Fact-checked by Grok 2 weeks ago

Atomic broadcast

Atomic broadcast is a fundamental communication primitive in distributed systems that guarantees all correct processes deliver the same set of messages in the same , even in the presence of asynchronous communication, process crashes, and unreliable networks. It extends reliable broadcast by enforcing a consistent delivery sequence across all participants, making it essential for maintaining agreement on system state without a central . The protocol satisfies four key properties to ensure correctness: validity, where if a correct broadcasts a , every correct eventually delivers it; agreement, ensuring that if one correct delivers a , all correct processes do so; uniform integrity, stipulating that a is delivered at most once and only if it was broadcast by some ; and , requiring that the delivery order of messages is identical for all correct processes. These properties hold in asynchronous environments tolerant to fewer than half (f < n/2) of the total processes using failure detectors such as Ω. Introduced in the context of fault-tolerant computing, atomic broadcast is equivalent to the problem in asynchronous crash-prone systems, allowing solutions to one to be transformed into solutions for the other. Seminal work by and Toueg demonstrated its solvability using unreliable failure detectors, which provide eventual accuracy about process failures without perfect synchrony. It plays a critical role in applications such as for high-availability services, where replicas must process client requests in the same sequence to preserve consistency, and in crash-recovery models that support process restarts using stable storage. Modern implementations often optimize for scalability in large-scale systems, incorporating techniques like hierarchical structures or to reduce latency and message overhead.

Introduction and Fundamentals

Definition and Motivation

In distributed systems, multiple independent processes interact by sending and receiving messages across an unreliable communication network, typically in an asynchronous setting where there are no guaranteed bounds on message transmission delays or processing times. This asynchrony, combined with the potential for network partitions or node variability, makes it challenging for processes to maintain a consistent shared state without specialized coordination mechanisms. Atomic broadcast, also referred to as or broadcast, is a core communication primitive designed to address these issues by ensuring that a message broadcast by a correct sender is either delivered to all correct processes in the exact same sequence or not delivered to any, thereby achieving atomicity in group communication. This primitive abstracts away the complexities of unreliable channels, providing a reliable and ordered service that all correct participants can rely on for synchronization. The primary motivation for atomic broadcast stems from the need to overcome common pitfalls in asynchronous distributed environments, such as message losses, out-of-order arrivals, and transient faults, which can otherwise result in divergent system behaviors or inconsistent data across nodes. It enables the development of robust higher-level abstractions, notably , where client requests are totally ordered and executed identically on all replicas to ensure fault-tolerant service continuity. For example, in a scenario, atomic broadcast allows update transactions from a client to be disseminated such that all replicas apply them in the same order, preserving global consistency without a .

Historical Development

The concept of atomic broadcast emerged in the as part of research on reliable protocols within fault-tolerant distributed systems, particularly through efforts to ensure consistent message delivery in groups. A foundational contribution came from Kenneth Birman's work on virtual synchrony, which provided a model for group communication where es maintain a consistent view of membership changes and message orders, underpinning early implementations of atomic broadcast-like primitives. This approach was integrated into toolkit, a distributed programming environment developed at that supported virtually synchronous groups for reliable dissemination. The 1985 Fischer-Lynch-Paterson (FLP) impossibility result profoundly influenced atomic broadcast designs by demonstrating that no deterministic algorithm can guarantee termination in fully asynchronous systems prone to even a single crash failure, prompting researchers to incorporate failure detectors or partial synchrony assumptions into broadcast protocols. In the , the primitive was formally defined and refined, with Tushar and Sam Toueg establishing the equivalence between atomic broadcast and in asynchronous crash-prone systems, enabling reductions that facilitated uniform atomic broadcast implementations using unreliable failure detectors. Research evolved in the late and to address more adversarial failure models, shifting from faults to Byzantine faults where processes could behave maliciously; this was exemplified by protocols integrating atomic broadcast into Byzantine fault-tolerant , such as those building on practical BFT mechanisms. Concurrently, atomic broadcast was embedded in systems like , the successor to , which extended virtual synchrony to handle dynamic group memberships and fault recovery more efficiently. Luís and Michel Raynal further advanced the field by developing atomic broadcast algorithms for asynchronous -recovery models, ensuring message ordering even when processes could and later recover without stable storage. By the 2010s, atomic broadcast saw increased adoption in cloud computing environments to support scalable replication and consistency in distributed databases and services, with optimized protocols leveraging modern hardware like RDMA for low-latency delivery in geo-replicated systems.

