Academia.eduAcademia.edu

Distributed Deadlock

description24 papers
group23 followers
lightbulbAbout this topic
Distributed deadlock refers to a situation in a distributed computing system where two or more processes are unable to proceed because each is waiting for a resource held by another, creating a cycle of dependencies that prevents any of the involved processes from making progress.
lightbulbAbout this topic
Distributed deadlock refers to a situation in a distributed computing system where two or more processes are unable to proceed because each is waiting for a resource held by another, creating a cycle of dependencies that prevents any of the involved processes from making progress.

Key research themes

1. How can Petri nets and siphon-based methods be used to prevent deadlocks in multithreaded and automated manufacturing distributed systems?

This theme explores formal modeling and control techniques for deadlock prevention in distributed systems using Petri nets, particularly Gadara nets and siphon/trap methods. Preventing deadlocks in multithreaded and multi-agent manufacturing systems requires understanding resource allocation dynamics and employing efficient supervisory control policies that guarantee system liveness and maximal permissiveness. Petri net models allow structural analysis to identify problematic siphons or resource allocations that lead to deadlocks, which can be constrained with monitors or supervisor places to enable live operation without deadlocks.

Key finding: The paper proposes an iterative deadlock prevention policy for ordinary Gadara nets modeling multithreaded software, focusing on minimal resource-number emptiable siphons (MRES). At each iteration, it identifies all MRESs,... Read more
Key finding: Two Petri net-based methods for deadlock prevention in automated manufacturing systems are compared: one using place invariants (P-invariants) and the other utilizing siphons and traps. Both approaches add additional monitor... Read more
Key finding: The paper presents c-INPRES, a strategy for minimizing the number of locked agents needed to prevent cyclic instability (deadlocks) in ambient intelligence systems by leveraging the strong and weak coupling concepts in the... Read more
Key finding: The paper describes a distributed algorithm for detecting deadlocks by constructing and searching a distributed transaction wait-for graph (TWFG). Each site only knows a local portion of the TWFG and transactions may have... Read more

2. What are the design and performance trade-offs in distributed lock management over RDMA-enabled networks to prevent starvation and deadlock?

With the emergence of fast RDMA networks, distributed lock managers must balance scalability, latency, starvation prevention, and deadlock handling while avoiding centralized bottlenecks. This theme investigates decentralized lock management algorithms that achieve starvation-free, first-come-first-serve scheduling without explicit queues, exploiting RDMA atomic primitives (fetch-and-add) rather than compare-and-swap. It evaluates performance improvements in throughput, latency, and fairness essential to modern distributed transactional systems.

Key finding: The paper introduces DSLR, a fully decentralized, starvation-free lock manager tailored for RDMA-enabled networks, which adapts Lamport’s bakery algorithm with fetch-and-add (FA) atomic operations to prevent blind retries and... Read more

3. How can deadlock detection scheduling and algorithmic enhancements optimize the trade-off between deadlock detection overhead and resolution cost in distributed systems?

Effective distributed deadlock handling requires balancing detection frequency to minimize combined overhead of detection message complexity and resolution duration. This theme studies stochastic and analytical models identifying optimal deadlock detection intervals as functions of system parameters like deadlock formation rate and process count. It also explores algorithms for efficient detection and resolution of generalized deadlocks with minimal message, time complexity, and streamlined termination detection.

Key finding: Presents a theoretical analysis determining that the optimal frequency of initiating deadlock detection in distributed systems is asymptotically O((λn)^{1/3}), where λ is the deadlock formation rate and n is the number of... Read more
Key finding: Proposes a novel algorithm for detecting generalized deadlocks under the P out-of Q resource request model by diffusing probes and collecting replies carrying direct dependency information, minimizing message length and... Read more
Key finding: Develops a priority-based edge-chasing distributed deadlock detection algorithm that initiates detection only on antagonistic conflicts (when a higher priority transaction waits for a lower priority one), thus reducing... Read more

All papers in Distributed Deadlock

In this paper we present a comparison between six novel approaches to the fundamental problem of cyclic instability in Ambient Intelligence. These approaches are based on different optimization algorithms, Particle Swarm Optimization... more
In this paper the problem of cyclic instability in dynamic environments is presented. This cyclic instability is generated when binary rule-based nomadic agents (agents entering or leaving the environment) interact in complex ways,... more
Deadlock is a phenomenon in which a system or a part of it remains indefinitely blocked and cannot terminate its task. Such phenomenon often implies disaster in man-made systems and, therefore, must be carefully handled by system... more
Download research papers for free!