Giwon On

The effectiveness of realistic replication strategies on quality of availability for peer-to-peer systems

We take an availability-centric view on quality of service (QoS) and propose a model and mechanisms for studying the effectiveness of realistic replication schemes on availability QoS for peer-to-peer (P2P) systems. We especially tackle the dynamic replica placement (RP) problem where our focus is on choosing dynamically the number and location of replicas while (1) meeting different availability QoS requirements for all individual peers and (2) taking the intermittent connectivity of peers explicitly into account. We model P2P systems as a dynamic stochastic graph in which the nodes go up and down depending on their assigned up probability. We develop some simple heuristic algorithms for solving the RP problem, which are fully distributed and adaptive. Through an event-driven simulation study we compare and evaluate the achieved availability QoS of the proposed RP algorithms. Simulation results show that (1) even simple heuristics can achieve reasonably high availability QoS, and (2) satisfying availability QoS requires more replicas than for only increasing the hit rate.

Quality of availability: replica placement for widely distributed systems

In this paper, we take an availability-centric view on Quality of Service (QoS) and focus on the issues of providing availability guarantees for widely distributed systems such as web servers and peer-to-peer (P2P) file sharing systems. We propose a concept called Quality of Availability (QoA) in which the availability is treated as a new controllable QoS parameter. The newly refined fine-grained availability definitions and QoA metrics enable the specification and evaluation of the different level of availability for different users and applications. We tackle specifically the replica placement (RP) problem where our focus is on choosing the number and location of replicas while (1) meeting different availability QoS requirement levels for all individual users and (2) taking the intermittent connectivity of system nodes explicitly into account. We decompose the RP problem into two sub-problems: (1) improving QoA and (2) guaranteeing QoA. We investigate a number of simulations - for full and partial replication models and static and dynamic placements - to compare and evaluate the achieved availability QoS of the developed RP algorithms. Our proposed QoA concept and model can be used as a base mechanism for further study on the effectiveness of realistic replication schemes on both availability and performance QoS for widely distributed systems.

On Availability QoS for Replicated Multimedia Service and Content

Recently, it has been realized that the importance of satisfying service availability is becoming one of the most critical factors for the success of Internet-based services and applications. In this paper, we take an availability-centric view on QoS where the availability is treated as a new controllable QoS parameter and focus on the issues of providing availability guarantees for distributed and replicated multimedia services and contents. We especially tackle the replica placement (RP) problem and study the effects of number and location of replicas on the achieved availability. From a simulation study, we find that (1) the location of replica is a more relevant factor than their number for satisfying the availability QoS requirements for all individual users, and (2) the heuristic methods, in general, cannot give any guarantee for their achieved availability QoS, while they are very efficient for large size graphs.

Replay for debugging MPI parallel programs

The cyclic debugging approach often fails for parallel programs because parallel programs reveal nondeterministic characteristics due to message race conditions. This paper addresses the execution replay algorithm for debugging MPI parallel programs. The lexical analyzer identifies the MPI events which affect nondeterministic executions, and then an execution is controlled in order to make it equivalent to a reference execution by keeping their orders of events in two executions identical. The proposed replay system uses the logical time stamping algorithm and the derived data types provided by MPI standard. It also presents the method of how to replay the blocking and nonblocking message passing events. The proposed replay system was applied to the bitonic-merge sort and other parallel programs. We found that re-execution has reproducible behavior and the replay system is useful to find the communication errors.

QoS-Controlled Dynamic Replication in Peer-to-Peer Systems

ABSTRACT This paper presents a study of dynamic replication for peer-topeer networks. We take an availability-centric view on quality of service (QoS) and focus on the issues of satisfying availability requirements for distributed multimedia services running on large Peer-to-Peer (P2P) systems. We especially tackle the replica placement problem where our focus is on choosing dynamically the number and location of replicas while (1) satisfying the availability QoS requirement for all individual peers and (2) taking the intermittent connectivity of peers explicitly into account. For this purpose, we model P2P systems as a dynamic stochastic graph in which the nodes go up and down depending on their assigned up probability and issue content access events with a certain level of availability requirement. Through an event-driven simulation study we compare and evaluate replication schemes which are fully distributed and adaptive and which satisfy the availability QoS requirements. Simulation results show that (1) satisfying availability QoS requires more replicas than for only increasing the hit rate, (2) the location of replicas is a more relevant factor than their number for satisfying availability QoS, and (3) even simple heuristics can achieve reasonably high availability QoS. Our proposed replication model can be used for further study on the dual availability and performance QoS for dynamically changing, large-scale P2P systems.

Design and Implementation of a QoS-Aware Replication Mechanism for a Distributed Multimedia System

This paper presents the design and implementation architecture of a replication mechanism for a distributed multimedia system medianode which is currently developed as an infrastructure to share multimedia-enhanced teaching materials among lecture groups. The proposed replication mechanism supports the quality of service (QoS) characteristics of multimedia data and the availability of system resources. Each type of data handled and replicated are classified according to their QoS characteristics and replication requirements. The main contribution of this paper is the identification of new replication requirements in distributed multimedia systems and a multicast-based update propagation mechanism by which not only the update events are signalled, but also the updated data are exchanged between replication managers. By prototyping the proposed replication mechanism in medianode, we prove the feasibility of our concept for combining the QoS concept with replication mechanisms.

Replication for a distributed multimedia system

Replicating data and services at multiple networked computers increases the service availability of distributed systems. This paper presents the design and implementation architecture of a replication mechanism for a distributed multimedia system medianode which is developed as an infrastructure to share multimedia-enhanced teaching materials among lecture groups. With the replication mechanism, medianode provides enhanced access to presentation materials in both connected and disconnected operation modes. The main contribution of this paper is the identification of new replication requirements in distributed media systems and a multicast-based update propagation mechanism by which not only the update events are signaled, but also the updated data are exchanged between replication managers.

A Scheme for Building Visual Debugging Environment with Dynamic Debugging Method for Parallel Systems

Cyclical debugging with replay support and visualization are two key techniques to assist in debugging parallel programs. In this contribution we present the concept of an integrated visual debugging environment for debugging parallel programs in both distributed and shared memory computational models. This environment is designed to offer several graphical views of a programs behavior while giving textual informations as result of cyclical debugging. This is achieved by mapping the particular textual point into the corresponding graphical view point automatically. The user can repeatedly stop the program during execution, examines the state, and then continues or re-execute the program, while analyzing or animating the corresponding trace event in graphical views. The proposed textual/graphical view mapper bridges the cyclical and visual debugging and is useful for error isolation in a parallel programming environment.