Naohiro Hayashibara

Failure detectors for large-scale distributed systems

This paper discusses the problem of implementing a scalable failure detection service for grid systems. More specifically, traditional implementations of failure detectors are often tuned for running over local networks and fail to address important problems found in wide-area distributed systems, such as grid systems. We identify some of the most important problems raised in the context of grids. We then survey recent propositions that can help in solving some of these problems.

A Distributed Coordination Protocol for a Heterogeneous Group of Peer Processes

In peer-to-peer (P2P) applications like computer supported cooperative work (CSCW), multiple peer processes are required to cooperate to make a global decision, e.g. fix a meeting schedule of multiple persons. We discuss how multiple peer processes make a decision to achieve some objectives in a peer-to-peer (P2P) overlay network. Here, every process is assumed to be peer and autonomous. That is, there is no centralized coordination. A domain of a process is a collection of possible values which the process can take. Each process first takes a value v in its domain and notifies the other processes of the value v. A process can change the value with another value on receipt of values from other processes. However, a process can take only some value depending on the value v. For example, a process may abort after notifying commit but cannot commit after notifying abort in the commitment control. An existentially (E)-precedent relation shows what values a process can take after taking a value. In addition, a process takes a more preferable value if the process can take one of multiple values. Thus, values are ordered in the preferentially (P)-precedent relation. Based on the E- and P- precedent relations, each process takes the most preferable one in the values which can be changed from the current value v. In this paper, we discuss how every process makes an agreement on a value while each process can change the value according to the relations. In this paper, we discuss a coordination protocol in a type of heterogeneous system where every pair of processes have different E-precedent relation and P-precedent relation on the same domain. Each process learns a part of the precedent relations of another process through exchanging values.

Performance comparison of a rotating coordinator and a leader based consensus algorithm

Protocols that solve agreement problems are essential building blocks for fault tolerant distributed systems. While many protocols have been published, little has been done to analyze their performance, especially the performance of their fault tolerance mechanisms. In this paper, we compare two well-known asynchronous consensus algorithms. In both algorithms, a leader process tries to impose a decision, and another leader retries if the leader fails doing so. The algorithms elect leaders differently: the Chandra-Toueg algorithm has a rotating leader, whereas processes in the Paxos algorithm elect leaders directly. We investigate the performance implications of this difference. In the system under study, processes send atomic broadcasts to each other. Consensus is used to decide the delivery order of messages. We evaluate the steady state latency in (1) runs with neither crashes nor suspicions, (2) runs with crashes and (3) runs with no crashes in which correct processes are wrongly suspected to have crashed, as well as the transient latency after (4) one crash and (5) multiple correlated crashes. The results show that the Paxos algorithm tolerates frequent wrong suspicions (3) and correlated crashes (5) better, while the performance is comparable in all other scenarios.

A fault-tolerant model for wireless sensor-actor system

In a wireless sensor and actor network (WSAN), a group of sensors and actors are geographically distributed and linked by wireless networks. Sensors gather information sensed for an event in the physical world and send them to actors. Actors perform appropriate actions on actuation devices by making a decision on receipt of sensed information from sensors. Sensors are low cost, low powered devices with limited energy, computation, and wireless communication capabilities. Sensors may not only stop by fault but also suffer from arbitrary faults. Furthermore, wireless communication is less reliable due to noise and shortage of power of sensors. Reliable real time communication among sensors, actors, and actuation devices, is required in WSAN applications. We newly propose a multi-actor/multi-sensor (MAMS) model. In addition, multiple actors may perform actions on receipt of sensed information. Multiple redundant execution of an action on each device has to be prevented and conflicting actions on each device from multiple actors have to be serialized. In this paper, we discuss how to make WSAN reliable and available and how to reliably and non-redundantly perform actions with realtime constraints.

CBF: LOOK-UP PROTOCOL FOR DISTRIBUTED MULTIMEDIA OBJECTS IN PEER-TO-PEER OVERLAY NETWORKS

Various types of applications take usage of multimedia objects like music and movies. Multimedia objects are distributed in peer-to-peer (P2P) overlay networks since the objects are downloaded and personalized in computers, mainly personal computers interconnected with the Internet. An application would like to take some service of a target object. A target peer is a peer which can manipulate a target object. First, the application has to find target peers which can support enough quality of service (QoS) of target multimedia objects. Due to the scalability and variety of peers in P2P overlay networks, it is difficult, possibly impossible to maintain a centralized directory showing in which peer each object is distributed. In this paper, we newly take an acquaintance approach where each peer maintains its acquaintance peers. An acquaintance peer of a peer p is a peer whose service the peer p knows and with which the peer p can directly communicate in P2P overlay networks. We discuss types of acquaintance relations of peers with respect to what objects each peer holds, is allowed to manipulate, and can grant access rights on. In addition, we discuss a new type of flooding algorithm named charge-based flooding (CBF) algorithm to find target peers based on charge and acquaintance concepts so that areas in P2P overlay networks where target peers are expected to exist are more deeply searched. We evaluate the charge-based flooding (CBF) algorithm compared with a traditional TTL-based flooding algorithm in terms of the number of messages transmitted in networks.

