W. D. Clinger and L. T. Hansen. Generational garbage collection and the radioactive decay model. In PLDI '97: Proceedings of the ACM SIGPLAN 1997.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....in our experiments are never again used, it means that after an active period of accessing, most objects become useless, but they can not be garbage collected immediately. This seems to support past research that most objects die young [30] and that old generations have longer life expectancies [8]. 6.1 Open Issues While we have learned a great deal about efficient offloading of memory in this paper several issues still remain. Do our results apply more widely For example to other virtual machines and applications. ChaiVM is a fairly representative JVM to use, since it supports the ....

W. D. Clinger and lars T. Hansen. Generational Garbage Collection and the Radioactive Decay Model. In SIGPLAN PLDI'97, pages 97--108, 1997.

....hybrids. The most common segregation policy is by age: two or three agebased regions are common, but some collectors use more [27] Generational age based collectors exploit the weak generational hypothesis [34] Older first collection [32] and renewal older first (here called older first mix) [15, 19] are premised upon the just allocated (Older First) or copied (renewal older first) objects being likely to stay reachable for a while. Beltway configurations exploit these characteristics with multiple increments on FIFO belts. Beltway X.X.100 is complete, unlike older first. More importantly, ....

William D. Clinger and Lars T. Hansen. Generational garbage collection and the Radioactive Decay model. In Proceedings of the 1997.

....memory deallocation. However, garbage collection imposes a performance overhead since it must identify and reuse memory that is no longer accessible by the program, while the program is executing. The performance of heap allocation and collection techniques has been the focus of much research [10, 12, 9, 16, 1, 13, 31, 6]. The goal of most of this prior work has been to provide generalpurpose mechanisms that enable high performance execution across all applications. However, other prior research [5, 14, 33] has shown that the efficacy of a memory management system (the allocator and the garbage collector) is ....

CLINGER, W., AND HANSEN, L. T. Generational garbage collection and the radioactive decay model. In Proceedings of the ACM SIGPLAN '97 Conference on Programming Language Design and Implementation (May 1997), pp. 97--108.

....wastes time copying live objects from the younger to older generation, and collects the entire heap just as often as a nongenerational GC. When assumption 2 fails, minor collections become expensive, and may promote garbage due to back pointers from dead older objects. Assumption 1 is questionable [5, 3]. Assumption 2 may be violated by certain programs [19] It appears that the success of GGC may derive more from its structuring of the heap and collections of di erent frequencies, than from the accuracy of its heuristic assumptions. 1.2 Leveled Garbage Collection We propose a new method, ....

....on medium to large problem sizes of the benchmark saturated. A heap of 10 MB is used, and hence no paging cost is involved. In this gure, F = 0:5 S for LGC. 6 Related Work Recently, Clinger and Hansen proposed a nonpredictive generational collector that doesn t keep track of objects ages [5]. However, it is used for collecting objects in the oldest generation only. Also, it makes no e ort to achieve better paging performance than conventional generational collectors. On their web pages, the authors don t claim performance improvements over the conventional generational collectors. ....

William D. Clinger and Lars T. Hansen. Generational garbage collection and the radioactive decay model. In Proceedings of the

....information about the lifespans of objects, a number of proposals have been put forward to improve the efficiency of older generations. Most recently, work by Will Clinger and Lars Hansen, and by Darko Stefanovic, Eliot Moss, and Kathryn McKinley has examined the idea of oldest first collectors [6, 14]. Our proposal differs in that we attempt to improve the collector by having knowledge about the objects in the heap. In this sense, it is orthogonal to oldest first collection techniques and may even improve these techniques by placing objects where they are likely to be collected shortly after ....

W. D. Clinger and L. T. Hansen. Generational garbage collection and the Radioactive Decay model. In Proceedings of SIGPLAN'97 Conference on Programming Languages Design and Implementation, ACM SIGPLAN Notices, pages 97--108, Las Vegas, Nevada, June 1997. ACM Press.

