R. Guerraoui, P. Felber, B. Garbinato and K. Mazouni, "System support for object groups", Proceedings of the ACM Conference on Object Oriented Programming Systems, Languages and Applications, OOPSLA 98.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....groups in a transparent way, as if they were a single, and non replicated object. Objects forming a group cooperate in order to provide a reliable and high available service to their clients. This cooperation is established through the facilities offered by a group communication service (GCS) [4, 9], that enables the creation of dynamic groups of objects that communicate through reliable multicast primitives. Objects forming a group are kept informed about the current membership of the group itself, that may vary at run time due to accidental events such as failures and repairs, or to ....

R. Guerraoui, P. Felber, B. Garbinato, and K. Mazouni. System Support for Objects Groups. In Proceedings of the 13th ACM Conference on Object Oriented Programming Systems, Languages and Applications (OOPSLA'98), volume 33(10) of Sigplan Notice, Vancouver, British Columbia, Canada, Oct. 1998.

....uses inter replica communication. 6. Conclusions Replication schemes by their nature impose a significant performance overhead. Measurements reported for several existing approaches to replication indicate performance overheads ranging from three to ten times that of nonreplicated systems [1 3, 25]. Until recently, only single thread applications were replicated, since multithreading does not easily conform to the Errors were injected (one per experiment) in each bit of each byte in the portion of the text segment corresponding to the selected functions. Table 5. Text injections into ....

R. Guerraoui et al. System support for object groups. In ACM Conf. on Object-Oriented Programming Systems, Languages and Applications, 1998.

....These characteristics are investigated with fault injection. 1 Introduction Software based active replication is a well known technique for providing fault tolerance using space redundancy and fault masking. Typically, replication is expensive in terms of performance overhead [17] 23] 21] [12]. Multithreading can help improve performance by exploiting concurrency in thread execution; however, since the thread process scheduling performed by operating systems such as UNIX is asynchronous with replica execution, multithreaded replicas can exhibit nondeterministic behavior. Solutions to ....

....6. Conclusions Replication schemes by their nature impose a significant performance overhead. For example, measurements reported for several existing approaches to replication indicate performance overheads ranging from three to ten times that of nonreplicated systems [17] 23] 21] [12]. Until recently, only single thread applications were replicated, since multithreading does not easily conform to the state machine approach [26] widely used in software replication. However, if the replicas are multithreaded, then performance overhead due to replication can be lessened. A ....

R. Guerraoui, P. Felber, B. Garbinato, and K. R. Mazouni. System support for object groups. In ACM Conference on Object-Oriented Programming Systems, Languages and Applications, 1998.

....the main technology for fault tolerance is very similar to that in process groups [13, 14] There are other fault tolerant middleware systems based on message logging [23] The discussion here only focuses on the object group approaches. Three main efforts in CORBA object groups are introduced [15 17]. We follow with a discussion of the Coign project, which is built on the COM middleware system [18, 19] An API instrumentation technique is used in the Coign project. We conclude with a discussion of a COM system on IP multicast transport built by P. Emerald Chung et al. in Bell Lab [20] 2.2.1 ....

....Dynamic Invocation Interface (DII) and the Dynamic Skeleton Interface (DSI) to their associated OGS objects, which then coordinate with each other to perform the operation on the replicas of the object and to return the results appropriately. Namely, the object group mechanisms are put into OGS [15]. The advantage of this approach is that it is fully compliant with the CORBA standard and requires no proprietary mechanisms. However, the object group is now visible to the application objects since the application objects must be aware of the existence of the OGS objects in order to utilize ....

Guerraoui, R., et al. System support for object groups. in ACM Conference on ObjectOriented Programming Systems, Languages and Applications. 1998.

.... with schedulability analysis is [5] The Delta 4 system proposes a different approach based on semi active replication following a leader follower model [2] There are some works providing CORBA interfaces for object replication like the Eternal system [14] and the object group service (OGS) [9]. OGS guarantees determinism by executing request sequentially in the total order they have been delivered. The Eternal system provides support for replicated multithreaded CORBA servers. In their approach, although replicated servers are multithreaded (using any of the CORBA multithreading ....

