#### DMCA

## Fast Randomized Test-and-Set and Renaming

### Cached

### Download Links

Citations: | 14 - 8 self |

### Citations

2170 | Randomized Algorithms
- Motwani, Raghavan
- 1995
(Show Context)
Citation Context ...sses, with high probability. A tempting, yet unsuccessful approach to bound the total number of calls before each test-and-set is accessed once would be to use the well-known coupon collector process =-=[25, 26]-=-, which guarantees that n distinct coupons will be discovered using O(n log n) independent random trials. Note, however, that the strong adversary controls the scheduling of the trials, which causes t... |

1707 | Impossibility of distributed consensus with one faulty process
- Fischer, Lynch, et al.
- 1985
(Show Context)
Citation Context ...d to abandon it. The simple solution would be to run a distributed consensus algorithm to agree on which process owns each name. However, asynchronous, wait-free deterministic consensus is impossible =-=[16]-=-, and the randomized version is inherently expensive, requiring Ω(n 2 ) total steps [7]. – How are processes scheduled? If the processes are scheduled in a synchronous fashion, then resolving contenti... |

1159 | Linearizability: a correctness condition for concurrent objects
- Herlihy, Wing
- 1990
(Show Context)
Citation Context ...onsensus number 2 (see [22] for details). We present an efficient randomized implementation that guarantees the desired properties with probability 1, and is linearizable, following the definition in =-=[23]-=-. Our implementation is adaptive, in that the complexity of an operation depends on the total contention k at the object, and not on n, the total number of processes. 4.1 The RatRace Algorithm The Rat... |

849 | Wait-free synchronization
- Herlihy
- 1991
(Show Context)
Citation Context ...sequential specification is provided in Figure 1. Note that one-shot test-and-set cannot be implemented deterministically wait-free in asynchronous shared memory, since it has consensus number 2 (see =-=[22]-=- for details). We present an efficient randomized implementation that guarantees the desired properties with probability 1, and is linearizable, following the definition in [23]. Our implementation is... |

518 |
Probability and Computing: Randomized Algorithms and Probabilistic Analysis
- Mitzenmacher, Upfal
- 2005
(Show Context)
Citation Context ...sses, with high probability. A tempting, yet unsuccessful approach to bound the total number of calls before each test-and-set is accessed once would be to use the well-known coupon collector process =-=[25, 26]-=-, which guarantees that n distinct coupons will be discovered using O(n log n) independent random trials. Note, however, that the strong adversary controls the scheduling of the trials, which causes t... |

151 | The topological structure of asynchronous computability
- Herlihy, Shavit
- 1999
(Show Context)
Citation Context ...eparation in terms of complexity between randomized tight renaming and randomized consensus in asynchronous shared-memory. The impossibility of wait-free renaming in a namespace smaller than (2n − 1) =-=[4, 5]-=- is circumvented by the use of randomization. There exist infinite length executions of infinitesimal probability weight, in which the algorithms do not terminate. Also, note that our test-and-set imp... |

112 |
Renaming in an asynchronous environment
- Attiya, Bar-Noy, et al.
- 1990
(Show Context)
Citation Context ...ique. The size of the resulting namespace should only depend on n and on t. Note that, in our algorithms, we relax the assumption of unique initial identifiers, made in the original problem statement =-=[6]-=-. We assume t ≤ n − 1, hence our solutions are wait-free. The complexity of our solutions is measured in terms of total steps (reads and writes, including random coin flips). In the following, we say ... |

83 |
Immediate Atomic Snapshots and Fast Renaming
- Borowsky
- 1993
(Show Context)
Citation Context ...for ≤ (2n−2) names for specific parameter values. The complexity of deterministic shared-memory renaming implementations has been an active research topic. Burns and Peterson [17], Borowski and Gafni =-=[18]-=-, Anderson and Moir [2], Moir and Garay [3] were among the first to propose wait-free, one-shot, deterministic algorithms into a namespace of size (2n − 1). These solutions have very high total step c... |

43 | Adaptive mutual exclusion with local spinning
- Anderson, Kim
- 2000
(Show Context)
Citation Context ...in the system. They use k for the maximum number of processes that may participate in an execution (which we denote by n).is similar to the adaptive algorithm for mutual exclusion by Anderson et al. =-=[15]-=-, although the problem and the fault model we analyze are different. We use the twoprocess randomized test-and-set algorithm by Tromp and Vitànyi [16] as a building block. The renaming problem has bee... |

42 |
The ambiguity of choosing
- Burns, Peterson
- 1989
(Show Context)
Citation Context ...renaming may be possible for ≤ (2n−2) names for specific parameter values. The complexity of deterministic shared-memory renaming implementations has been an active research topic. Burns and Peterson =-=[17]-=-, Borowski and Gafni [18], Anderson and Moir [2], Moir and Garay [3] were among the first to propose wait-free, one-shot, deterministic algorithms into a namespace of size (2n − 1). These solutions ha... |

