Denning PJ. The working set model for program behavior. Communications of the ACM 1968; 11(5):323--333.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....p s CPU share relies on two measurements: the overall system load NR and as p s past load n(p) 5.5.2 Memory Physical memory on Linux and many other OSes is not a hard constraint: applications can use more virtual memory than the available physical memory. However, if the combined working set [22] of all applications exceeds the available physical memory, then the system thrashes. Thus, I can define the supply of memory to an application as the amount of memory it can use without the system thrashing (Section 2.3.3) There are three kinds of virtual memory pages that an application can ....

P. J. Denning. The working set model of program behavior. Communications of the ACM, 11(5):323--333, May 1968.

....SAD and OLR are easily extended to maintain that information, again allowing for accurate simulations without modification to the simulator. We will show how SAD and OLR can be applied to simulations of a compressed cache [WKS99] and of a variable space allocation policy such as Working Set (WS) [Den76]. In a shorter, earlier paper [KSW99] we presented the SAD and OLR algorithms, and provided a comparison to SD. In this paper, we present several extra results. The main result is the proof of optimality of the OLR algorithm. We also present the SAD2 algorithm, which is a supplement to SAD, and ....

....can then be reduced to a single reference. This reduction does not affect the simulation of time independent paging algorithms algorithms that do not consider the exact time of each reference in making replacement decisions. Such algorithms are LRU, OPT, etc. but not, for instance, Working Set [Den76], which must track every single memory reference. Blocking is so widely applicable that it is practically assumed in most simulation work. For the remainder of this paper, when we refer to an original trace, we are referring to a blocked trace. Blocking is also interesting in that it is ....

Peter J. Denning. The working set model for program behavior. Communications of the ACM, 19(5):285- 294, 1976.

....caching worthwhile. These results suggest that our framework can allow a more principled understanding of how components of the Web interact, leading to better insight into the entire Web as a system. II. BACKGROUND AND RELATED WORK The notion of temporal locality was first recognized by Denning [6] in his definition of the working set, which is the set of unique references contained within some fixed number of past references. This and other early work was in the domain of program memory references, as opposed to Web requests. Throughout this paper we refer simply to object references in ....

P. J. Denning, "The working set model for program behavior," Communications of the ACM, vol. 11, no. 5, pp. 323--333, May 1968.

....store the most recently accessed data from main memory. Due to the principle of locality, these recently accessed parts of memory are the most likely to be re used by the processor in the near future. Caches improve performance because most programs run on processors have large amounts of locality [3]. Locality means that after a processor accesses point x in memory, x and other locations close to x in memory tend to be accessed soon. Typically, researchers divide locality into two types. Temporal locality occurs when the processor reuses the same location in memory shortly after a previous ....

P. J. Denning. The working set model for program behavior. Communications of the ACM, 11(5):323-333, May 1968.

....focused on the behavior of individual applications. The memory and I O behavior of sequential applications received great attention during the early development of virtualmemory and file cache mechanisms and policies in traditional operating systems [22] For example, Denning s working set model [10], initially applied to memory access patterns, has also been examined as a model for sequential I O behavior [19] As with file systems, studies of commercial workloads [6, 18] have become more common in recent years. However, in this domain, the interaction or pipeline behavior of sequential ....

P. J. Denning. The working set model for program behavior. In Proceedings of the ACM symposium on Operating System Principles (SOSP), 1967.

....based on observations of past behavior has been studied extensively. Most of them exploit the recency and or frequency of past references. In traditional paging systems, temporal locality (recency information) is dominant and hence, a number of studies, such as the working set model by Denning [4], tried to characterize the temporal locality of page references. However, reference characteristics of Web environments are quite different from those of traditional computing environment due to the following reasons. First, reference streams seen by Web caches are generally the union of streams ....

P. J. Denning, "The Working Set Model for Program Behavior," Comm. ACM, vol.11. no.5, pp.323-333, 1968.

....hinges on the existence of a working sets at the function level and methods for identifying them accurately and eciently. In this paper we report on the experiments which show that programs exhibit a working set nature at the function level. While there is a large body of work on program behavior [24, 14, 9, 3, 21], to the best of our knowledge, there has been no e ort to study locality at the level of functions. 3 Background 3.1 De nitions Let N = f1; 2; ng be the set of functions which make up the program. Its dynamic behavior can be modelled as a sequence of accesses to the functions. Let = f ....

