F.Panzieri and S. K. Shrivastava,"Rajdoot: A Remote Procedure Call Mechanism Supporting Orphan Detection and Killing," IEEE Transactions on SoftwareEngineering,vol. SE-14, no. 1, pp. 30-37, January 1988.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....a remote call and the RMI implementation is either unable to make the call or makes the call but detects a failure before the call has returned. The server is not informed if a client node fails whilst a server is executing a request on its behalf. RMI does not implement orphan extermination [8, 16]. Hence, RMI has exactly once semantics in the absence of failures and at most once semantics in the presence of failures [18] 4 Minimal Integration between RTSJ and RMI (Level 0 Integration) Interfaces in Java provide a mechanism for defining a contract between a client and a server. They ....

F. Panzieri and S.K. Shrivastava, Rajdoot: A Remote Procedure Call Mechanism Supporting Orphan Detection and Killing, IEEE TOSE, 14(1), pp 30-37, 1988.

....a remote call and the RMI implementation is either unable to make the call or makes the call but detects a failure before the call has returned. The server is not informed if a client node fails whilst a server is executing a request on its behalf. RMI does not implement orphan extermination [8, 16]. Hence, RMI has exactly once semantics in the absence of failures and at most once semantics in the presence of failures [ 18] 4 Minimal Integration between RTSJ and RMI (Level 0 Integration) Interfaces in Java provide a mechanism for defining a contract between a client and a server. They ....

F. Panzieri and S.K. Shrivastava, Rajdoot: A Remote Procedure Call Mechanism Supporting Orphan Detection and Killing, IEEE TOSE, 14(1), pp 30-37, 1988.

....and the failure of the communication network has no effect [22] This means that remote procedure calls are made in an environment in which failures 7 are common. To enumerate the possible scenarios, the classification scheme for the semantics of remote procedure calls described in [37] and [42] is adopted. This is given by: At least once: The remote procedure has been executed one or more times if the invocation terminates normally. If it terminates abnormally, nothing can be said about the number of times the remote procedure has executed. It may have executed partially, zero, one, ....

....this RPC, nothing can be said about whether the server s counter was actually incremented or not. There are mechanisms to avoid these scenarios caused by network or processor failure for example, the client may retry the remote procedure call. However, this causes another problem orphans [42], which are the unwanted execution of remote procedures. These can also arise because of timing or synchronization problems. The choice of the transport protocol also plays a role in the fault tolerance of an application. Typical RPC implementations provide mechanisms for use of either UDP or TCP ....

F. Panzieri et al., "Rajdoot: A Remote Procedure Call Mechanism Supporting Orphan Detection and Killing", IEEE Transactions on Software Engineering, Jan. 1988.

....the same set of properties as existing RPC services. For example, the basic RPC service described in [18] corresponds to a service configured with RPCMain, Blocking, Acceptance(1) Collation with a function that returns the first response, ReliableCommunication, and UniqueExecution. Rajdoot [19] corresponds to the same set, plus BoundedTermination and TerminateOrphans. Among group RPC services, the one to many RPC described in [20] corresponds to the set RPCMain, Blocking, Acceptance(N) Collation with a function consisting of identity and comparison to detect inconsistencies at the ....

F. Panzieri and S. Shrivastava. Rajdoot: A remote procedure call mechanism supporting orphan detection and killing. IEEE Transactions on Software Engineering, SE-14(1):30--37, Jan 1988.

....with communication protocols [HP91] and runs standalone on a network of Sun workstations. The orientation towards supporting multiple paradigms distinguishes FT SR from other languages [Lis85, EFH82, LW85] language extensions [SCP91, CGR88, KU87] and language libraries [BSS91, Coo85, PS88, HW87] related to fault tolerance, which are typically oriented around a particular paradigm. Support for a single paradigm has been shown to be constraining in many situations [Bal91] and is particularly inappropriate for constructing systems, where different paradigms may be used at different ....

Fabio Panzieri and Santosh K. Shrivastava. Rajdoot: A remote procedure call mechanism supporting orphan detection and killing. IEEE Transactions on Software Engineering, 14(1):30--37, January 1988.

