L. Rodrigues, K. Guo, P. Verissimo, and K. Birman. A dynamic light-weight group service. Journal on Parallel and Distributed Computing, (60):1449--1479, 2000.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....in order to implement Virtual Synchrony: end points send synchronization messages to their designated leaders, which in turn exchange only the cumulative information among themselves. Also, the number of messages exchanged to synchronize multiple groups can be reduced, as suggested in [11, 39], by aggregating information pertaining to multiple groups into a single message. Another optimization addresses the use of stable storage. Recall that in Section 4 we assumed that end points keep their running states on stable storage, and therefore, recover with their state intact. However, our ....

L. Rodrigues, K. Guo, A. Sargento, R. van Renesse, B. Glade, P. Verissimo, and K. Birman, A dynamic light-weight group service, in 15th IEEE International Symposium on Reliable Distributed Systems (SRDS), Oct. 1996.

....for message dissemination in order to implement VS: end points send synchronization messages to their designated leaders, which in turn exchange only the cumulative information among themselves. Also, the number of messages exchanged to synchronize multiple groups can be reduced, as suggested in [11, 39], by aggregating information pertaining to multiple groups into a single message. Another optimization addresses the use of stable storage. Recall that in section 4 we assumed that end points keep their running states on stable storage, and therefore, recover with their state intact. However, our ....

L. Rodrigues, K. Guo, A. Sargento, R. van Renesse, B. Glade, P. Verissimo, and K. Birman, A dynamic light-weight group service, in Proceedings of the 15th IEEE International Symposium on Reliable Distributed Systems (SRDS '96), 1996.

....systems do not tolerate uncertainty about the participants. Therefore, a race condition between a join and a concurrent failure can cause such light weight group services (e.g. 11] to violate consistency. Those light weight group services that do preserve consist membership semantics (e.g. [19]) do incur extra delivery latencies whenever joins and leaves occur. Other work on group membership in synchronous and real time systems, e.g. 15, 14] has focused on membership maintenance in a static, fairly small, group of processes, where processes are subject to failures but no new ....

L. Rodrigues, K. Guo, A. Sargento, R. van Renesse, B. Glade, P. Verissimo, and K. Birman. A dynamic light-weight group service. In 15th IEEE Int. Symp. on Reliable Dist. Sys. (SRDS), pp. 23--25, Oct 1996.

....range of existing local resource managers, to deploy its beneficial capabilities without requiring wholesale replacement of existing infrastructure. Results from GRAM testbeds have shown the feasibility of mapping TSLAs onto a range of local job schedulers, as well as simple time sharing computers [15, 5, 32]. The GARA prototype has shown how RSLAs and BSLAs can be mapped down to contemporary network QoS systems [20, 21, 33] Following this model, SNAP manager services represent adaptation points between the SNAP protocol domain and local RM mechanisms. 6.3 Monitoring A fundamental function for RM ....

L. Rodrigues, K. Guo, P. Verissimo, and K. Birman. A dynamic light-weight group service. Journal on Parallel and Distributed Computing, (60):1449--1479, 2000.

....in order to implement Virtual Synchrony: end points send synchronization messages to their designated leaders, which in turn exchange only the cumulative information among themselves. Also, the number of messages exchanged to synchronize multiple groups can be reduced, as suggested in [18, 81], by aggregating information pertaining to multiple groups into a single message. Another optimization addresses the use of stable storage. Recall that in Chapter 4 we assumed that end points keep their running states on stable storage, and therefore, recover with their state intact. However, our ....

Luis Rodrigues, Katherine Guo, Antonio Sargento, Robbert van Renesse, Brad Glade, Paulo Verissimo, and Ken Birman. A dynamic light-weight group service. In 15th IEEE International Symposium on Reliable Distributed Systems (SRDS), pages 23--25, October 1996.