R. Guerraoui, P. Felber, B. Garbinato, and K. Mazouni. System Support for Object Groups. In Proc. of ACM Conf. Object Oriented Programming Systems, Languages and Applications, 1998.

....provoked by multithreading and asynchronous events. In the advent of a rollback this log is used to replay a former consistent state of the replica. There are some systems providing CORBA interfaces for object replication like the Eternal system [NMMS99] the object group service (OGS) GFGM98] AQuA [CRS 98] Electra [LM97] and FTS [FH01] We will focus our comparison on the Eternal system since it is the only one (to the best of our knowledge) that provides support for replicated CORBA servers in the presence of non determinism such as multithreading. In their approach, although ....

R. Guerraoui, P. Felber, B. Garbinato, and K. R. Mazouni. System support for object groups. In ACM Conference on Object-Oriented Programming Systems, Languages and Applications (OOPSLA98), October 1998.

....6 provides performance overhead reported for several existing approaches to replication. Table 6 clearly indicates that most replication strategies impose a significant performance overhead. The exception is Eternal with a reported overhead of 3 . Most of the overhead measurements [20] 22] [13] use an echo server (or effectively the same functionality) An echo server accepts messages incoming from clients and echoes them back to the clients. The only computation performed by the server is moving the data received from the client to an internal buffer and sending the data stored in ....

....3 Transparent replication to both CORBA applications and ORB. 10 overhead on Solaris. LSA framework Active replicated echo server with majority voting. Ensemble as communication layer. 200 30 overhead for multithreaded Apache server with a CGI program creating a 1 KB HTML page. Bast [13] Smalltalk method invocation. 900 Replicas immediately send a reply to the client. OGS [13] CORBA method invocation. Solaris 2.5 on SPARC. 900 Arjuna [24] System call invocation. Solaris 2.3 on SPARC. 600 (for write system call) Object oriented framework for fault tolerant distributed ....

[Article contains additional citation context not shown here]

R. Guerraoui, P. Felber, B. Garbinato, and K. R. Mazouni. System support for object groups. In ACM Conference on Object-Oriented Programming Systems, Languages and Applications, 1998.

.... 10 20 30 40 50 60 speedup cpus ASP GMI mpiJava 0 10 20 30 40 50 60 0 10 20 30 40 50 60 cpus LEQ GMI mpiJava 0 10 20 30 40 50 60 0 10 20 30 40 50 60 cpus QR GMI mpiJava Figure 12: Speedups of benchmarks serves the purposes of fault tolerance and location transparency [6, 13, 27]. Efficiency is less important in this context. The Common Object Request Broker Architecture (CORBA) 18] is similar in functionality to RMI. CORBA uses a special interface definition language (IDL) to define remote interfaces. As with RMI, these interfaces are used to generate stubs and ....

R. Guerraoui, P. Felber, B. Garbinato, and K. Mazouni. System Support for Object Groups. In Proc. OOPSLA'98, pages 244--258, Vancouver, B.C., Oct. 1998.

....requires the participating sites to coordinate their activities by exchanging messages. In practice, this can have a significant impact on the overall behavior of the protocol. On the one side, CPU cycles are lost in dealing with the messages (flattening, sending, receiving, and unflattening) [10]. On the other side, the network bandwidth might be exceeded, resulting in additional delays. The message overhead affects not only the scalability but also the availability [16] In this section we study the message overhead of the different protocols. 6.1 Message cost To get an accurate ....

R. Guerraoui, P. Felber, B. Garbinato, and K. R. Mazouni. System support for object groups. In ACM OOPSLA'98, Oct. 1998.

....and application speed, if implemented well. Our RepMI mechanism provides object replication for the purpose of efficiently accessing objects from parallel applications. In the field of distributed systems, object replication mainly serves the purposes of fault tolerance and location transparency [7, 18, 33], efficiency is less important in this context. An alternative approach to improving the access time of shared objects is object caching as implemented in systems like Hyperion [3] and Jackal [32] However, neither replication nor caching of objects provides the flexibility and expressiveness of ....