....also interfere with new calls issued by a recovered client. Options for dealing with orphans include interference avoidance, where the orphans finish their computation before the recovered client is allowed to issue new requests, and orphan termination, where orphans are terminated upon detection [18, 20]. ffl Communication semantics specify properties about the communication between the client and server. Here, we concentrate on reliable communication, which can be implemented by message acknowledgements and retransmissions. Of course, if the reliability guarantees provided by the underlying ....

....and the client thread returns within a bounded, specified time. If the server has not responded by the deadline, the call returns with an indication of failure. We point out that our classification of failure semantics subsumes more traditional distinctions, which can be summarized as follows [18]. At least once guarantees that if the invocation terminates normally, the remote procedure has been executed one or more times, and if it terminates abnormally, no conclusion is possible [21] Exactly once guarantees that if the invocation terminates normally, the remote procedure has been ....

[Article contains additional citation context not shown here]

F. Panzieri and S. K. Shrivastava. Rajdoot: A remote procedure call mechanism supporting orphan detection and killing. IEEE Transactions on Software Engineering, SE14 (1):30--37, Jan 1988.

....with new calls issued by a recovered client. Options for dealing with orphans include interference avoidance, where the orphans finish their computation before the recovered client is allowed to issue new requests, and orphan termination, where orphans are terminated upon detection [Shr83, PS88] ffl Communication semantics specify properties about the communication between the client and server. Here, we concentrate on reliable communication, which can be implemented by message acknowledgements and retransmissions. Of course, if the reliability guarantees provided by the underlying ....

....and the client thread returns within a bounded, specified time. If the server has not responded by the deadline, the call returns with an indication of failure. We point out that our classification of failure semantics subsumes more traditional distinctions, which can be summarized as follows [PS88] At least once guarantees that if the invocation terminates normally, the remote procedure has been executed one or more times, and if it terminates abnormally, no conclusion is possible [Spe82] Exactly once guarantees that if the invocation terminates normally, the remote procedure has been ....

F. Panzieri and S. K. Shrivastava. Rajdoot: A remote procedure call mechanism supporting orphan detection and killing. IEEETransactions on Software Engineering, SE-14(1):30--37, Jan 1988.

....invalid or contradictory actions. Providing integrity for distributed threads is similar to providing integrity for other distributed computation abstractions. For example, orphan detection and elimination has been studied in the context of nested transactions [17, 18, 31] remote procedure calls [36, 39, 45], and message logging protocols [2] A specific TMAR protocol was developed for Alpha, based on assumptions about the expected operating environment [37] This included assumptions about the workload, the user and application requirements, the system fault model, and the resource base and its ....

F. Panzieri and S. K. Shrivastava. Rajdoot: A Remote Procedure Call Mechanism Supporting Orphan Detection and Killing. IEEE Transactions on Software Engineering, 14(1):30--37, January 1988.

....the probability of certain objects becoming undetectable garbage, our default mechanism can be extended by the addition of garbage collection driven by the objects themselves. In other distributed systems [Parrington95] Rosenburg92] ANSA91] OMG93] this is typically referred to as orphan detection [Panzieri88], although we term it object driven garbage detection. If lost references are detected (e.g. through the use of periodic keep alive messages) then the objects automatically reduce their reference count for the appropriate references. Our system provides reference driven garbage collection for all ....

F. Panzieri and S. K. Shrivastava, "Rajdoot: A Remote Procedure Call Mechanism Supporting Orphan Detection and Killing," IEEE Transactions on Software Engineering, Vol. 14, No. 1, pp. 30-37, January 1988.

....about the number of times the remote procedure has been executed, both in the case when the invocation terminates normally and when it terminates abnormally; the latter occurs, for example, when the server crashes prior to completing execution of the remote procedure. The most common are [Nel81, PS88] ffl At Least Once: The remote procedure has been executed one or more times if the invocation terminates normally. If it terminates abnormally, no conclusion is possible, i.e. it may have been executed one time or multiple times, may have been partially through an execution, or may not have ....

....client reissues an RPC after failing to receive a response from the server within a specified period of time even though the server is still up and running. Orphans can cause consistency problems by interfering with subsequent calls to the server or with other computations going on in the system [PS88] In addition, orphans waste system resources. The possible creation of orphans is a fundamental problem for any RPC mechanism, so techniques must be provided to detect such computations and eliminate them. In addition, if the semantics of the mechanism require that there be no side effects in ....