....has the same residual lifetime as the lifetime of a newly allocated object, and this greatly simplifies the analysis. Thus, the exponential distribution was used by Clinger and Hansen in the analysis (and to inspire the design) of a non predictive collector, outside the generational realm [3]. In our examination of a generalized form of that collector [13] we decided to use not only the exponential distribution s t e rt , but also a variation with decreasing mortality s t e rt as being in agreement with the strong generational hypothesis, as well as a ....

W. D. Clinger and L. T. Hansen. Generational garbage collection and the radioactive decay model. SIGPLAN Notices, 32(5):97--108, May 1997. Proceedings of the ACM SIGPLAN '97 Conference on Programming Language Design and Implementation.

....distributions we use to study the three collectors in Section 5. The remaining sections describe collection space and time costs, provide mathematical and simulation analysis of these costs for the three collectors. 2 Related work The primary related work is a recent study by Clinger and Hansen [3]. They considered the behavior of heaps under the exponential distribution (radioactive decay model) and proposed a non predictive collector. Rather than collecting the youngest data like a traditional generational collector, the non predictive collector processes the youngest portion of the ....

....Square exponential Survivor function Mortality function Density function Figure 3: Analytical distributions. Exponential survival: The survivor function is s(t) e Gammalt , mortality is m(t) l, and the probability density function is 1 Also known as the radioactive decay model [2, 3]. 4 f (t) le Gammalt . The mortality is constant for all ages. For this distribution, we also have analytical solutions for collector performance. We used them to validate the simulation procedure. Square root exponential survival: The survivor function is s(t) e Gamma p bt , m(t) ....

William D. Clinger and Lars T. Hansen. Generational garbage collection and the radioactive decay model. SIGPLAN Notices, 32(5):97--108, May 1997. Proceedings of the ACM SIGPLAN '97 Conference on Programming Language Design and Implementation.

....has the same residual lifetime as the lifetime of a newly allocated object, and this greatly simplifies the analysis. Thus, the exponential distribution was used by Clinger and Hansen in the analysis (and to inspire the design) of a non predictive collector, outside the generational realm [3]. In our examination of a generalized form of that collector [13] we decided to use not only the exponential distribution s(t) e rt , but also a variation with decreasing mortality s(t) e p rt as being in agreement with the strong generational hypothesis, as well as a variation with ....

W. D. Clinger and L. T. Hansen. Generational garbage collection and the radioactive decay model. SIGPLAN Notices, 32(5):97--108, May 1997. Proceedings of the ACM SIGPLAN '97 Conference on Programming Language Design and Implementation.

....has the same residual lifetime as the lifetime of a newly allocated object, and this greatly simplifies the analysis. Thus, the exponential distribution was used by Clinger and Hansen in the analysis (and to inspire the design) of a non predictive collector, outside the generational realm [CH97]. In our examination of a generalized form of that collector [SMM98] we decided to use not only the exponential distribution s(t) e rt , but also a variation with decreasing mortality s(t) e p rt as being in agreement with the strong generational hypothesis, as well as a variation with ....

William D. Clinger and Lars T. Hansen, Generational garbage collection and the radioactive decay model, SIGPLAN Notices 32 (1997), no. 5, 97--108, Proceedings of the ACM SIGPLAN '97 Conference on Programming Language Design and Implementation.

....5. The remaining sections describe collection space and time costs, and provide mathematical and simulation analyses of these costs for the three collectors, including their behavior on actual mature object traces. 2 Related work The primary related work is a recent study by Clinger and Hansen [Clinger and Hansen, 1997]. They considered the behavior of heaps under the exponential distribution (radioactive decay model) and proposed a nonpredictive collector. Rather than collecting the youngest data like a traditional generational collector, the nonpredictive collector processes the youngest portion of the ....

