Closed form bounds for clock synchronization under simple uncertainty assumptions

Biaz, Saad; Welch, Jennifer L.

doi:10.1016/s0020-0190(01)00151-x

Cited by 38 publications

(56 citation statements)

References 2 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Note that the figure shows that PulseSync features a clearly sublinear dependence of the skew on D, as predicted by our analysis. This justifies the choice of a more optimistic system model, as it is known for long that the skew must be linear in D in the worst case [2]. These findings are not artifacts of unrealistic simulation parameters or network diameters; careful experiments confirm considerable improvements already for 5-10 hops.…”

Section: Introductionmentioning

confidence: 73%

“…This bound is probabilistic, whereas the worstcase accuracy that can be achieved is no smaller than D/2 in a general model where message delays vary arbitrarily in the range [0, 1] and clocks are perfect [2]. A worst-case upper bound of (1 + ρ)D can be achieved by a simple algorithm even if in addition clock speeds maybe vary arbitrarily between 1−ρ and 1+ρ [11].…”

Section: Related Workmentioning

confidence: 97%

See 1 more Smart Citation

PulseSync: An Efficient and Scalable Clock Synchronization Protocol

Lenzen

Sommer

Wattenhofer

2015

IEEE/ACM Trans. Networking

108

View full text Add to dashboard Cite

Abstract-Clock synchronization is an enabling service for a wide range of applications and protocols in both wired and wireless networks. We study the implications of clock drift and communication latency on the accuracy of clock synchronization when scaling the network diameter. Starting with a theoretical analysis of synchronization protocols, we prove tight bounds on the synchronization error in a model that assumes independently and randomly distributed communication delays and slowly changing drifts. While this model is more optimistic than traditional worst-case analysis, it much better captures the nature of real-world systems such as wireless networks.The bound on the synchronization accuracy, which is roughly the square-root of the network diameter, is achieved by the novel PulseSync protocol. Extensive experiments demonstrate that PulseSync is able to meet the predictions from theory and tightly synchronizes large networks. This contrasts against an exponential growth of the skew incurred by the state-of-the-art protocol for wireless sensor networks. Moreover, PulseSync adapts much faster to network dynamics and changing clock drifts than this protocol.

show abstract

Section: Introductionmentioning

confidence: 73%

Section: Related Workmentioning

confidence: 97%

PulseSync: An Efficient and Scalable Clock Synchronization Protocol

Lenzen

Sommer

Wattenhofer

2015

IEEE/ACM Trans. Networking

108

View full text Add to dashboard Cite

show abstract

“…In clock synchronization research [13,6,20,4,7,14], system models are considered where the uncertainty comes from varying message delays, failures, and drifting clocks. Denoted "Partially Synchronous Reliable/Unreliable Models" in [23], such models are nowadays called (non-lockstep) synchronous models in literature.…”

Section: Classic Computing Modelmentioning

confidence: 99%

Optimal Clock Synchronization Revisited: Upper and Lower Bounds in Real-Time Systems

Moser

Schmid

2006

Lecture Notes in Computer Science

View full text Add to dashboard Cite

MotivationExecutions of distributed algorithms are typically modeled as sequences of atomic computing steps that are executed in zero time. With this assumption, it does not make a difference, for example, whether messages arrive at a processor simultaneously or nicely staggered in time: The messages are processed instantaneously when they arrive. The zero step-time abstraction is hence very convenient for analysis, and a wealth of distributed algorithms, impossibility results and lower bounds have been developed for models that employ this assumption [15].In real systems, however, computing steps are neither instantaneous nor arbitrarily preemptable: A computing step triggered by a message arriving in the middle of the execution of some other computing step is usually delayed until the current computation is finished. This results in queueing phenomenons, which depend not only on the actual message arrival pattern but also on the queueing/scheduling discipline employed. The real-time systems community has established powerful techniques for analyzing such effects [22], such that the resulting worst-case response times and end-to-end delays can be computed.This paper introduces a real-time distributed computing model for message-passing systems, which reconciles the distributed computing and the real-time systems perspective: By just replacing the zero step-time assumption with non-zero step times, we obtain a real-time distributed computing model that admits real-time analysis without invalidating standard distributed computing analysis techniques and results: We show that a system adhering to the real-time model can simulate a system that adheres to the classic model and vice versa.

show abstract

“…First, it requires synchronized clocks. Highspeed networks like SAN may not increase the synchronization accuracy dramatically for atleast two reasons: synchronization accuracy is bound by message delay variance and not by the absolute delay ( [14], [4]); also, clock synchronization messages -being few tens of bytes long, may not see significant latency reduction even in gigabit networks. However, highspeed networks will need to handle higher number of active transactions in a given time-slice and hence require better clocks.…”

Section: Concurrency Controlmentioning

confidence: 99%

“…It is understood that different ordering mechanisms -FIFO, casual, causally constrained total order and unconstrained total order 4 -guarantee different consistency semantics with cost and strictness increasing in that order. iSAN, being storage-consumer aware, deploys the suitable message ordering that suffices the storage-consumer's requirements.…”

Section: Concurrency Controlmentioning

confidence: 99%

iSAN – An Intelligent Storage Area Network Architecture

Narayan

Gopinath

2004

Lecture Notes in Computer Science

View full text Add to dashboard Cite

Abstract. This paper describes the motivation, architecture and implementation of iSAN, an "intelligent" storage area network. The main contributions of this work are: (1) how to architect an intelligent SAN that understands the storageconsumers 1 to serve them better; (2) how to realise this abstract architecture using existing technologies; and (3) to demonstrate the benefits that would accrue from such an intelligent SAN. Our results show that the iSAN approach has important benefits when compared with conventional SANs: iSAN facilitates storage sharing, is secure and offers better throughput. In this paper, we discuss two case studies as well as discuss how iSAN approach has been generic enough to capture a wide range of other requirements of SANs.

show abstract

Closed form bounds for clock synchronization under simple uncertainty assumptions

Cited by 38 publications

References 2 publications

PulseSync: An Efficient and Scalable Clock Synchronization Protocol

PulseSync: An Efficient and Scalable Clock Synchronization Protocol

Optimal Clock Synchronization Revisited: Upper and Lower Bounds in Real-Time Systems

iSAN – An Intelligent Storage Area Network Architecture

Contact Info

Product

Resources

About