[Article contains additional citation context not shown here]

F. Panzieri and S. K. Shrivastava. Rajdoot: A remote procedure call mechanism supporting orphan detection and killing. IEEE Transactions on Software Engineering, SE-14(1):30--37, Jan 1988.

....copy of objects to take care of cases when operations are aborted or if there a failure in the system. Replication of objects are also allowed in the architecture. An interesting feature of Arjuna is that server processes are created on the fly. This is achieved by the special Rajdoot RPC system [PS88] It is an enhanced RPC, providing support for object group management such as activation, invocation, and replication of objects. It even detects orphan processes which can occur during failures. We observe that Arjuna has features similar to that of an OODBMS. All the objects created have to ....

F. Panzieri and S. K. Shrivastava. Rajdoot: A remote procedure call mechanism supporting orphan detection and killing. IEEE Transactions on Software Engineering, 14(1), January 1988.

....can also be used to implement different semantic variants of the same property. For example, with RPC, there may be multiple micro protocols implementing different policies for how the request is handled if the server fails, such as exactly once, at least once, or at most once semantics [34]. A system is then configured based on the particular properties needed for the given application. This micro protocol approach is realized by augmenting the x kernel s standard hierarchical object composition model with the ability to internally structure protocol objects. The result is a ....

F. Panzieri and S. K. Shrivastava. Rajdoot: A remote procedure call mechanism supporting orphan detection and killing. IEEE Trans. on Softw. Eng., SE-14(1):30--37, Jan 1988.

....6: Tree of degree m = 4 and of height p = 1: This algorithm has been implemented in C. For the Fortran analogue see Appendix A. The interface between client and server is defined in the file scalar.sc which does not differ for C and Fortran applications. 3 In contrast, the RPC system Radjoot [20] provides a sophisticated orphan treatment mechanism supporting crash recovery and multiple clients INTERFACE Scalar; scalar.sc: Interface Definition Module for distributed scalar product ) CONST LDQ = 1000; TYPE Vector = PACKED DYNAMIC ARRAY [LDQ] OF LONGREAL; PROCEDURE scalar (IN myid, ....

F. Panzieri and S. K. Shrivastava. Rajdoot: A remote procedure call mechanism supporting orphan detection and killing. IEEE Transactions on Software Engineering, 14:30--37, 1988.

....can also be used to implement different semantic variants of the same property. For example, with RPC, there may be several micro protocols implementing distinct policies for how a request is handled if the server fails, such as exactly once, at least once, or at most once semantics [PS88] A system is then configured based on the particular properties needed for the given application. This micro protocol approach is realized by augmenting the x kernel s standard hierarchical object composition model with the ability to internally structure protocol objects. The result is a ....

....to send a client request to a group of servers, but no server responses are allowed and there are no ordering guarantees. If responses are required, the servers use separate RPC calls that are manually collated by the client. Other RPC services are provided by [SB90, BALL90, Cou81, BN84, ATK91, PS88] 2.2.3 Group RPC Systems Sun RPC has been used as a basis for a fault tolerant group RPC service in [YJT88] In this system, the servers are organized as a linearly ordered group, with the first process in the group the primary receiving all RPC calls and sending all replies. The primary ....

F. Panzieri and S. K. Shrivastava. Rajdoot: A remote procedure call mechanism supporting orphan detection and killing. IEEE Trans. on Software Engineering, SE-14(1):30--37, January 1988. 130

....also interfere with new calls issued by a recovered client. Options for dealing with orphans include interference avoidance, where the orphans finish their computation before the recovered client is allowed to issue new requests, and orphan termination, where orphans are terminated upon detection [PS88, Shr83] Communication semantics specify properties about the communication between the client and server. Here, we concentrate on reliable communication, which can be implemented by message acknowledgments and retransmissions. Of course, if the reliability guarantees provided by the underlying ....

....terminates and the client thread returns within a bounded, specified time. If the server has not responded by the deadline, the call returns with an indication of failure. Note that our classification of failure semantics subsumes more traditional distinctions, which can be summarized as follows [PS88] At least once guarantees that if the invocation terminates normally, the remote procedure has been executed one or more times, and if it terminates abnormally, no conclusion is possible [Spe82] Exactly once guarantees that if the invocation terminates normally, the remote procedure has been ....