....collector behavior with mature object distributions determined from actual traces; and we simulate collector behavior on actual mature object traces. There have been a number of papers about object lifetimes and the generational hypotheses; Clinger and Hansen offer a good overview of these [Clinger and Hansen, 1997, Section 9 (pp. 106 107) We could not find any recent studies of mature object lifetime distributions that go beyond anecdotal reports, with the signal exception of Hayes data [Hayes, 1993] One of us has worked on collection techniques directed at mature objects, the Mature Object Space ....

[Article contains additional citation context not shown here]

Clinger, W. D. and Hansen, L. T. (1997). Generational garbage collection and the radioactive decay model. SIGPLAN Notices, 32(5):97-- 108. Proceedings of the ACM SIGPLAN '97 Conference on Programming Language Design and Implementation.

....until they have had time to become older and die. The work closest to ours in this respect is Clinger and Hansen s non predictive collector, which organizes objects according to the time elapsed since last collected and performs well with a hypothetical exponential distribution of lifetimes [Clinger and Hansen, 1997]. However, as we shall see in our empirical study in Chapter 5, that particular organization does not work remarkably well. 2.1.4.1 Pointer maintenance In order to collect one region (a set of youngest generations) without collecting the remainder of the heap, the set of pointers crossing the ....

....This discussion was entirely independent of the actual shape of the survivor and mortality curves, but the relative merits of generational and other collectors will certainly depend on them. Assumed or measured object lifetime distributions can be used to predict the performance of collectors [Clinger and Hansen, 1997]. However, the implicit assumption that object lifetimes are independent identically distributed random variables is far from true: phase behavior of programs must lead to highly correlated object lifetimes. The collector performance calculations are tied to the assumption of a steady state, but ....

[Article contains additional citation context not shown here]

William D. Clinger and Lars T. Hansen. Generational garbage collection and the radioactive decay model. SIGPLAN Notices, 32(5):97--108, May 1997. Proceedings of the ACM SIGPLAN '97 Conference on Programming Language Design and Implementation.

....but we combine it with efficient mechanisms to keep pointer tracking costs in check, even without the youngest first discipline. Clinger and Hansen proposed a collector for Scheme that does not base collection decisions on object age, but rather on the time elapsed since last collection [11], and focuses on objects for which that time is longest. There have been historical precursors to this idea [2, 18, 6] Although this algorithm is not age based, it prompted us to investigate similarly flexible age based ones; in the context of object oriented languages that we examined, we ....

.... we recognize that our study draws on previous experience with generational garbage collection implementations [19, 27, 20, 24, 28, 35] their policies [29, 30, 31, 34, 1, 13] their write barrier mechanisms [33, 15, 14] and their evaluation with respect to object allocation and lifetime behavior [3, 26, 11]. Achieving performance improvements with generational collection critically depends on setting or adapting the configuration parameters right incorrectly chosen generation sizes can cause performance to degrade severely. We have confirmed these matters in our observations of multi generational ....

Clinger, W. D., and Hansen, L. T. Generational garbage collection and the radioactive decay model. SIGPLAN Notices 32, 5 (May 1997), 97--108. Proceedings of the ACM SIGPLAN '97 Conference on Programming Language Design and Implementation.

....pause times [Hudson and Moss, 1992] The work closest to ours is Clinger and Hansen s nonpredictive collector, which organizes objects according to the time elapsed since last collected and is applied to the mature space only. It performs well with an exponential distribution of lifetimes [Clinger and Hansen, 1997]. That work prompted our exploration of alternative GC schemes, but we use the real age of objects and explore the application of the collectors to the entire heap. 3 Intuition and motivation In this section, we examine the above mentioned two beliefs about object lifetimes and pointer ....

....to GYF. youngest R R S S C C R R Collection 1 Collection 2 oldest Figure 9: Deferred older first collection. youngest R R R C S S C Collection 1 Collection 2 oldest Figure 10: Deferred younger first collection. 4. 2 Renewal age schemes The collector proposed recently by Clinger and Hansen [Clinger and Hansen, 1997] and its generalization, the oldest first collector [Stefanovic et al. 1998] are also age based, but according to renewal age, and older objects form the collected region. A renewal oldest first collector always chooses a leftmost subsequence of the sequence of heap objects but the survivors ....