R. Guerraoui, P. Felber, B. Garbinato, and K. Mazouni. System Support for Object Groups. In Proc. OOPSLA'98, pages 244--258, Vancouver, B.C., Oct. 1998.

....this section, we want to argue that the virtual synchrony model, which is originally a process group model, doesn t fit as nicely with the object model, at least when used to replicate entities for faulttolerance. We ll first take a somewhat indirect route and present the problems the authors of Guerraoui et. al 1998 faced when trying to build an object group system on top of an existing process group system (namely, Isis) which they attribute to the mismatch between process groups and object groups. Thus, their conclusion is that one should build an object group system from scratch, and not on top of a ....

.... entity, this other entity will receive the same message more than once; this would even happen with an object group within one and the same process, so that this cannot be a consequence of the mismatch between the rigidity of the process group model and the flexible nature of object interactions (Guerraoui et al. 1998, p. 1) The same applies to mixed scheduling, which is rather a consequence of the mismatch between two different threading models, where only one of these is really the appropriate one for virtually synchronous processing (namely, the Isis model) A generic object group system clearly has to ....

[Article contains additional citation context not shown here]

Guerraoui, R., P. Felber, B. Garbinato, and K. Mazouni. 1998. "System Support for Object Groups." Proceedings OOPSLA 1998. ACM.

....is located physically closer. Replica group Client Replica consistency protocol Figure 1: Replica group interactions. Replication protocols can be categorised by the type of consistency they maintain between the replicas: strong consistency, where all replicas have identical states (e.g. [1,2]) and weak consistency, where replica states are allowed to diverge (e.g. 3] Within these categories there are basically two classes of protocols: active replication: more than one copy of an object is activated on distinct nodes and all copies perform processing [1] Active replication is ....

....active replication whereas the other passive, and each domain may tolerate different types of member failures. Externally such protocols are not visible to users, so any implementation can be used. Consistency protocols will normally require support from some form of object group service (e.g. [2]) Cons istency protocol Consistency domain Replica Figure 4: Replication and cacheing model. To maintain levels of (in )consistency between the object copies, the domains must co operate to propagate updates; the protocols used to connect these domains and maintain their levels of ....

R. Guerraoui, P. Felber, B. Garbinato and K. Mazouni, "System support for object groups", Proceedings of the ACM Conference on Object Oriented Programming Systems, Languages and Applications, OOPSLA 98.

....to be also not portable. Furthermore, most existing group toolkits are not adapted to a CORBA environment: they are designed for process groups instead of object groups, and they do not provide adequate primitives for group to group communication, making it difficult to support client replication [GFGM98] An interesting variant of the integration approach, which has not been explored by existing systems, would consist in providing a second object adapter in addition to the Basic Object Adapter. We would then have the basic adapter for standard objects and a dedicated adapter a Group Object ....

....in the context of OGS and the underlying mechanisms used to provide it. Transparency in OGS. Transparency is the property of a system to be invisible, i.e. the degree to which application programs are unaware of the system. Transparency appears at different levels in the OGS architecture [GFGM98] and may be classified as follows: Plurality transparency: the application objects have the illusion that they deal with singleton objects, although they interact with object groups. Behavior transparency: the application objects are not aware of the replication policy and the protocols ....

[Article contains additional citation context not shown here]

R. Guerraoui, P. Felber, B. Garbinato, and K. Mazouni. System support for object groups. In Proceedings of the ACM Conference on Object Oriented Programming Systems, Languages and Applications (OOPSLA'98), 1998.

....making development of portable applications almost impossible. In addition, most of these systems suffer from a number of problems related to the mismatch between the underlying process group model and distributed objects, such as group proliferation, request duplication, or mixed scheduling [7]. To alleviate these limitations, the Object Management Group (OMG) has initiated a standardization process for fault tolerant CORBA. Several individuals involved in previous fault tolerant CORBA implementations (namely [15, 13, 4] have contributed to this specification, and therefore the ....