Core Properties

Validity and Integrity

In atomic broadcast, the validity property ensures that messages originating from correct processes are reliably delivered. Specifically, if a correct process broadcasts a message, then it will eventually deliver that message. This liveness guarantee, together with agreement, ensures delivery to all correct processes and prevents the loss of legitimate messages in the presence of benign failures, providing a foundation for dependable message dissemination in distributed systems. Formally, validity can be stated as: for all correct processes p and messages m, if p broadcasts m, then p eventually m. This property aligns with the requirements of reliable broadcast but is extended in atomic broadcast to support subsequent and ordering among all correct participants. The formulation emphasizes by the broadcaster, with system-wide assured by , distinguishing it from point-to-point communication where is not assured system-wide. The integrity property complements validity by safeguarding against fabrication and duplication. It stipulates that every —correct or faulty—delivers a at most once, and only if that was previously broadcast by some . This no-creation and no-duplication assurance protects the system from spurious or repeated deliveries, even if faulty processes attempt to inject invalid . Formally, uniform integrity is expressed as: for all processes p and m, p delivers m at most once, and only if some q previously broadcast m. The "" aspect applies the constraint to all processes, strengthening the property beyond non-uniform variants that might allow faulty processes to deliver multiple times or incorrectly. This is crucial for maintaining the authenticity and uniqueness of in fault-tolerant environments. These properties differentiate atomic broadcast from weaker like unreliable , which offers no guarantees against message loss, duplication, or fabrication—potentially leading to inconsistent or incomplete deliveries across processes. In contrast, validity and uniform integrity ensure that only intended messages from correct sources propagate without alteration or , forming the core guarantees for higher-level in distributed applications.

Agreement and Total Order

In atomic broadcast, the agreement property ensures that messages are delivered uniformly across all correct processes, preventing scenarios where some correct processes receive a message while others do not. Specifically, agreement guarantees that if a process delivers a message m, then all correct processes will eventually deliver m as well. This property is crucial for maintaining consistency in distributed systems, as it ensures the set of delivered messages is identical for all correct participants, regardless of timing or network delays. The formal statement of agreement is: for any process p and all correct processes q, and for any message m, if p delivers m, then eventually q delivers m (denoted as \deliver(p,m) \rightarrow \Diamond \deliver(q,m) \ \forall q \ correct). This formulation, introduced in foundational analyses of broadcast, emphasizes the among correct processes even in the presence of failures. Complementing , the property imposes a global sequencing on delivered messages, ensuring that all correct processes observe the same relative order. specifies that if two processes p and q both deliver messages m_1 and m_2, then p delivers m_1 before m_2 if and only if q delivers m_1 before m_2. This creates a single, linear history of messages shared by all correct processes, which is essential for applications requiring synchronized state updates, such as replicated databases. Formally, is expressed as: for all processes p and q, and for any messages m_1 and m_2, \deliver(p,m_1) < \deliver(p,m_2) \leftrightarrow \deliver(q,m_1) < \deliver(q,m_2) if both deliver them. This bidirectional equivalence, as defined in surveys of broadcast protocols, guarantees that no correct process can perceive a different ordering, thereby avoiding conflicts in concurrent operations. Together, agreement and provide a stronger guarantee than weaker ordering semantics like first-in-first-out () or causal ordering. While preserves per-sender sequence and causal order respects chains, atomic broadcast's properties enforce a global across all messages from any sender, ensuring full coordination without partial views.

