Abstract
Embedding CRDT counters has shown to be a challenging topic, since their introduction in Riak Maps. The desire for obliviousness, where all information about a counter is fully removed upon key removal, faces problems due to the possibility of concurrency between increments and key removals. Previous state-based proposals exhibit undesirable reset-wins semantics, which lead to losing increments, unsatisfactorily solved through manual generation management in the API. Previous operation-based approaches depend on causal stability, being prone to unbounded counter growth under network partitions. We introduce a novel embeddable operation-based CRDT counter which achieves both desirable observed-reset semantics and obliviousness upon resets. Moreover, it achieves this while merely requiring FIFO delivery, allowing a tradeoff between causal consistency and faster information propagation, being more robust under network partitions.

Abstract
TCP is typically the default transport protocol of choice for its supposed reliability, even for message-oriented middleware (e.g., ZeroMQ) or inter-actor communication (e.g., distributed Erlang). However, under network issues, TCP connections can fail, which requires ensuring both at-least-once and at-most-once delivery at the upper middleware layer. Moreover, the use of TCP at scale, in highly concurrent systems, can lead to drastic performance loss due to the need for TCP connection multiplexing and the resulting head-of-line blocking. This paper introduces Exon, an oblivious exactly-once messaging protocol, and a corresponding lightweight library implementation. Exon uses a novel strategy of a per-message four-way protocol to ensure oblivious exactly-once messaging, with on-demand protocol-level soft half-connections that are established when needed and safely discarded. This achieves correctness, obliviousness, and performance, through merging and pipelining basic protocol messages. The empirical evaluation of Exon demonstrates significant improvements in throughput and latency under packet loss, while maintaining a negligible overhead over TCP in healthy networks.

Abstract
TCP is the default transport protocol of choice, namely for message-oriented middleware protocols (e.g., ZMTP, AMQP, MQTT) or distributed language runtimes (e.g., distributed Erlang), where exactly-once (EO) messaging is paramount. However, EO is only guaranteed within the TCP session, since reality shows that TCP connections can fail under many circumstances. Ensuring EO delivery ends up at the middleware layer, at the cost of higher complexity and lack of obliviouness - due to the use of permanent per-peer state. Moreover, using TCP at scale in highly concurrent systems leads to the need for TCP connection multiplexing, and possibly drastic performance loss due to head-of-line blocking. This paper introduces Exon, an oblivious exactly-once messaging protocol, and a corresponding lightweight (requiring no persistent storage, minimal memory, and low computation) library implementation over UDP. Exon uses a novel strategy of a per-message four-way protocol to ensure oblivious exactly-once messaging, with on-demand protocol-level "soft half-connections", established when needed and safely discarded. Obliviousness here refers to the protocol's ability to discard connection-specific state between incarnations, although some global information is retained. Exon achieves simultaneously: correctness with no timing assumptions, obliviousness, and performance through merging and pipelining basic protocol messages. Exon also employs a reliable delegation technique to handover the sending responsibility to a mediating node, without violating EO, when the sender the receiver are directly unreachable to each other and even if the message had already been delivered. The empirical evaluation of Exon demonstrates significant improvements (40%) over TCP in throughput and latency under packet loss, while maintaining a negligible (8%) overhead in healthy networks.

Abstract

Abstract

Abstract
Conflict-free Replicated Data Types (CRDTs) allow optimistic replication in a principled way. Different replicas can proceed independently, being available even under network partitions and always converging deterministically: Replicas that have received the same updates will have equivalent state, even if received in different orders. After a historical tour of the evolution from sequential data types to CRDTs, we present in detail the two main approaches to CRDTs, operation-based and state-based, including two important variations, the pure operation-based and the delta-state based. Intended for prospective CRDT researchers and designers, this article provides solid coverage of the essential concepts, clarifying some misconceptions that frequently occur, but also presents some novel insights gained from considerable experience in designing both specific CRDTs and approaches to CRDTs.