Providing strong fault-tolerant guarantees for the modern cloud is difficult, as application developers must coordinate between independent stateful services and ephemeral compute and handle various failure-induced an...
详细信息
Providing strong fault-tolerant guarantees for the modern cloud is difficult, as application developers must coordinate between independent stateful services and ephemeral compute and handle various failure-induced anomalies. We propose Composable Resilient Steps (CReSt), a new abstraction for resilient cloud applications. CReSt uses fault-tolerant steps as its core building block, which allows participants to receive, process, and send messages as a single uninterruptible atomic unit. Composability and reliability are orthogonally achieved by reusable CReSt implementations, for example, leveraging reliable message queues. Thus, CReSt application builders focus solely on translating application logic into steps, and infrastructure builders focus on efficient CReSt implementations. We propose one such implementation called DARQ (for Deduplicated Asynchronously Recoverable Queues). At its core, DARQ is a storage service that encapsulates CReSt participant state and enforces CReSt semantics; developers attach ephemeral compute nodes to DARQ instances to implement stateful distributed components. Services built with DARQ are resilient by construction, and CReSt-compatible services naturally compose without loss of resilience. For performance, we propose a novel speculative execution scheme to execute CReSt steps without waiting for message persistence in DARQ, effectively eliding cloud persistence overheads; our scheme maintains CReSt's fault-tolerance guarantees and automatically restores to a consistent system state upon failure. We showcase the generality of CReSt and DARQ using two applications: cloud streaming and workflow processing. Experiments show that DARQ is able to achieve extremely low latency and high throughput across these use cases, often beating state-of-the-art customized solutions.
A distributed database system often operates in an asynchronous communication model where messages can be arbitrarily delayed. This communication model causes nondeterministic effects like unpredictable arrival orders...
详细信息
A distributed database system often operates in an asynchronous communication model where messages can be arbitrarily delayed. This communication model causes nondeterministic effects like unpredictable arrival orders of messages. Nonetheless, in general we want the distributed system to be deterministic;the system should produce the same output despite the nondeterministic effects on messages. Previously, two interpretations of determinism have been proposed. The first says that all infinite fair computation traces produce the same output. The second interpretation is a confluence notion, saying that all finite computation traces can still be extended to produce the same output. A decidability result for the confluence notion was previously obtained for so-called simple transducer networks, a model from the field of declarative networking. In the current article, we also present a decidability result for simple transducer networks, but this time for the first interpretation of determinism, with infinite fair computation traces. We also compare the expressivity of simple transducer networks under both interpretations.
The CALM-conjecture, first stated by Hellerstein [2010] and proved in its revised form by Ameloot et al. [2013] within the framework of relational transducer networks, asserts that a query has a coordination-free exec...
详细信息
The CALM-conjecture, first stated by Hellerstein [2010] and proved in its revised form by Ameloot et al. [2013] within the framework of relational transducer networks, asserts that a query has a coordination-free execution strategy if and only if the query is monotone. Zinn et al. [2012] extended the framework of relational transducer networks to allow for specific data distribution strategies and showed that the nonmonotone win-move query is coordination-free for domain-guided data distributions. In this article, we extend the story by equating increasingly larger classes of coordination-free computations with increasingly weaker forms of monotonicity and present explicit Datalog variants that capture each of these classes. One such fragment is based on stratified Datalog where rules are required to be connected with the exception of the last stratum. In addition, we characterize coordination-freeness as those computations that do not require knowledge about all other nodes in the network, and therefore, can not globally coordinate. The results in this article can be interpreted as a more fine-grained answer to the CALM-conjecture.
Motivated by a recent conjecture concerning the expressiveness of declarative networking, we propose a formal computation model for "eventually consistent" distributed querying, based on relational transduce...
详细信息
Motivated by a recent conjecture concerning the expressiveness of declarative networking, we propose a formal computation model for "eventually consistent" distributed querying, based on relational transducers. A tight link has been conjectured between coordination-freeness of computations, and monotonicity of the queries expressed by such computations. Indeed, we propose a formal definition of coordination-freeness and confirm that the class of monotone queries is captured by coordination-free transducer networks. Coordination-freeness is a semantic property, but the syntactic class of "oblivious" transducers we define also captures the same class of monotone queries. Transducer networks that are not coordination-free are much more powerful.
To enhance the retrieval efficiency, a User-centered cloud-based script programming Environment is proposed (UPE). This environment consist three main components. Firstly, we designed a set of primitives of car image ...
详细信息
ISBN:
(纸本)9783319155548;9783319155531
To enhance the retrieval efficiency, a User-centered cloud-based script programming Environment is proposed (UPE). This environment consist three main components. Firstly, we designed a set of primitives of car image retrieval and implemented in Julia, a high performance scripting language. The feature extracting algorithms (SIFT, Scale-Invariant Feature Transform) and feature matching were implemented on multi-/many-core platform. Secondly, we constructed a function library to auto-adapt heterogeneous platform. Users orchestrated primitives in Julia script to construct their applications and debug/run it in UPE. Finally, this UPE is coded in Java to edit, debug, and run Julia codes. UPE help users passing codes as a message onto a remote supercomputer. Users needed not concern themselves to the supercomputer architecture details. The experimental results show that the whole retrieval process is speed up 26.35x, and the accuracy enhanced to 33.63% with a 100% recall ratio. The performance loss of Julia is about 2.65% compared with C implementation.
The CALM-conjecture, first stated by Hellerstein [23] and proved in its revised form by Ameloot et al. [13] within the framework of relational transducer networks, asserts that a query has a coordination-free executio...
详细信息
ISBN:
(纸本)9781450323758
The CALM-conjecture, first stated by Hellerstein [23] and proved in its revised form by Ameloot et al. [13] within the framework of relational transducer networks, asserts that a query has a coordination-free execution strategy if and only if the query is monotone. Zinn et al. [32] extended the framework of relational transducer networks to allow for specific data distribution strategies and showed that the non monotone win-move query is coordination-free for domain guided data distributions. In this paper, we complete the story by equating increasingly larger classes of coordination free computations with increasingly weaker forms of mono tonicity and make Datalog variants explicit that capture each of these classes. One such fragment is based on stratified Datalog where rules are required to be connected with the exception of the last stratum. In addition, we characterize coordination-freeness as those computations that do not require knowledge about all other nodes in the network, and therefore, can not globally coordinate. The results in this paper can be interpreted as a more fine-grained answer to the CALM-conjecture.
programming models for cloud computing has been a research focus recently. Current programming models for cloud computing mainly focus on improving the efficiency of the cloud computing platforms but little has been d...
详细信息
ISBN:
(纸本)3642128130
programming models for cloud computing has been a research focus recently. Current programming models for cloud computing mainly focus on improving the efficiency of the cloud computing platforms but little has been done on the performance predictability of models. This paper presents a cloud computing programming model, called BSPcloud. The BSPcloud has the advantages of predictable performance. BSPcloud uses a hybrid of distributed-memory and shared-memory bulk synchronous parallel (BSP) programming model. Computing tasks are first divided into a set of coarse granularity bulks which are computed by the distributed-memory BSP model, and each coarse granularity bulk is further divided into a set of bulk threads which are computed by the shared-memory BSP model. BSPcloud makes full use of the multi-core architecture, and more importantly, the performance of the BSPcloud is predictable. We have implemented a proof-of-concept BSPcloud parallel programming library in java. We give the implementation of BSPcloud parallel programming library and show how the library implements hybrid programming. We detail applying the BSPcloud library on matrix multiplication. The performance predictability and speedup are evaluated in the cloud platform. We give the results of experiments.
暂无评论