Evaluation of Reliable Data Transmission Protocol in Wireless Sensor-Actuator Network

In a wireless sensor-actuator network (WSAN), sensor and actuator nodes exchange messages in a wireless channel. Sensor nodes can deliver messages to only close nodes while actuator can deliver to distant nodes. Messages sent by nodes might be lost due to collision and noise. The more number of sensor nodes are located in an event area, the more number of messages are lost due to collision. In the redundant data transmission (RT) protocol, a sensor node sends a message with not only its sensed value but also sensed values received from other sensor nodes. Even if a message with a sensed value v from a sensor node is lost, an actuator node can take the value v from other messages. We evaluate the RT protocol compared with the CSMA protocol in terms of how much sensing data a node can receive in presence of messages loss.

Scalable peer-to-peer multimedia streaming model in heterogeneous networks

In a peer-to-peer (P2P) overlay network, a large number and various types of peer processes are cooperating by using multimedia contents like movies. Multimedia streaming is a key technology to realize multimedia applications. Here, multimedia contents are required to be reliable and continuously delivered to processes in a realtime manner. In this paper, we newly discuss a heterogeneous asynchronous multi-source streaming (HAMS) model where multiple contents peers transmit packets of a multimedia content to a requesting leaf peer to increase the throughput, reliability, and scalability in P2P overlay networks.

Causally ordered delivery with global clock in hierarchical group

In peer-to-peer (P2P) networks, large number of peer processes are cooperating. In this paper, we discuss a scalable group of processes where processes are widely distributed in networks. Clocks of computers in every local network are synchronized by using the network time protocol (NTP) with a GPS time server. We discuss a global clock group (GCG) protocol where messages are causally ordered by using the physical time stamps. Messages not to be ordered by physical clock are furthermore ordered by using linear clock. We evaluate the protocol in terms of the number of messages ordered compared with the vector clock.

HAMS: scalable peer-to-peer multimedia streaming model in heterogeneous networks

In a peer-to-peer (P2P) overlay network, a large number and various types of peer processes are interconnected in networks and are cooperating by using multimedia contents like movies and music. Here, multimedia contents are in nature distributed to peers in various ways like downloading and caching to the peers. Multimedia streaming is a key technology to realize multimedia applications in networks. In multimedia streaming applications, multimedia contents are required to be reliable and continuously delivered to processes in a real-time manner. Some contents peer may not send packets of a content at a required rate due to limited computation resource and a communication channel may not support enough Quality of Service (QoS) due to congestions and faults. Thus, P2P overlay networks are in nature heterogeneous. In this paper, we newly discuss a heterogeneous asynchronous multisource streaming (HAMS) model where multiple contents peers transmit packets of a multimedia content to a requesting leaf peer to increase the throughput, reliability, and scalability in P2P overlay networks. Here, some pair of channels between contents and leaf peers may support different QoS. Peers may be faulty and some pair of contents peers may have different transmission rates. Finally, we show the HAMS model can support higher throughput and shorter transmission time than the other models in the evaluation.

Efficient Data Transmission in a Lossy and Resource Limited Wireless Sensor-Actuator Network

In a wireless sensor-actuator network (WSAN), sensors send sensed values to actuators in a wireless channel. Messages sent by nodes might be lost due to collision and noise if multiple nodes simultaneously send the messages. If a message is detected to be lost, the source node retransmits the message to the destination nodes. It takes at least three rounds to detect and retransmit a lost message since the source node sends the message. The more number of messages are retransmitted, the more number of messages collide. We have to reduce the number of messages retransmitted to realize the efficient data transmission. We newly propose a redundant data transmission (RT) protocol where a sensed value is redundantly carried to the destination node by multiple messages. Even if a message is lost, the destination node can receive data in the lost message if the destination node receives other redundant messages. In addition, we have to reduce the energy consumption of a sensor node. A sensor node mainly consumes the energy to send and receive messages. Even if an event occurs, only some number of sensor nodes sensing the event send the sensed values to reduce the total energy consumption. We discuss an energy-efficient data transmission protocol. We evaluate the RT protocol compared with the CSMA protocol in terms of how much sensing data a node can receive in presence of messages loss