department of Computer Science, Cornell University, Jan 1998.

....messages issued by an ORB are intercepted and mapped on to calls of a group communication system. Well known examples of this approach are the Eternal [10,11] and AQuA [12] systems; Eternal uses the Totem group communication system [6] whereas AQuA uses the Ensemble group communication system [3]. The need to intercept calls makes these systems platform dependent. Both Eternal and AQuA make use of group communication for supporting object replication only (and not for other uses of group communication, such as collaborative applications) They do so by using the inclusive approach, and ....

M. Hayden, "The Ensemble system", PhD thesis, Dept. of Computer Science, Cornell University, 1998.

....We then resort to group management techniques to facilitate the (replicated) servers to exchange messages and cooperate with each other. The proposed implementation assumes two basic services: reliable multicast and group membership, which many of the existing group management middleware systems [3 9] can readily provide. By exploiting the fact that a server is internally replicated (over a synchronous network) we circumvent the unsolveable problem [10] of accurate failure detection in an asynchronous network (e.g. the Internet) which the servers would use to communicate with each other. The ....

....guarantee that the survivors are in agreement on the last multicast the old primary made before it crashed so that the transfer of the processing role from the old to the new primary remains correct. The specification and protocols for RM i can be found in [29] and for GM i the specification in [3 9] and protocols (that use NA2) in [30 32] Note that with passive replication, while every replica may receive the inputs, only the primary sends the server output to IS i and to other servers. Next, we describe how the (passively replicated) servers exchange episode messages. For simplicity we ....

[Article contains additional citation context not shown here]

M. Hayden, "The Ensemble system", PhD thesis, Dept. of Computer Science, Cornell University, 1998.

....failure) and uses a mechanism (such as round robin DNS) to determine which (functioning) server to direct an incoming message to. 5 processing role from the old to the new primary remains correct. The specification and protocols for RM i can be found in [15] and for GM i the specification in [3 9] and protocols (that use NA2) in [16 18] Note that with passive replication, while every replica may receive the inputs, only the primary sends the server output to IS and to other servers. Next, we describe how the (passively replicated) servers exchange episode messages. For simplicity we will ....

....3. We assume a reliable fifo multicast service (RM G ) and a group membership service (GM G ) within G. Using RM G , only) primaries would multicast episode messages which are received by every member of G. Note that RM G and GM G must be implemented with NA1 alone. Many groupbased systems e.g. [3 9], can provide these services just with NA1. However, they use failure suspectors to handle crashes which must be switched off and membership changes be effected by failure notification multicast (in G) by a S i 2 when primary crash is detected through GM i . Observe that S i 2 can reliably ....

M. Hayden, "The Ensemble system", PhD thesis, Dept. of Computer Science, Cornell University, 1998.

....at the network level, it might be necessary to switch between available network links depending on the quality of service currently on offer. Higher up the protocol stack a range of adaptations are also possible such as switching to a different transport protocol (or set of micro protocols [Hayden97]) changing the parameters of a particular protocol element, or introducing new protocol elements. It might also be useful to change the encoding of data, e.g. by introducing compression and de compression elements. Crucially, many aspects of adaptation are also carried out either in the ....

Hayden, M., "The Ensemble System", PhD Dissertation, Dept. of Computer Science, Cornell University, USA, 1997.

....the COOL platform has been extended as shown in figure 5. Currently COOL ORB supports two basic communications infrastructures: TCP IP and CHORUS IPC. In order to support higher degree of configurability, we have extended COOL with a third communications infrastructure based on Ensemble Maestro [10]. Ensemble, developed at Cornell University, is the successor to Horus [23] The software is written in Objective Caml (a dialect of ML) and provides a modular framework for constructing protocol stacks, specifically protocol stacks for group communication (point to point communication is treated ....

Hayden, M., "The Ensemble System", PhD Dissertation, Dept. of Computer Science, Cornell University, USA, 1997.

No context found.

M.G. Hayden. The ensemble system. PhD thesis, 1998.

No context found.

Hayden, M., "The Ensemble System", PhD Dissertation, Dept. of Computer Science, Cornell University, USA, 1997.

No context found.

Hayden, M., "The Ensemble System", PhD Dissertation, Dept. of Computer Science, Cornell University, USA, 1997.

No context found.

M. Hayden, "The Ensemble system", PhD thesis, Dept. of Computer Science, Cornell University, 1998.

No context found.

M.G. Hayden, "The Ensemble System," PhD thesis, Dept. of Computer Science, Cornell Univ., 1997.

No context found.

M. Hayden, "The Ensemble system," Ph.D. dissertation, Dept. of Computer Science, Cornell University, USA, 1997.

No context found.

Mark Hayden, "The Ensemble system," Technical Report TR98-1662, Department of Computer Science, Cornell University, Jan. 8, 1998.

No context found.

M. Hayden, "The Ensemble system", PhD thesis, Dept. of Computer Science, Cornell University, 1998.

No context found.

M. Hayden, "The Ensemble system", PhD thesis, Dept. of Computer Science, Cornell University, 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