Search

In this paper, we present the design of our LogPlayer that is a component responsible for fault-tolerant delivery of transactional mutations recorded on a WAL to the backend storage shards. LogPlayer relies on gRPC for asynchronous streaming. However, the design provided in this paper can be used with other asynchronous streaming platforms. We model check the correctness of LogPlayer by TLA+. In particular, our TLA+ specification shows that LogPlayer guarantees in-order exactly-once delivery of WAL entries to the storage shards, even in the presence of shards or LogPlayer failures. Our experiments show LogPlayer is capable of efficient delivery with sub-millisecond latency, and it is significantly more efficient than Apache Kafka for designing a WAL system with exactly-once guarantee., Comment: 12 pages, 13 figures

Causal consistency is an intermediate consistency model that can be achieved together with high availability and performance requirements even in presence of network partitions. In the context of partitioned data stores, it has been shown that implicit dependency tracking using timestamps is more efficient than explicit dependency tracking. Existing time-based solutions depend on monotonic psychical clocks that are closely synchronized. These requirements make current protocols vulnerable to clock anomalies. In this paper, we propose a new time-based algorithm, CausalSpartanX, that instead of physical clocks, utilizes Hybrid Logical Clocks (HLCs). We show that using HLCs, without any overhead, we make the system robust on physical clock anomalies. This improvement is more significant in the context of query amplification, where a single query results in multiple GET/PUT operations. We also show that CausalSpartanX decreases the visibility latency for a given data item compared with existing time-based approaches. In turn, this reduces the completion time of collaborative applications where two clients accessing two different replicas edit same items of the data store. CausalSpartanX also provides causally consistent distributed read-only transactions. CausalSpartanX read-only transactions are non-blocking and require only one round of communication between the client and the servers. Also, the slowdowns of partitions that are unrelated to a transaction do not affect the performance of the transaction. Like previous protocols, CausalSpartanX assumes that a given client does not access more than one replica. We show that in presence of network partitions, this assumption (made in several other works) is essential if one were to provide causal consistency as well as immediate availability to local updates.

Eventual consistency is a popular consistency model for geo-replicated data stores. Although eventual consistency provides high performance and availability, it can cause anomalies that make programming complex for application developers. Session guarantees can remove some of these anomalies while causing much lower overhead compared with stronger consistency models. In this paper, we provide a protocol for providing session guarantees for NuKV, a key-value store developed for services with very high availability and performance requirements at eBay. NuKV relies on the Raft protocol for replication inside datacenters, and uses eventual consistency for replication among datacenters. We provide modified versions of conventional session guarantees to avoid the problem of slowdown cascades in systems with large numbers of partitions. We also use Hybrid Logical Clocks to eliminate the need for delaying write operations to satisfy session guarantees. Our experiments show that our protocol provides session guarantees with a negligible overhead when compared with eventual consistency.

Causal consistency for key-value stores has two main requirements (1) do not make a version visible if some of its dependencies are invisible as it may violate causal consistency in the future and (2) make a version visible as soon as possible so that clients have the most recent information (to the extent feasible). These two requirements conflict with each other. Existing key-value stores that provide causal consistency (or detection of causal violation) utilize a static approach in the trade-off between these requirements. Depending upon the choice, it assists some applications and penalizes some applications. We propose an alternative where the system provides a set of tracking groups and checking groups. This allows the application to choose the settings that are most suitable for that application. Furthermore, these groups can be dynamically changed based on application requirements.

We present our framework DKVF that enables one to quickly prototype and evaluate new protocols for key-value stores and compare them with existing protocols based on selected benchmarks. Due to limitations of CAP theorem, new protocols must be developed that achieve the desired trade-off between consistency and availability for the given application at hand. Hence, both academic and industrial communities focus on developing new protocols that identify a different (and hopefully better in one or more aspect) point on this trade-off curve. While these protocols are often based on a simple intuition, evaluating them to ensure that they indeed provide increased availability, consistency, or performance is a tedious task. Our framework, DKVF, enables one to quickly prototype a new protocol as well as identify how it performs compared to existing protocols for pre-specified benchmarks. Our framework relies on YCSB (Yahoo! Cloud Servicing Benchmark) for benchmarking. We demonstrate DKVF by implementing four existing protocols --eventual consistency, COPS, GentleRain and CausalSpartan-- with it. We compare the performance of these protocols against different loading conditions. We find that the performance is similar to our implementation of these protocols from scratch. And, the comparison of these protocols is consistent with what has been reported in the literature. Moreover, implementation of these protocols was much more natural as we only needed to translate the pseudocode into Java (and add the necessary error handling). Hence, it was possible to achieve this in just 1-2 days per protocol. Finally, our framework is extensible. It is possible to replace individual components in the framework (e.g., the storage component).