Fault Tolerance Aspects

Failure Models

Atomic broadcast protocols are designed to operate under various failure models that capture different types of and communication faults in distributed systems. These models provide the environmental assumptions under which the protocol's properties—such as agreement and —are guaranteed, assuming a bounded number of failures. The crash-fault model, also known as the crash-stop model, assumes that faulty es halt execution abruptly and never recover, behaving correctly until the crash occurs. This is the simplest failure model for atomic broadcast, as it does not involve malicious or partial behaviors, allowing protocols to tolerate up to f < n/2 faulty es among n total es. In this model, atomic broadcast can be reduced to reliable broadcast combined with a mechanism for sequencing messages, such as using a leader or failure detector to establish a . The omission-fault model extends the crash model by allowing processes to fail to send or receive specific messages while otherwise operating correctly; this includes send-omission failures, where a process does not insert a message into its outgoing channel buffer, and receive-omission failures, where a message in the incoming buffer is not processed. Omission failures are more challenging than crashes because they can lead to incomplete message dissemination without halting the process entirely, requiring protocols to handle sporadic non-delivery while ensuring eventual consistency. The Byzantine-fault model represents the most severe case, where faulty processes can exhibit arbitrary, potentially malicious behavior, such as sending conflicting messages, colluding with other faulty processes, or deviating from the protocol in any way. This model encompasses and omission failures as subsets and demands stronger mechanisms like digital signatures or to detect and mitigate inconsistencies; atomic broadcast protocols under this model can tolerate up to f < n/3 Byzantine faults. Atomic broadcast protocols typically operate under partial synchrony or fully asynchronous timing assumptions, where message delays and processing times are either bounded after some global stabilization point or unbounded, respectively, without requiring a global clock for coordination. These models ensure that properties like uniform agreement are tested and preserved across the scenarios described.

Resilience Bounds and Limits

In asynchronous systems subject to crash faults, atomic broadcast protocols can tolerate up to f < n/2 faulty processes, where f denotes the maximum number of crashes and n the total number of processes. This bound ensures that a of processes remain operational to achieve and . However, without any synchrony assumptions, deterministic atomic broadcast becomes impossible even for f \geq 1 crash failure, extending the FLP impossibility result originally established for in asynchronous environments. For Byzantine fault models, where processes may exhibit arbitrary malicious behavior, the resilience of uniform atomic broadcast is restricted to f < n/3. This stricter threshold arises from the fundamental lower bound demonstrated in the Byzantine Generals Problem, which shows that —and by equivalence, uniform atomic broadcast—cannot be solved if fewer than three-quarters of processes are correct. Resilience metrics for atomic broadcast further highlight practical limits. Message complexity reaches O(n^2) in the worst case, as seen in foundational protocols like Bracha's reliable broadcast, which forms a building block for atomic variants and requires quadratic exchanges to propagate and verify messages across all processes. Under partial synchrony, where bounds on delays hold eventually after a global stabilization time (), time complexity stabilizes to a constant number of rounds post-GST, enabling efficient termination while preserving up to the respective bounds.

Relation to Consensus

Theoretical Equivalence

In theory, atomic broadcast is formally equivalent to in asynchronous systems tolerant to failures, meaning that each problem can be solved the other can be solved under the same fault model. This equivalence holds specifically for their uniform variants: uniform atomic broadcast, where no two processes deliver different messages or sequences, is solvable uniform , where no two processes decide differently, is solvable. The reduction from atomic broadcast to consensus uses a sequence of consensus instances, where each instance decides on a batch of messages that have been reliably broadcast but not yet delivered. To atomically broadcast a message m, a process first reliably broadcasts it; when there are pending messages, all processes invoke with the set of pending messages as the proposal, deliver the decided batch in a deterministic order (e.g., by sender ID and ), and proceed to the next instance upon termination. This ensures the uniform agreement, , and validity properties of atomic broadcast, as guarantees that all correct processes agree on the same batch for each slot and eventually terminate. The reverse reduction, from to atomic broadcast, is achieved by having each atomically broadcast its proposed value (or the null value \perp if none). All processes then decide on the first non-\perp value delivered, leveraging the property to ensure agreement and validity. This equivalence implies that atomic broadcast inherits the fundamental impossibility results of . In particular, by the Fischer-Lynch-Paterson (FLP) theorem, no deterministic protocol for atomic broadcast (or uniform ) exists in an with crash failures, even tolerating just one fault, without additional assumptions such as failure detectors or partial synchrony.