27 | Adaptive and efficient algorithms for lattice agreement and renaming
- Attiya, Fouren
(Show Context)
Citation Context ...ep complexity and the size of the namespace depend only on total contention k, not on the maximum number of participating processes n. The first adaptive algorithm was introduced by Attiya and Fouren =-=[19]-=-. They achieve a namespace of (6k − 1) names, with a total complexity of O(k 2 log k). Afek and Merritt [7] build on the previous algorithm in order to achieve adaptive wait-free (2k−1)-renaming with ... |

24 | Efficient adaptive collect using randomization
- Attiya, Kuhn, et al.
(Show Context)
Citation Context ...omplexity depends on k, the actual number of competitors (not on n, the total number of possible competitors). The algorithm efficiently combines the idea of a randomized splitter tree, first used in =-=[10]-=-, with the tournament tree algorithm by Afek et al. [11]. Our renaming algorithms rely on both the adaptivity and anonymity properties of this implementation. Given the power of test-and-set to simpli... |

22 | Tight bounds for asynchronous randomized consensus
- Attiya, Censor
(Show Context)
Citation Context ...to agree on which process owns each name. However, asynchronous, wait-free deterministic consensus is impossible [8], and the randomized version is inherently expensive, requiring Ω(n 2 ) total steps =-=[9]-=-. In this paper, we present two efficient randomized renaming algorithms for an asynchronous, fault-prone system subject to a strong, adaptive adversary. The key building block for both algorithms is ... |

21 | Randomized naming using wait-free shared variables
- Panconesi, Papatriantafilou, et al.
- 1998
(Show Context)
Citation Context ...with ours. This reference also contains an excellent overview of prior work on renaming.The feasibility of randomized renaming in an asynchronous system has been first considered by Panconesi et al. =-=[21]-=-. They present a wait-free solution that ensures a namespace of size n(1 + ɛ) for ɛ > 0, with expected O(M log 2 n) total step complexity, using only single-writer multiple-reader registers. This solu... |

20 |
wait-free (2k-1)-renaming
- Fast
- 1999
(Show Context)
Citation Context ...rithms that tolerate crash failures [6, 13, 20]. Even loose renaming, where the namespace is of size (2n − 1), can be quite expensive, as the best known solutions require at least Θ(n 3 ) total steps =-=[2, 24]-=-. Yet in practice, most people believe that renaming is relatively easy: simply choose a name at random; if more than one process selects the same name, then try again. Several subtle problems occur w... |

19 | Using local-spin k-exclusion algorithms to improve wait-free object implementations
- Anderson, Moir
- 1997
(Show Context)
Citation Context ... deterministic algorithms into a namespace of size (2n − 1). These solutions have very high total step complexity; for some, the total step complexity is exponential (e.g. [3, 17]). Anderson and Moir =-=[1]-=- propose a variant of renaming that attains a tight namespace of n names using stronger set-first-zero objects. Note that their algorithm could be rephrased using our one-shot test-and-set implementat... |

16 |
long-lived renaming improved and simplified
- Fast
- 1998
(Show Context)
Citation Context ...mespace. For example, nearly every networked device has an ethernet address, and yet the namespace is so large as to reduce the usefulness of such names. Thus, a significant amount of research (e.g., =-=[4, 13, 20, 24, 25]-=-) has analyzed the feasibility and complexity of the renaming problem in a crash-prone distributed system. Unfortunately, renaming in a fault-prone system can be expensive, if not impossible. For exam... |

14 | The complexity of synchronous iterative do-all with crashes - Georgiou, Shvartsman |

13 |
New Combinatorial Topology Upper and Lower Bounds for Renaming
- Castañeda, Rajsbaum
- 2008
(Show Context)
Citation Context ...eparation in terms of complexity between randomized tight renaming and randomized consensus in asynchronous shared-memory. The impossibility of wait-free renaming in a namespace smaller than (2n − 1) =-=[4, 5]-=- is circumvented by the use of randomization. There exist infinite length executions of infinitesimal probability weight, in which the algorithms do not terminate. Also, note that our test-and-set imp... |

12 | long-lived renaming improved and simplified
- Moir, Fast
- 1998
(Show Context)
Citation Context ...lues. The complexity of deterministic shared-memory renaming implementations has been an active research topic. Burns and Peterson [17], Borowski and Gafni [18], Anderson and Moir [2], Moir and Garay =-=[3]-=- were among the first to propose wait-free, one-shot, deterministic algorithms into a namespace of size (2n − 1). These solutions have very high total step complexity; for some, the total step complex... |

12 | Fully-adaptive algorithms for long-lived renaming
- Brodsky, Ellen, et al.
(Show Context)
Citation Context ...e little total memory O(n log(M/n)). In comparison, our algorithms pre-allocate O(n2 ) memory, and use O(n polylog n) total memory, without assuming any bound M on the initial namespace. Ellen et al. =-=[20]-=- analyze the complexity of long-lived adaptive renaming (i.e., processes may release their names) in shared-memory, under various synchrony assumptions. Their asynchronous algorithm ensures Θ(k) overh... |

