2022
Trigeorgi, Andria; Nicolaou, Nicolas; Georgiou, Chryssis; Hadjistasi, Theophanis; Stavrakis, Efstathios; Cadambe, Viveck; Urgaonkar, Bhuvan
Invited Paper: Towards Practical Atomic Distributed Shared Memory: An Experimental Evaluation Inproceedings
In: Devismes, Stéphane; Petit, Franck; Altisen, Karine; Luna, Giuseppe Antonio Di; Anta, Antonio Fernandez (Ed.): Stabilization, Safety, and Security of Distributed Systems, pp. 35–50, Springer International Publishing, Cham, 2022, ISBN: 978-3-031-21017-4.
@inproceedings{trigeorgi_invited_2022,
title = {Invited Paper: Towards Practical Atomic Distributed Shared Memory: An Experimental Evaluation},
author = {Andria Trigeorgi and Nicolas Nicolaou and Chryssis Georgiou and Theophanis Hadjistasi and Efstathios Stavrakis and Viveck Cadambe and Bhuvan Urgaonkar},
editor = {Stéphane Devismes and Franck Petit and Karine Altisen and Giuseppe Antonio Di Luna and Antonio Fernandez Anta},
isbn = {978-3-031-21017-4},
year = {2022},
date = {2022-01-01},
booktitle = {Stabilization, Safety, and Security of Distributed Systems},
pages = {35--50},
publisher = {Springer International Publishing},
address = {Cham},
abstract = {Distributed Shared Storage Services may serve as building blocks to yield complex, decentralized, cloud applications in emerging technologies (e.g., IoT, VR/AR), as they offer a transparent cloud storage space where distributed applications can store, retrieve, and coordinate over shared data. Ideally, distributed applications would like to communicate through a “cloud” memory layer that may provide similar guarantees as a centralized sequential memory. Atomic Distributed Shared Memory (ADSM) provides the illusion of a sequential memory space despite asynchrony, network perturbations, and device failures. A plethora of algorithmic solutions along with proven correctness guarantees have been proposed to provide ADSM in a message passing system. None of them, however, has been adopted in a real working solution: commercial solutions avoid the use of ADSM algorithms, mainly due to their communication overhead. But what is exactly the performance overhead of an ADSM algorithm over existing commercial solutions? In this work we want to provide a first answer to this question by performing an in-depth experimental comparison of the state-of-the-art dynamic ADSM algorithm ARES, with two well-established open-source distributed storage solutions, Cassandra and Redis. The results show that ARES's performance is comparable with the commercial systems, with respect to scalability, object size and throughput.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Distributed Shared Storage Services may serve as building blocks to yield complex, decentralized, cloud applications in emerging technologies (e.g., IoT, VR/AR), as they offer a transparent cloud storage space where distributed applications can store, retrieve, and coordinate over shared data. Ideally, distributed applications would like to communicate through a “cloud” memory layer that may provide similar guarantees as a centralized sequential memory. Atomic Distributed Shared Memory (ADSM) provides the illusion of a sequential memory space despite asynchrony, network perturbations, and device failures. A plethora of algorithmic solutions along with proven correctness guarantees have been proposed to provide ADSM in a message passing system. None of them, however, has been adopted in a real working solution: commercial solutions avoid the use of ADSM algorithms, mainly due to their communication overhead. But what is exactly the performance overhead of an ADSM algorithm over existing commercial solutions? In this work we want to provide a first answer to this question by performing an in-depth experimental comparison of the state-of-the-art dynamic ADSM algorithm ARES, with two well-established open-source distributed storage solutions, Cassandra and Redis. The results show that ARES's performance is comparable with the commercial systems, with respect to scalability, object size and throughput.
2021
Theophanis Hadjistasi Chryssis Georgiou, Nicolas Nicolaou; Schwarzmann, Alexander A.
Implementing Three Exchange Read Operations for Distributed Atomic Storage Journal Article Forthcoming
In: Journal of Parallel and Distributed Computing, Forthcoming.
@article{GHNS21,
title = {Implementing Three Exchange Read Operations for Distributed Atomic Storage},
author = {Chryssis Georgiou, Theophanis Hadjistasi, Nicolas Nicolaou, and Alexander A. Schwarzmann},
year = {2021},
date = {2021-09-01},
journal = {Journal of Parallel and Distributed Computing},
abstract = {Communication latency typically dominates the performance of message-passing systems, and consequently de nes the eciency of operations of algorithms implementing atomic read/write objects in asynchronous, crash-prone, message-passing systems. Here latency is measured in terms of the number of communication exchanges (or simply exchanges) involved in each operation. We present four algorithms, two for the SWMR and two for the MWMR environments, that allow reads to take two or three exchanges, advancing the state-of-the-art in this area. Writes take the same number of exchanges as in prior works (i.e., two for SWMR and four for MWMR settings). In contrast with existing efficient implementations, ours come with no constraints on reader participation in both settings , and on the number of writers in the MWMR setting. Correctness of algorithms is rigorously argued. We conclude with an empirical study demonstrating the practicality of the algorithms, and identifying settings in which their performance is clearly superior compared to relevant algorithms.},
keywords = {},
pubstate = {forthcoming},
tppubtype = {article}
}
Communication latency typically dominates the performance of message-passing systems, and consequently de nes the eciency of operations of algorithms implementing atomic read/write objects in asynchronous, crash-prone, message-passing systems. Here latency is measured in terms of the number of communication exchanges (or simply exchanges) involved in each operation. We present four algorithms, two for the SWMR and two for the MWMR environments, that allow reads to take two or three exchanges, advancing the state-of-the-art in this area. Writes take the same number of exchanges as in prior works (i.e., two for SWMR and four for MWMR settings). In contrast with existing efficient implementations, ours come with no constraints on reader participation in both settings , and on the number of writers in the MWMR setting. Correctness of algorithms is rigorously argued. We conclude with an empirical study demonstrating the practicality of the algorithms, and identifying settings in which their performance is clearly superior compared to relevant algorithms.