Practical Implications

The equivalence between atomic broadcast and enables modular design in distributed systems, where atomic broadcast is frequently implemented by layering a —such as —for message sequencing, allowing developers to leverage established mechanisms for guarantees while composing higher-level abstractions like . This approach facilitates reusable components but introduces additional from the rounds required per message batch, as each decision on order involves multiple communication phases among nodes. Performance trade-offs arise primarily from the overhead of invocations, where delivering a single message can require up to message exchanges in a of n nodes due to quorum-based , leading to latencies dominated by network round-trip times (e.g., 45-70 ms across data centers). Optimizations mitigate this by batching multiple messages into fewer instances to amortize costs or using view changes to re-elect leaders efficiently during failures, thereby improving throughput in high-load scenarios without sacrificing safety. In partially synchronous models, the supports deterministic decisions once a global stabilization time () elapses, ensuring progress under bounded delays and influencing system choices: atomic broadcast is preferred for applications needing ordered (e.g., replicated logs), while direct suffices for simpler agreement tasks. Real-world implementations diverge from pure asynchronous theory by incorporating synchrony assumptions—such as timeouts and heartbeats—to guarantee termination and bypass the FLP impossibility result, which prohibits (and thus atomic broadcast) in fully asynchronous settings with even one crash failure.

Algorithms and Implementations

Foundational Algorithms

The foundational algorithms for atomic broadcast emerged in the early 1990s to solve the problem in asynchronous distributed systems subject to crash failures, providing the theoretical and practical basis for subsequent developments. A seminal contribution is the algorithm by T. D. Chandra and S. Toueg, which reduces atomic broadcast to reliable broadcast and primitives to achieve uniform properties such as uniform agreement and uniform , ensuring all correct processes deliver the same messages in the same sequence despite crashes. The underlying uniform reliable broadcast in this construction uses three phases—echo, ready, and commit—to ensure message dissemination. In the echo phase, processes relay the broadcast message to confirm reception; the ready phase allows processes to signal preparedness for delivery once a threshold of echoes is received; and the commit phase finalizes delivery once a threshold of readies is received. The overall atomic broadcast leverages this flooding-based reliable broadcast for initial propagation (PER-PROPOSE messages) and consensus instances to resolve ordering, tolerating up to f < n/2 crash failures where n is the number of processes. In the crash-fault model, the message complexity is O(n^2), arising from the quadratic overhead of message relays in reliable broadcast combined with consensus rounds. A high-level pseudocode outline for the broadcast operation is as follows:
upon event ABroadcast(m):
  ReliableBroadcast(PER-PROPOSE, m)

upon event R-Deliver(PER-PROPOSE, m) from sender s:
  add m to local set of pending messages
  if consensus instance decides on a set S of messages (including m if proposed by correct sender):
    deliver messages in S in deterministic order (e.g., sequence number or lexical)
    for each m' in S: A-Deliver(m')
This construction invokes reliable broadcast to propose messages and uses to agree on batches for delivery, ensuring atomicity without requiring synchronous timing assumptions. An extension to handle crash-recovery failures was proposed by L. Rodrigues and M. Raynal, adapting the consensus-based for systems where processes can restart using stable storage, while tolerating up to f < n/2 crash failures. Their approach builds on the consensus-based through mechanisms for dissemination and optional logging for recovery. Key innovations across these foundational works include the integration of unreliable failure detectors to enable progress in asynchronous environments without perfect synchrony. These algorithms establish atomic broadcast as reducible to in the crash-fault setting, providing a modular foundation where reliable broadcast handles dissemination and enforces agreement.