Causal consistency is in an intermediate consistency model that can be achieved together with high availability and high performance requirements even in presence of network partitions. There are several proposals in the literature for causally consistent data stores. Thanks to the use of single scalar physical clocks, GentleRain has a throughput higher than other proposals such as COPS or Orbe. However, both of its correctness and performance relay on monotonic synchronized physical clocks. Specifically, if physical clocks go backward its correctness is violated. In addition, GentleRain is sensitive on the clock synchronization, and clock skew may slow write operations in GenlteRain. In this paper, we want to solve this issue in GenlteRain by using Hybrid Logical Clock (HLC) instead of physical clocks. Using HLC, GentleRain protocl is not sensitive on the clock skew anymore. In addition, even if clocks go backward, the correctness of the system is not violated. Furthermore, by HLC, we timestamp versions with a clock very close to the physical clocks. Thus, we can take causally consistency snapshot of the system at any give physical time. We call GentleRain protocol with HLCs GentleRain+. We have implemented GentleRain+ protocol, and have evaluated it experimentally. GentleRain+ provides faster write operations compare to GentleRain that rely solely on physical clocks to achieve causal consistency. We have also shown that using HLC instead of physical clock does not have any overhead. Thus, it makes GentleRain more robust on clock anomalies at no cost.

This paper focuses on the problem of consistency in distributed data stores.We define strong consistency model which provides a simple semantics for application programmers, but impossible to achieve with availability and partition-tolerance. We also define weaker consistency models including causal and eventual consistency. We review COPS and GentleRain as two causally consistent data stores as well as Dynamo as an eventually consistent data store. We provide insights about scenarios where each of these methods is suitable, and some future research directions., Comment: 12 pages

We focus on the problem of adding fault-tolerance to an existing concurrent protocol in the presence of {\em unchangeable environment actions}. Such unchangeable actions occur in practice due to several reasons. One instance includes the case where only a subset of the components/processes can be revised and other components/processes must be as is. Another instance includes cyber-physical systems where revising physical components may be undesirable or impossible. These actions differ from faults in that they are simultaneously {\em assistive} and {\em disruptive}, whereas faults are only disruptive. For example, if these actions are a part of a physical component, their execution is essential for the normal operation of the system. However, they can potentially disrupt actions taken by other components for dealing with faults. Also, one can typically assume that fault actions will stop for a long enough time for the program to make progress. Such an assumption is impossible in this context. We present algorithms for adding stabilizing fault-tolerance, failsafe fault-tolerance and masking fault-tolerance. Interestingly, we observe that the previous approaches for adding stabilizing fault-tolerance and masking fault-tolerance cannot be easily extended in this context. However, we find that the overall complexity of adding these levels of fault-tolerance remains in P (in the state space of the program). We also demonstrate that our algorithms are sound and complete.

We focus on a protocol for auditable restoration of distributed systems. The need for such protocol arises due to conflicting requirements (e.g., access to the system should be restricted but emergency access should be provided). One can design such systems with a tamper detection approach (based on the intuition of "break the glass door"). However, in a distributed system, such tampering, which are denoted as auditable events, is visible only for a single node. This is unacceptable since the actions they take in these situations can be different than those in the normal mode. Moreover, eventually, the auditable event needs to be cleared so that system resumes the normal operation. With this motivation, in this paper, we present a protocol for auditable restoration, where any process can potentially identify an auditable event. Whenever a new auditable event occurs, the system must reach an "auditable state" where every process is aware of the auditable event. Only after the system reaches an auditable state, it can begin the operation of restoration. Although any process can observe an auditable event, we require that only "authorized" processes can begin the task of restoration. Moreover, these processes can begin the restoration only when the system is in an auditable state. Our protocol is self-stabilizing and has bounded state space. It can effectively handle the case where faults or auditable events occur during the restoration protocol. Moreover, it can be used to provide auditable restoration to other distributed protocol., Comment: 10 pages