R. Guerraoui, P. Felber, B. Garbinato, and K. Mazouni. System support for object groups. In Proceedings of the ACM Conference on Object Oriented Programming Systems, Languages and Applications (OOPSLA'98), Oct. 1998.

....database tier makes it complicated, 36 if not impossible, to use standard, off the shelf database systems. Other issues such as how to coordinate many to many communication (between a group of servers and a group of databases) are not obvious and might induce considerable performance penalties [23]. In our case, we rely on off the shelf clustering technology to provide quick recovery for databases. However, we do not assume that each database can be viewed as a failure free entity: we still need to handle the case of transaction aborts because of a database crash. Our primary backup ....

R. Guerraoui, P. Felber, B. Garbinato, and K. Mazouni, "System support for object groups," in Proceedings of ACM Conference on Object-Oriented Programming Systems, Languages and Applications (OOPSLA'98), pp. 244--258, October 1997.

....Service follows the OMG service specification design guidelines and complies with many of the requirements of the OMG Request For Proposal on CORBA Fault Tolerance. This paper does not detail OGS design or implementation. More information on OGS, including performance figures, can be found in [FGG96, FGS98, GFGM98], or at http: lsewww.epfl.ch OGS where binary code versions of OGS (for Orbix and Visibroker) are available. We rather focus in this paper on the use of OGS in distributed programming. More precisely, we give an overview of OGS configuration and programming models and we illustrate its use on ....

.... as a C dynamic library (OGSl) to be linked with C applications, or as a set of Java classes usable from Java applications (see Figure 2) This execution model is more efficient since interprocess communications are more costly than invocations between objects located in the same process [GFGM98]. Nevertheless, it enforces the code of the application to be written with the same programming language as the library and to support multi threading. The daemon execution model, with two separate processes, has the advantage of decoupling the service from the application, enabling several ....

R. Guerraoui, P. Felber, B. Garbinato, and K. Mazouni. System support for object groups. In Proceedings of the ACM Conference on Object Oriented Programming Systems, Languages and Applications (OOPSLA'98), 1998.

No context found.

R. Guerraoui, P. Felber, B. Garbinato and K. Mazouni, "System support for object groups", Proceedings of the ACM Conference on Object Oriented Programming Systems, Languages and Applications, OOPSLA 98.

No context found.

R. Guerraoui, P. Felber, B. Garbinato and K. Mazouni, "System support for object groups", Proceedings of the ACM Conference on Object Oriented Programming Systems, Languages and Applications, OOPSLA 98.

No context found.

R. Guerraoui, P. Felber, B. Garbinato and K. Mazouni, "System support for object groups", Proceedings of the ACM Conference on Object Oriented Programming Systems, Languages and Applications, OOPSLA 98.

No context found.

R. Guerraoui, P. Felber, B. Garbinato and K. Mazouni, "System support for object groups", Proceedings of the ACM Conference on Object Oriented Programming Systems, Languages and Applications, OOPSLA 98.

No context found.

R. Guerraoui, P. Felber, B. Garbinato and K. Mazouni, "System support for object groups", Proceedings of the ACM Conference on Object Oriented Programming Systems, Languages and Applications, OOPSLA 98.

No context found.

R. Guerraoui, P. Felber, B. Garbinato, and K. R. Mazouni. System support for object groups. In OOPSLA, 1998.

No context found.

R. Guerraoui, P. Felber, B. Garbinato, and K. Mazouni. System Support for Object Groups. In Proc. OOPSLA'98, pages 244-258, Vancouver, B.C., Oct. 1998.

No context found.

R. Guerraoui, P. Felber, B. Garbinato and K. Mazouni, "System support for object groups", Proceedings of the ACM Conference on Object Oriented Programming Systems, Languages and Applications, OOPSLA 98.

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