Modern Protocols and Systems

One prominent modern protocol for atomic broadcast is the Atomic Broadcast (ZAB), introduced in 2011, which provides for coordination in primary-backup systems. ZAB integrates to select a primary that sequences messages, followed by atomic phases ensuring all non-faulty processes agree on the despite failures, tolerating up to f < n/2 faulty processes in a system of n processes. This protocol underpins , enabling reliable state synchronization for distributed applications like and . In the 2020s, advancements have emphasized RDMA-based protocols to enhance performance in high-throughput environments, leveraging for low-latency, CPU-bypass communication. For instance, Acuerdo, proposed in 2022, achieves fast atomic broadcast using one-sided RDMA writes to replicate messages efficiently, reducing latency compared to traditional TCP-based approaches while maintaining crash-fault tolerance. Similarly, RamCast (2021) implements RDMA-accelerated atomic , an extension supporting multiple groups, which scales better for multicast scenarios by minimizing coordination overhead in datacenter networks. These protocols build on Fast variants, optimizing leader-driven sequencing for RDMA hardware to handle millions of operations per second in production clusters. Scalability improvements in modern atomic broadcast protocols address geo-replicated systems through extensions like , originally from 1988 but revisited and extended in works such as the 2024 Vertical Atomic Broadcast, which supports reconfiguration for dynamic membership changes with minimal downtime. These extensions enable geo-distribution by allowing view changes that adapt to node additions or failures across wide-area networks, ensuring without halting delivery. Regarding message complexity, protocols like Slim-ABC (2024) achieve optimizations approaching O(n log n) in certain phases for large n, reducing total communication from O(n^2) in classic designs to support scalable deployments in cloud environments. As of 2025, further advancements include Carry-the-Tail, a deterministic atomic broadcast in partial synchrony that guarantees constant latency after the global stabilization time (), and , a scalable sharding-based Byzantine fault-tolerant atomic broadcast supporting up to f < n/3 failures. ZAB's principles have influenced systems beyond , including etcd, where consensus provides equivalent atomic broadcast semantics for key-value coordination, supporting atomic transactions and watches in distributed storage. Post-2011 developments, such as Paxos (2015), further address dynamic membership by allowing subscriptions to message groups without full reconfiguration, enhancing adaptability in elastic cloud settings.

Applications and Extensions

Real-World Use Cases

Atomic broadcast plays a crucial role in (SMR), a technique for building fault-tolerant distributed systems by ensuring all replicas execute the same sequence of client commands in the same order, thereby maintaining identical states. In databases like Google Spanner, atomic broadcast primitives—realized through Paxos-based —facilitate the replication of logs across geographically distributed replicas, enabling and external for global-scale data management. In distributed coordination services, atomic broadcast supports reliable and configuration management. For example, employs the ZAB (ZooKeeper Atomic Broadcast) protocol, which uses atomic broadcast to propagate state updates in , allowing nodes to coordinate locks, barriers, and while tolerating failures. Blockchain platforms and permissioned ledgers leverage atomic broadcast to achieve on ordering without requiring full validation by all nodes. In Fabric, the ordering service implements atomic broadcast to deliver blocks of transactions in a consistent sequence to endorsing peers, supporting modular in enterprise environments. Atomic broadcast has several variants that relax some of its core properties to suit specific requirements, such as reduced ordering guarantees or handling dynamic group membership. Reliable broadcast is a weaker primitive that ensures all correct processes deliver the same message sent by a correct sender, but without any ordering requirements, making it suitable for applications like news dissemination where sequence does not matter. This variant traces back to early protocols using tree-based forwarding for fault tolerance against crash failures. Causal broadcast extends reliability by preserving the causal order of messages—ensuring that if one message causally precedes another, all correct processes deliver them in that order—but it does not enforce total ordering across unrelated messages, which is useful in distributed databases with dependent operations. Protocols for causal broadcast often incorporate vector clocks or similar mechanisms to track dependencies. View-synchronous broadcast combines reliable broadcast with group membership services to handle process joins and leaves, ensuring that messages sent within a stable view (group configuration) are delivered before the view changes, thus coordinating delivery with membership dynamics in process groups. Related primitives include atomic multicast, which generalizes atomic broadcast to multiple groups by ensuring total order delivery within each group but allowing independent orders across groups, addressing scalability issues in large systems where not all processes need every message. Atomic multicast is particularly relevant in partitioned networks or hierarchical systems. Ordering mechanisms in these primitives often fall into dissemination-based approaches, where order emerges from message propagation and history (e.g., using causal logs or merging functions), versus token-based methods, which rely on circulating tokens or sequencers to explicitly assign sequence numbers, trading off latency for stronger guarantees in fault-prone environments. Extensions of atomic broadcast address dynamic environments and adversarial settings. Dynamic atomic broadcast supports processes joining or leaving voluntarily (or crashing) by integrating membership protocols that coordinate delivery slots, allowing new members to catch up on prior messages without violating order, as seen in protocols that use failure detectors for reconfiguration. Byzantine-robust variants enhance resilience against malicious faults by incorporating digital signatures or threshold signatures to verify message authenticity and prevent equivocation, ensuring even with up to one-third faulty processes in asynchronous networks. For instance, some protocols aggregate signatures to reduce communication overhead while maintaining verifiability. Weaker ordering variants like FIFO broadcast, which enforces first-in-first-out delivery per sender but not total order, find application in resource-constrained sensor networks where full atomicity is unnecessary and would incur high overhead; such protocols use simple queuing at receivers to approximate order in multi-hop wireless body area networks. Recent 2020s research has explored accountable variants of atomic broadcast, particularly in Byzantine settings, where protocols provide cryptographic proofs of misbehavior (e.g., equivocation) without relying on signatures, enabling efficient dispute resolution in permissioned blockchains and multi-party computation.