11 | Randomized two-process wait-free testand-set
- Tromp, Vitányi
(Show Context)
Citation Context ...algorithm for mutual exclusion by Anderson et al. [15], although the problem and the fault model we analyze are different. We use the twoprocess randomized test-and-set algorithm by Tromp and Vitànyi =-=[16]-=- as a building block. The renaming problem has been introduced by Attiya et al. [6]. In the original paper, the authors present a wait-free solution using (2n − 1) names in an asynchronous message-pas... |

11 |
Aspnes and Orli Waarts. Randomized consensus in expected o(n log2 n) operations per processor
- James
- 1996
(Show Context)
Citation Context ...tch. Each two-player match is decided using the randomized two-process test-and-set algorithm of Tromp and Vitànyi [28]. (Alternatively, we could use a randomized consensus algorithm with n = 2, e.g. =-=[5]-=-, although the properties stay the same.) Note that the matches are decided in a wait-free manner, since a process wins automatically if the opponent does not show up. The Backup Grid. The backup grid... |

11 |
waitfree renaming with optimal name space and high throughput
- Long-lived
- 1998
(Show Context)
Citation Context ...m has to take because of adversarial scheduling or crashes. Our algorithm improves significantly on the total step complexity of previous randomized or deterministic namespace-optimal implementations =-=[2, 15]-=-, which have at least Θ(n 3 ) total step complexity. It guarantees unique names in a range from 1 to n in every execution, and terminates with probability 1. Adaptive Renaming. Our second renaming alg... |

10 | Writing-all deterministically and optimally using a nontrivial number of asynchronous processors - Kowalski, Shvartsman |

8 | Wait-free test-and-set (extended abstract
- Afek, Gafni, et al.
- 1992
(Show Context)
Citation Context ...s (not on n, the total number of possible competitors). The algorithm efficiently combines the idea of a randomized splitter tree, first used in [9], with the tournament tree algorithm by Afek et al. =-=[1]-=-. Our renaming algorithms rely on both the adaptivity and anonymity properties of this implementation. Given the power of test-and-set to simplify coordination in a distributed system, and the efficie... |

7 | Asynchronous exclusive selection
- Chlebus, Kowalski
- 2008
(Show Context)
Citation Context ...e adaptive deterministic solutions known so far by providing a smaller namespace than is otherwise feasible, and by improving the total step complexity. (The most efficient adaptive algorithm to date =-=[13]-=- has total step complexity O(k 2 ), and renames in (8k − log k − 1) names.) Discussion. Both algorithms are within logarithmic factors from the immediate lower bound of Ω(n) (or Ω(k)) on the total ste... |

4 |
waitfree renaming with optimal name space and high throughput
- Eberly, Higham, et al.
- 1998
(Show Context)
Citation Context ...m has to take because of adversarial scheduling or crashes. Our algorithm improves significantly on the total step complexity of previous randomized or deterministic namespace-optimal implementations =-=[7, 12]-=-, which have at least Θ(n 3 ) total step complexity. It guarantees unique names in a range from 1 to n in every execution, and terminates with probability 1. Adaptive Renaming. Our second renaming alg... |

3 | J.H.: Fast, Long-Lived Renaming (Extended Abstract
- Moir, Anderson
- 1994
(Show Context)
Citation Context ...c algorithms that tolerate crash failures [4–6]. Even loose renaming, where the namespace is of size (2n − 1), can be quite expensive, as the best known solutions require at least Θ(n 3 ) total steps =-=[2, 7]-=-. ⋆ The work of Dan Alistarh is supported by the Swiss NCCR MICS project. The work of Hagit Attiya is supported in part by the Israel Science Foundation (grant number 953/06).Yet in practice, most pe... |

3 |
Randomized Consensus in Expected O(nlog 2 n) Operations per Processor
- Aspnes, Waarts
- 1996
(Show Context)
Citation Context ...tch. Each two-player match is decided using the randomized two-process test-and-set algorithm of Tromp and Vitànyi [16]. (Alternatively, we could use a randomized consensus algorithm with n = 2, e.g. =-=[24]-=-, although the properties stay the same.) Note that the matches are decided in a wait-free manner, since a process wins automatically if the opponent does not show up. The Backup Grid. The backup grid... |

2 |
long-lived renaming (extended abstract
- Fast
- 1994
(Show Context)
Citation Context ...mespace. For example, nearly every networked device has an ethernet address, and yet the namespace is so large as to reduce the usefulness of such names. Thus, a significant amount of research (e.g., =-=[4, 13, 20, 24, 25]-=-) has analyzed the feasibility and complexity of the renaming problem in a crash-prone distributed system. Unfortunately, renaming in a fault-prone system can be expensive, if not impossible. For exam... |

1 |
wait-free (2k-1)-renaming,” in PODC ’99: Proceedings of the eighteenth annual ACM symposium on Principles of distributed computing
- Afek, Merritt, et al.
- 1999
(Show Context)
Citation Context ...c algorithms that tolerate crash failures [4–6]. Even loose renaming, where the namespace is of size (2n − 1), can be quite expensive, as the best known solutions require at least Θ(n 3 ) total steps =-=[2, 7]-=-. ⋆ The work of Dan Alistarh is supported by the Swiss NCCR MICS project. The work of Hagit Attiya is supported in part by the Israel Science Foundation (grant number 953/06).Yet in practice, most pe... |