[Article contains additional citation context not shown here]

F. Panzieri and S. Shrivastava. Rajdoot: A remote procedure call mechanism supporting orphan detection and killing. IEEE Transactions on Software Engineering, SE-14(1):30--37, Jan 1988.

....processors with fail silent semantics, although the approach generalizes to other failure models as well. The orientation towards a multi paradigm approach distinguishes FT SR from other languages [Lis85 , EFH82, LW85] language extensions [SCP91, CGR86, KU87] and language libraries [BSS91, Coo85, PS88, HW87] related to fault tolerance, which are typically oriented around a particular paradigm. Support for a single paradigm has been shown to be constraining in many situations [Bal91 ] and is particularly inappropriate for constructing systems, where different paradigms may be used at different ....

....runtime the mechanics of checkpointing. As another example, languages like FTCC [CGR86] provide the programmer with the ability to create and manage object groups and, as a result, the ability to build stable objects, but lack support for building restartable objects. Systems like Rajdoot [PS88] provide some support for restartable actions in the form of orphan detection and termination, allowing the programmer to re issue invocations to servers. These systems, however, make no provision for replication. We feel that FT SR provides an appropriate level of abstractions for a systems ....

F. Panzieri and S. K. Shrivastava. Rajdoot: A remote procedure call mechanism supporting orphan detection and killing. IEEE Transactions on Software Engineering, 14(1):30--37, January 1988.

....the statement in line 2 (figure 2) Then explicit steps must be taken to break the orphaned binding: the server node must detect this crash and break the binding. The functionality of a connection manager process can be embellished to include periodic checking of connections with client nodes [22]. Every active object is associated with some object server; this server uses get request and send reply to service operation invocations. One server may manage several objects (i.e. the correlation between server processes and objects may not be one to one) Any internal details of the server ....

....are the only visible entities; the client and server processes that do the actual work are hidden. In Arjuna, server processes are created dynamically as RPCs are made to objects; these servers are created using the facilities provided by the underlying RPC subsystem, Rajdoot, also built by us [22]. The current implementation of Arjuna makes use of the Unix file system for long term storage of objects, with a class ObjectStore providing an object oriented interface to the file system. The design and implementation of the Arjuna object store is discussed elsewhere [11] along with the ....

Panzieri, F. and S.K. Shrivastava, "Rajdoot: a remote procedure call mechanism supporting orphan detection and killing", IEEE Trans. on Software Eng. 14, 1, pp. 30-37, January 1988.

....requires that a server node must have a mechanism for breaking the binding if it suspects the crash of a client. This mechanism will also cope with a partition that prevents any communication between a client and a server. The Arjuna RPC level facilities for the detection and killing of orphans [Panzieri88] is responsible for such a cleanup, ensuring at the same time that an orphaned server (a server with no bindings) is terminated. The principal classes which make up the class hierarchy of Arjuna Atomic Action module are depicted below. StateManager Naming, persistence and recovery control ....

Panzieri, F. and S.K. Shrivastava, "Rajdoot: a Remote Procedure Call Mechanism Supporting Orphan Detection and Killing"' IEEE Trans. on Software Eng. 14, 1, pp. 30-37, January 1988.

No context found.

F.Panzieri and S. K. Shrivastava,"Rajdoot: A Remote Procedure Call Mechanism Supporting Orphan Detection and Killing," IEEE Transactions on SoftwareEngineering,vol. SE-14, no. 1, pp. 30-37, January 1988.

No context found.

F. Panzieri, S. Shrivastava, Rajdoot: A Remote Procedure Call Mechanism Supporting Orphan Detection and Killing, IEEE Transaction on Sofware Engineering, vol. SE-14, n 1, Jan 1988, pp. 30-37

No context found.

Panzieri88 F. Panzieri, and S.K. Shrivastava, "Rajdoot: A Remote Procedure Call Mechanism Supporting Orphan Detection and Killing," IEEE Transactions on Software Engineering, vol. SE-14, no. 1, pp. 30-37, January 1988.

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