References

  1. [1]
    [PDF] Unreliable Failure Detectors for Reliable Distributed Systems
    We prove that Consensus and Atomic Broadcast are redueible to each other in asynchronous systems with crash failures; thus, the above results also apply to ...
  2. [2]
    [PDF] Atomic Broadcast in Asynchronous Crash-Recovery Distributed ...
    Abstract—Atomic Broadcast is a fundamental problem of distributed systems: It states that messages must be delivered in the same order to their destination ...
  3. [3]
    A Leaderless Hierarchical Atomic Broadcast Algorithm
    Jan 17, 2023 · Atomic Broadcast is an essential broadcast primitive as it ensures the consistency of distributed replicas. However, it is notoriously non- ...
  4. [4]
    [PDF] atomic broadcast - CS - Huji
    Atomic broadcast is a protocol for distributed systems that enables correct processes to attain consistent knowledge of the system state, despite failures and ...
  5. [5]
    [PDF] Implementing Fault-Tolerant Services Using the State Machine ...
    The state machine approach is a method for implementing fault-tolerant services in distributed systems, using multiple servers and protocols to coordinate ...
  6. [6]
    Exploiting virtual synchrony in distributed systems
    We describe applications of a virtually synchronous environment for distributed programming, which underlies a collection of distributed programming tools ...
  7. [7]
    [PDF] EXPLOITING VIRTUAL SYNCHRONY IN DISTRIBUTED SYSTEMS
    This paper begins by exploring the concept of vir- tual synchrony and the ways that it is reflected in the interfaces provided by the 1SIS 2 toolkit. We ...
  8. [8]
    [PDF] Impossibility of Distributed Consensus with One Faulty Process
    2, April 1985, pp. 374-382. Page 2. Impossibility of Distributed ... It follows from our impossibility result that every commit protocol has such ...
  9. [9]
    Unreliable failure detectors for reliable distributed systems
    We prove that Consensus and Atomic Broadcast are reducible to each other in asynchronous systems with crash failures; thus, the above results also apply to ...
  10. [10]
  11. [11]
    [PDF] An Efficient Atomic Broadcast Protocol for Cloud Computing Services
    Jun 3, 2013 · Abstract. For internet servers, how to handle colossal data volumes and maintain data consistency is a major challenge.
  12. [12]
    [PDF] A Modular Approach to Fault-Tolerant Broadcasts and Related ...
    With Uniform Atomic Broadcast, inconsistency may only result from skipping a message that is delivered by all correct processes. Once a faulty process becomes ...
  13. [13]
    [PDF] Atomic Broadcasts and Consensus: A Survey
    Following Cristian et al., an atomic broadcast protocol must fulfill the three properties. Atomicity, Order and Termination [5]. However, all three.
  14. [14]
    Total order broadcast and multicast algorithms: Taxonomy and survey
    In short, the primitive ensures that messages sent to a set of processes are, in turn, delivered by all those processes in the same total order.Missing: seminal | Show results with:seminal
  15. [15]
    [PDF] Atomic Broadcast in Asynchronous Crash-Recovery Distributed ...
    [18] L. Rodrigues and M. Raynal, Atomic Broadcast and Quorum-based Replication in Asynchronous Crash-Recovery Distributed Systems. FCUL/DI Technical Report.
  16. [16]
    [PDF] Reaching Agreement in the Presence of Faults
    Reaching Agreement in the Presence of Faults. M. PEASE, R, SHOSTAK, AND L. LAMPORT. SRI Internatwnal, Menlo Park, California. ABSTRACT. The problem addressed ...
  17. [17]
    [PDF] Consensus in the Presence of Partial Synchrony - Research
    Abstract. The concept of partial synchrony in a distributed system is introduced. Partial synchrony lies between the cases of a synchronous system and an ...
  18. [18]
  19. [19]
  20. [20]
    Managing Critical State: Distributed Consensus for Reliability
    Atomic broadcast is a distributed systems primitive in which messages are received reliably and in the same order by all participants. This is an incredibly ...