Clinger, W. D. and Hansen, L. T. (1997). Generational garbage collection and the radioactive decay model. SIGPLAN Notices, 32(5):97--108. Proceedings of the ACM SIGPLAN '97 Conference on Programming Language Design and Implementation.

....until they have had time to become older and die. The work closest to ours in this respect is Clinger and Hansen s non predictive collector, which organizes objects according to the time elapsed since last collected and performs well with a hypothetical exponential distribution of lifetimes [Clinger and Hansen, 1997]. However, as we shall see in our empirical study in Chapter 5, that particular organization does not work remarkably well. 2.1.4.1 Pointer maintenance In order to collect one region (a set of youngest generations) without collecting the remainder of the heap, the set of pointers crossing the ....

....This discussion was entirely independent of the actual shape of the survivor and mortality curves, but the relative merits of generational and other collectors will certainly depend on them. Assumed or measured object lifetime distributions can be used to predict the performance of collectors [Clinger and Hansen, 1997]. However, the implicit assumption that object lifetimes are independent identically distributed random variables is far from true: phase behavior of programs must lead to highly correlated object lifetimes. The collector performance calculations are tied to the assumption of a steady state, but ....

[Article contains additional citation context not shown here]

William D. Clinger and Lars T. Hansen. Generational garbage collection and the radioactive decay model. SIGPLAN Notices, 32(5):97--108, May 1997. Proceedings of the ACM SIGPLAN '97 Conference on Programming Language Design and Implementation.

.... live object has a 50 probability of dying during the next interval whose duration is the half life that parameterizes the model [2] If this half life is short, then most objects die young, but a conventional generational collector will actually perform worse than a non generational collector [9]. A more sophisticated explanation for the effectiveness of generational garbage collectors is that they try to predict which objects are likely to die soon, and how well they work depends upon the accuracy of these heuristic predictions. This explanation is also incorrect. The radioactive decay ....

....collect young objects more often than old. These young objects haven t had much time to die, so collecting them doesn t recover much storage. An older first generational collector, which collects old objects more often than young, would recover more storage for a similar amount of effort [9]. Why then do younger first collectors usually perform well in practice Because most programs satisfy the weak generational hypothesis, which asserts that young objects die at a considerably faster rate than older objects [19] Hence the young generations often contain a higher fraction of ....

[Article contains additional citation context not shown here]

William D Clinger and Lars T Hansen. Generational Garbage Collection and the Radioactive Decay Model. Proceedings of the 1997.

No context found.

W. D. Clinger and L. T. Hansen. Generational garbage collection and the radioactive decay model. In PLDI '97: Proceedings of the ACM SIGPLAN 1997.

No context found.

CLINGER, W., AND HANSEN, L. T. Generational garbage collection and the radioactive decay model. In Proceedings of the ACM SIGPLAN '97 Conference on Programming Language Design and Implementation (May 1997), pp. 97--108.

No context found.

CLINGER, W., AND HANSEN, L. T. Generational garbage collection and the radioactive decay model. In Proceedings of the ACM SIGPLAN '97 Conference on Programming Language Design and Implementation (May 1997), pp. 97--108.

No context found.

William D. Clinger and Lars Thomas Hansen. Generational garbage collection and the radioactive decay model. In Programming Languages Design and Implementation (PLDI), 1997.

No context found.

CLINGER, W., AND HANSEN, L. T. Generational garbage collection and the radioactive decay model. In Proceedings of the ACM SIGPLAN '97 Conference on Programming Language Design and Implementation (May 1997), pp. 97--108.

No context found.

W. Clinger and L. Hansen. Generational garbage collection and the radioactive decay model. Proc. SIGPLAN 97 Conference on Programming Language Design and Implementation, May 1997.

No context found.

William Clinger and Lars Hansen. Generational garbage collection and the radioactive decay model. Proc. SIGPLAN 97 Conference on Programming Language Design and Implementation, May 1997.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC