Traces Synchronization in Distributed Networks

Clément, Eric; Dagenais, Michel R.

doi:10.1155/2009/190579

Cited by 5 publications

(5 citation statements)

References 13 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…For commodity computers, the NTP 26 ( Network Time Protocol ) is used by operating systems to have a correct clock, but with a coarse‐grain precision: only around 200

\mu

s on local networks. In research about distributed systems, for instance, Clément and Dagenais synchronize 27 event timestamps to trace events in OS kernel during distributed executions.…”

Section: Precise Distributed Tracesmentioning

confidence: 99%

Tracing task‐based runtime systems: Feedbacks from theStarPUcase

Denis,

Jeannot,

Swartvagher

et al. 2023

Concurrency and Computation

View full text Add to dashboard Cite

SummaryGiven the complexity of current supercomputers and applications, being able to trace application executions to understand their behavior is not a luxury. As constraints, tracing systems have to be as little intrusive as possible in the application code and performances, and be precise enough in the collected data. In this article, we present how works the tracing system used by the task‐based runtime system StarPU. We study the different sources of performance overhead coming from the tracing system and how to reduce these overheads. Then, we evaluate the accuracy of distributed traces with different clock synchronization techniques. Finally, we summarize our experiments and conclusions with the lessons we learned to efficiently trace applications, and the list of characteristics each tracing system should feature to be competitive. The reported experiments and implementation details comprise a feedback of integrating into a task‐based runtime system state‐of‐the‐art techniques to efficiently and precisely trace application executions. We highlight the points every application developer or end‐user should be aware of to seamlessly integrate a tracing system or just trace application executions.

show abstract

“…For commodity computers, the NTP 26 ( Network Time Protocol ) is used by operating systems to have a correct clock, but with a coarse‐grain precision: only around 200

\mu

s on local networks. In research about distributed systems, for instance, Clément and Dagenais synchronize 27 event timestamps to trace events in OS kernel during distributed executions.…”

Section: Precise Distributed Tracesmentioning

confidence: 99%

Tracing task‐based runtime systems: Feedbacks from theStarPUcase

Denis,

Jeannot,

Swartvagher

et al. 2023

Concurrency and Computation

View full text Add to dashboard Cite

show abstract

“…For any node in MST , the shortest paths to every other node are computed. The best‐time RN is the node with the shortest paths to all the nodes . Eventually, all the nodes in the network synchronise their time with the RN through those paths.…”

Section: Related Workmentioning

confidence: 99%

“…Clement and Dagenais proposed offline trace synchronisation in distributed systems. When tracing is finished, individual traces are collected and analysed together.…”

Section: Related Workmentioning

confidence: 99%

“…Because trace timestamps are locally generated on each node, some form of time synchronisation is required to be able to compare the timestamps from the traces on the different nodes . Then, we also need to select a time reference node (RN), for computing and displaying a global timeline of the events and interactions between nodes .…”

Section: Introductionmentioning

confidence: 99%

“…The links with the best accuracy (lowest error) are selected and form a minimum spanning tree (MST) . The remaining challenge is to find the optimal RN, given the MST of accurate links . The selection criteria for the RN is to insure that the total time conversion error should be minimised, to optimise the accuracy of the global analysis that compares timestamps from events on every node.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Reference node selection in dynamic tree

Jabbarifar

Dagenais

2014

Int J Network Mgmt

Self Cite

View full text Add to dashboard Cite

SUMMARYThe reference node (RN) is a central node that has minimum distance/hop count to all other nodes in the network. This central node can play several critical roles such as being the time reference in order to synchronise computer nodes. For synchronisation, the main goal is to minimise the sum of synchronisation errors. The time synchronisation error, known for each link between two nodes, accumulates for each hop along the path used for synchronisation between two nodes. In such a context, the best RN is defined as having the minimal sum of time synchronisation errors between itself and every other node. Thus, the first step for error minimisation is to select a minimum spanning tree (MST), formed by the links with minimum synchronisation error, as synchronisation path. The second step is to select an RN, which minimises the sum of synchronisation errors to all nodes in the MST, as time reference for synchronisation. In a dynamic network, where communication links appear and disappear, and synchronisation accuracy improves as more packets are exchanged, a static RN would entail suboptimal synchronisation accuracy. All existing models in this area are limited to static RNs because of the computing cost of updating the RN, yielding a suboptimal total synchronisation error over time and causing problems if the selected node is removed from the dynamic network. This paper presents a novel and efficient method for dynamic RN selection in dynamic networks. The approach proposed in this paper improves the performance of RN computation and update in live mode for dynamic networks. This new method concentrates on the altered path with respect to the RN, each time the MST is updated. This provides an efficient way to find and maintain a RN incrementally in an average time complexity of O(log n) per update, which n is the total number of nodes in the network. The proposed approach was tested with a huge dynamic network containing 60 000 simulated nodes, in a number of different situations. The proposed approach achieves excellent running time while minimising synchronisation error. Although this work is currently used for time synchronisation purposes, several dynamic network tools can benefit from an efficient incremental algorithm to calculate hop counts and select a central point for the network.

show abstract