  21. [21]
  22. [22]
    [PDF] Total Order Broadcast and Multicast Algorithms: Taxonomy and Survey
    Total order broadcast and multicast (also called atomic broadcast/multicast) present an important problem in distributed systems, especially with respect to ...
  23. [23]
    [PDF] Acuerdo: Fast Atomic Broadcast over RDMA
    Sep 1, 2022 · Acuerdo is a new protocol that improves atomic broadcast performance by using RDMA, which bypasses the CPU, and uses one-side RDMA writes.
  24. [24]
    [PDF] RamCast: RDMA-based Atomic Multicast
    Dec 6, 2021 · We consider Mu in our evaluation since it is the best-performing RDMA-based atomic broadcast protocol. Kernel Paxos [21] is a Multi-Paxos ...
  25. [25]
    [PDF] APUS: Fast and Scalable Paxos on RDMA - HKU CS
    APUS is an RDMA-based Paxos protocol designed to be fast and scalable, using fast RDMA to replicate requests concurrently.
  26. [26]
    [PDF] Viewstamped Replication Revisited - MIT
    This paper presents an updated version of Viewstamped. Replication, a replication technique that handles failures in which nodes crash.
  27. [27]
    [PDF] Vertical Atomic Broadcast and Passive Replication (Extended Version)
    Aug 16, 2024 · Atomic broadcast is a reliable communication abstraction ensuring that all processes deliver the same set of messages in a common global ...
  28. [28]
    [PDF] slim-abc: an optimized atomic broadcast protocol - arXiv
    Oct 5, 2024 · Traditional BA protocols in synchronous and partially synchronous networks often rely on leader-based approaches, which can suffer from high ...
  29. [29]
    etcd3 API | etcd
    Apr 6, 2024 · A transaction is an atomic If/Then/Else construct over the key-value store. It provides a primitive for grouping requests together in atomic ...Key-Value API · Key-Value pair · Transaction · Watch API
  30. [30]
    [PDF] Elastic Paxos: A Dynamic Atomic Multicast Protocol
    Paxos and other atomic broadcast algorithms ensure total order of commands within one partition (e.g., get and put commands), consistent cross-partition ...Missing: implications | Show results with:implications
  31. [31]
    A Comprehensive Study on Failure Detectors of Distributed Systems
    Jul 26, 2020 · In distributed systems, failure detectors are used to monitor the processes and to reduce the risk of failures by detecting them before ...
  32. [32]
    None
    ### Summary of Reliable, Atomic, and Causal Broadcast Definitions and Differences
  33. [33]
    [PDF] Distributed Systems Group Membership and View Synchronous ...
    and reliable broadcast. • View synchronous broadcast ensures that the delivery of messages is coordinated with the installation of views. View Synchrony. Page ...
  34. [34]
    [PDF] Optimistic Atomic Multicast - USI
    Atomic multicast addresses the (lack of ) scalability of atomic broadcast. While atomic broadcast propagates messages to all system mem- bers, atomic multicast ...
  35. [35]
    [PDF] Early-Delivery Dynamic Atomic Broadcast
    May 20, 2002 · Processes joining (or voluntarily leaving) the service coordinate their own join (or leave) by selecting a join-slot (or leave-slot) and ...
  36. [36]
    [PDF] Signature-Free Atomic Broadcast with Optimal O(n2) Messages and ...
    We present SQ, the first IT-secure and signature-free asyn- chronous ABC protocol that achieves optimal resilience, 𝑂(𝑛2) messages, and 𝑂(1) expected time (Sec ...
  37. [37]
    [PDF] Parsimonious Asynchronous Byzantine-Fault-Tolerant Atomic ...
    Aug 19, 2005 · We address the problem of asyn- chronous atomic broadcast when up to t < n/3 parties may exhibit Byzantine behavior.Missing: shift | Show results with:shift
  38. [38]
    [PDF] A Modular Approach to Fault-Tolerant Broadcasts and Related ...
    Each broadcast type T has a Uniform counterpart obtained by replacing every one of the properties of T (except Validity) with the corresponding uniform version.