....condition between a join and a concurrent failure can cause such light weight group services (e.g. 23, 12, 15] to violate the semantics of the underlying heavy weight membership services. Those light weight group services that do preserve the underlying heavy weight membership semantics (e.g. [24]) do incur extra delivery latencies whenever joins and leaves occur. Other work on group membership in synchronous and real time systems, e.g. 19, 18] has focused on membership maintenance in a static, fairly small, group of processes, where processes are subject to failures but no new ....

L. Rodrigues, K. Guo, A. Sargento, R. van Renesse, B. Glade, P. Verissimo, and K. Birman. A dynamic light-weight group service. In 15th IEEE International Symposium on Reliable Distributed Systems (SRDS), pages 23-25, Oct. 1996.

....overhead induced by OVS is very small. We describe the implementation and present our performance measurements in Section 4. The method we used to implement OVS in Transis can be applied in any group communication system that allows multiple processes to send messages concurrently (e.g. Horus [14, 23], Ensemble [15] Relacs [8] and those of [18, 24] We believe that it is possible to design a similar method for supporting OVS in tokenbased systems (e.g. Totem [4] but we have not explored this possibility. 1.1. Evaluating OVS Optimism does not provide additional capabilities for the ....

....any group communication system in which processes communicate over fifo channels, can send messages spontaneously without waiting for a token, and send synchronization (or flush) messages to each other to denote the end of the message stream in a terminating view. Examples of such systems include [14, 15, 8, 18, 24, 23]. 4.2. Performance Measurements In this section we describe the measured performance of OVS implemented on top of Transis. The tests were run on three 333 Megahertz Sun UltraSparc 5 10s, running SunOS version 5.6. The machines were connected via 10MBit sec Ethernet. The machines were not being ....

[Article contains additional citation context not shown here]

L. Rodrigues, K. Guo, A. Sargento, R. van Renesse, B. Glade, P. Verissimo, and K. Birman. A dynamic light-weight group service. In 15th IEEE Intl. Symp. on Reliable Dist. Systems (SRDS), pages 23--25, Oct. 1996.

....needed by such a system. Ours is the only membership algorithm that we are aware of that never delivers views which it knows to be obsolete. As explained in Section 2.1, this feature is very important in WANs. 8. 3 Light weight group membership services Light weight group membership algorithms [25, 5, 24, 7, 43, 31, 46, 16] employ a client server approach to both virtual synchrony and membership maintenance. In these algorithms, there are two levels of membership, heavy weight and light weight. The servers are part of the heavy weight membership, and they use virtually synchronous communication among them. The ....

L. Rodrigues, K. Guo, A. Sargento, R. van Renesse, B. Glade, P. Verissimo, and K. Birman. A dynamic light-weight group service. In 15th IEEE International Symposium on Reliable Distributed Systems (SRDS), pages 23--25, Oct. 1996. also Cornell University Technical Report, TR96-1611, August, 1996.

....first of these tasks. This is a critical point in the overall design as, if no performance advantages can be obtained by mapping several Lwgs in a single Hwg , the implementation of mapping and switching strategies becomes pointless (mapping and switching heuristics are discussed in another report [12]) 4 Protocols This section describes the protocols that implement the Light Weight Group Service. These protocols perform the several tasks required to offer virtual synchrony: join a group, leave a group, and multicast messages in a group. Additionally, the switching protocol is also ....

L. Rodrigues, K. Guo, A. Sargento, R. van Renesse, B. Grade, P. Ver'issimo, and K. Birman. A dynamic light weight group service. Technical report, INESC/Cornell Univ., March 1996.

....but it assumes that modules are implemented using such a construct. More recently, we are developing on the Appia framework[25] One of the early prototypes of NAVTECH used the Horus run time, combined with our own protocols[10] A subset of NAVTECH protocols, namely the Light Weight Group Service[36], have been implemented on the Ensemble system. Recent works address scalability through probabilistic algorithmic approaches, trading off determinism for scalability[21, 15] 3 System Model Like in any other architectural framework, we paid special attention to the system model, and its main ....

....enough to allow the protocol designer to exploit its hierarchical nature to improve the performance or reliable communication protocols. There are a few published protocols exploiting the innovative attributes of NAVTECH , namely topology and group model. For example, a lightweight group service[36] and a reliable remote group access protocol[37] both taking advantage from the sender member and site participant duality of the group model, and benefit from the topology to have credible efficiency. Another protocol is a clock synchronization protocol based on an hierarchical composition of a ....

L. Rodrigues, K. Guo, P. Verssimo, and K. Birman. A dynamic light-weight group service. Journal of Parallel and Distributed Computing, 60(12):1449--1479, December 2000.

....Because of the timely control information provided by TFDS, it is possible to reach an agreement before making a decision about the way that reconfiguration should be done. The use of light weight groups (that is the focus of this paper) is also something that has been used by other researchers [23, 19, 25]. However, they were used in a different perspective not related to real time and the handling of timing failures. Having group members that exclude themselves from the group upon failure detection is also used in [1] However, they do not use a hierarchical structure for groups and the excluding ....

L. Rodrigues, K. Guo, A. Sargento, R. van Renesse,B. Glade, P. Verssimo, and K. Birman. A dynamic light-weight group service. In Proceedings of the 15th IEEE Symposium on Reliable Distributed Systems, Niagara on the Lake, Canada, Oct. 1996. (also available as Cornell University Technical Report, TR96-1611).

....a single HWG , the implementation of mapping and switching strategies becomes pointless. In this paper we describe the protocol for switching between groups but not the decision process that leads to the invocation of this protocol (mapping and switching heuristics are discussed in another report [12]) 4 Protocols This section describes the protocols that implement the Light Weight Group Service. These protocols perform sev eral tasks required to offer virtual synchrony: join a group, leave a group, and multicast messages in a group. For self containment, the switching protocol is also ....

....procedure consists of two main steps, as illustrated in Figure 4. In the first step, a map is established between the LWG and some HWG (l. 401) To minimize accesses to the name service, the joining process proposes a mapping based on its own local HWGs according to the mapping heuristics [12]. Then, in a single access to the name service it commits this mapping or, in the case where the LWG is already mapped onto some other HWG, obtains the existing mapping (l. 404) Additionally, if the process is not a member of the selected HWG , it joins the HWG before executing the second step ....

[Article contains additional citation context not shown here]

L. Rodrigues, K. Guo, A. Sargento, R. van Renesse, B. Glade, P. Ver ssimo, and K. Birman. A dynamic light weight group service. Technical report, INESC/Cornell Univ., Mar. 1996.

No context found.

L. Rodrigues, K. Guo, P. Verissimo, and K. Birman. A dynamic light-weight group service. Journal on Parallel and Distributed Computing, (60):1449--1479, 2000.

No context found.

L. Rodrigues, K. Guo, A. Sargento, R. van Renesse, B. Glade, P. Verissimo, and K. Birman. A dynamic light-weight group service. In 15th IEEE International Symposium on Reliable Distributed Systems (SRDS), pages 23--25, Oct. 1996.

No context found.

Luis Rodrigues, Katherine Guo, Antonio Sargento, Robbert van Renesse, Brad Glade, Paulo Verissimo, and Ken Birman. A dynamic light-weight group service. In 15th IEEE International Symposium on Reliable Distributed Systems (SRDS), pages 23-25, October 1996. also Cornell University Technical Report, TR96-1611, August, 1996.

No context found.

Luis Rodrigues, Katherine Guo, Antonio Sargento, Robbert van Renesse, Brad Glade, Paulo Verissimo, and Ken Birman. A dynamic light-weight group service. In 15th IEEE International Symposium on Reliable Distributed Systems (SRDS), pages 23{ 25, October 1996. also Cornell University Technical Report, TR96-1611, August, 1996.

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