Peter J. Denning. The working set model for program behavior. In Proceedings of the ACM Symposium on Operating System Principles, pages 323-333. ACM, November 1967.

....hinges on the existence of a working sets at the function level and methods for identifying them accurately and eciently. In this paper we report on the experiments which show that programs exhibit a working set nature at the function level. While there is a large body of work on program behavior [25, 11, 12, 10, 22], to the best of our knowledge, there has been no e ort to study locality at the level of functions. 3 The Simulation Platform Program simulation techniques can be classi ed into one of 3 categories: trace driven, program driven or execution driven. In the trace driven approach a memory ....

Peter J. Denning. The working set model for program behavior. In Proceedings of the ACM Symposium on Operating System Principles, pages 323-333. ACM, November 1967.

....does not have the design challenge. Just like LRU, it chooses the victim block in the global stack. However, it can use the advantages provided by the detection based schemes. 2. 4 Working Set Models Lastly, we would like to compare our work with the working set model, an early work by Denning [6]. A working set of a program is a set of its recently used pages. Specifically, at virtual time t, a program s working set W t (#) is the subset of all pages of the program, which have been referenced in the previous # virtual time units (the working set window) A working set replacement ....

P. J. Denning, "The Working Set Model for Program Behavior", Communications of the ACM, Vol. 11, No. 5, May, 1968, pp. 323-333.

....of different deltas in order to introduce a new trailer. This environment can be compared with a client server software development environment that consists of workstations connected to a file server. The latency for sequentially browsing a file (video) should be less than 100 milliseconds [3]. Similarly, jumping from one line (frame) of the file (video) to another and referencing different files (video clips) should not be noticeable. The client server nature should not impact the correctness or behavior of a program (previewing of an edited clip) Otherwise, it is difficult to ....

P. J. Denning. The Working Set Model for Program Behavior. Communications of the ACM, 11(5):323-333, 1968.

....trac. Our observations are illustrated by comparing the footprints of real and synthetic traces, and by simulating routing table lookups driven by both types of traces. 1 Introduction It is well established that memory reference strings of computer programs exhibit spatial and temporal locality [1]. This locality is the motivating concept behind the use of instruction and data caches, which play a critical role in improving the performance of contemporary computer systems. In this paper, we apply the concepts of locality and workload modeling originally developed for investigating program ....

....This is because we only consider generating repeating addresses when the hit index is below LRU POINTER. Thus, in terms of the total fill LRU stack with unique items InitializeStack; for count: 1 to SyntheticTraceLength do begin generate index from uniform random real in [0, 1] U: Random1; if (U 1 Theta) then index : round( U ( A Theta) Theta) 1 (1 Theta) else index : round(random Cc) adjust if index 1 then index : 1; determine the address to output R = Random2 if index LRU POINTER and R SelfRepeatRate then address = ....

P. J. Denning. Working set model for program behavior. Communications of the ACM, 11(5):323-333, 1968.

....interconnection device between LANs) In particular, we propose four new measures of network traffic locality: persistence, concentration, address re use, and reference density. These measures are based, in part, on locality characteristics identified in other domains, such as program behaviour [4], and file referencing behaviour [12] Definitions are provided for each measure, and measurement data collected from a local area network environment is used to quantify each measure. The remainder of this paper is organized as follows. Section 2 motivates and defines our four locality measures. ....

....the performance of a network interconnection device. Finally, Section 5 presents our conclusions, and suggested directions for future research. 2 Locality Characteristics Studies of memory and file referencing behaviour have found that such references typically display non random characteristics [4, 12]. References are distributed nonuniformly in both time and space with small subsets of information favoured over relatively long time intervals. The term locality of reference has been used to describe this phenomenon. The General Principle of Locality [5] has been described as: 1. During any ....

[Article contains additional citation context not shown here]

P. Denning, "Working Set Model for Program Behavior", Communications of the ACM, Vol. 11, No. 5, May 1968, pp. 323-333.

....Locality of reference, working sets, compiled communication, communication cost, interconnection networks, multiprocessor communication. 2 1 Introduction Locality of reference was first identified as being important in the context of managing a program s use of pages of physical memory [6]. It has since been recognized as a concept that can be applied to a wide variety of contexts, and has been exploited to improve the performance of many functions of modern computers. It is the fundamental reason for the effectiveness of caches used for both memory and file accesses. It has been ....

P. Denning. The working set model for program behavior. Communications of the ACM, 11(5):323--333, May 1968.

....and a CMG, we develop new characterization streams based on the Inter Reference Gap (IRG) model [91] 2.2. 1 Procedure based Inter Reference Gap Modeling There has been a significant amount of prior work on modeling the interaction and locality of code segments, with most of them focusing on page [92, 93, 94] and or cache line [95, 96, 97] locality and liveness. In [96] Quong presents an estimation model that predicts miss rates for caches of arbitrary associativity using profile information. This model requires a compressed form of the trace in order to approximate the cache miss ratio, thus ....

P. Denning. The Working Set Model for Program Behavior. Communications of the ACM, 11(5):323--333, May 1968.

....page replacement algorithms; a good survey can be found in [16] Among these theoretical studies, of particular interests to our study are the ones on program behavior modelling and optimal online algorithms for each model. The models include independent reference [1] LRU stack [30] working set [11], access graphs [5] Markov model [21] For each of these models optimal online algorithms are found, for example, LRU for LRU stack model [16] access graph algorithm [18] and Markov paging algorithm [21] Our study suggests that the access graph model might be the appropriate model for the ....

Peter J. Denning. The working set model for program behavior. Communications of the ACM, 11(5):323--333, May 1968.

....effectively utilized by that program. It is important to characterize these as it does not make sense to use an application to evaluate a multiprocessor system in a regime where the speedups are essentially flat or going down as number of processors is increased. The working sets of a program [Den68, RSG93] correspond to points where the cache miss rate drops sharply as the cache size is increased, and they are indicative of the temporal locality in an application. A detailed characterization of the working sets and how they scale with number of processors and data set size is key to determining ....

Peter J. Denning. The Working Set Model for Program Behavior. Communications of the ACM, 11(5):323-333.

....of communicating processes: A = fG 1 [ G 2 [ G nG g, for i 6= j; G i G j = nA = the size of the process group. nG = the number of clusters (sub groups) G i = cluster i. g i = the number of elements of cluster i. Following the definition of working set model introduced by Denning [7], we define the communication working set of process A k at time t as: CA k (t; Delta) collection of processes that communicate with process A k during the time interval (t Gamma Delta; t) Delta = working set parameter For a given time window [t Gamma Delta; t] CA k (t; Delta) can be ....

P. J. Denning. The working set model for program behavior. Communications of the ACM, 11(5):323--333, May 1968.

....can then be reduced to a single reference. This reduction does not affect the simulation of time independent paging algorithms algorithms that do not consider the exact time of each reference in making replacement decisions. Such algorithms are LRU, OPT, etc. but not, for instance, Working Set [Den76] Blocking is so widely applicable that it is practically assumed in most simulation work. For the remainder of this paper, when we refer to an original trace, we are referring to a blocked trace. Blocking is also interesting in that it is exploiting a different kind of regularity than most ....

Peter J. Denning. The working set model for program behavior. Communications of the ACM, 19(5):285--294, 1976.

....in the main memory. The set of pages currently present in the main memory is referred to as the resident set. The mechanism for selecting pages to be removed is called the removal policy or replacement algorithm. One of the most well known removal policies is the Denning Working Set policy, DWS) [3]. It belongs to the class of variable space policies as it tries to adjust the memory allocation to the current needs of the program. In order to achieve this, the DWS estimates the locality set of a program by its working set (WS) All the pages that have been referenced during the last T memory ....

P. J. Denning. The working set model for program behavior. Communications of the ACM, 11:323--333, May 1968.

....becomes overcommitted, each client doesn t have enough memory to hold the pages it needs to make forward progress, so most page accesses cause faults, slowing progress to a near halt. Denning conjectured that each program needs to have its working set of data present to be able to make progress [Denning 68] He defined a program s working set as the number of pages accessed in the last t seconds, where t is the working set parameter. The working set policy included an admission criterion that prevented new jobs from being admitted until their working set could be accommodated. Another issue the ....

....level allocates frames among applications, the second level manages those frames for each application. This provides some isolation between applications. To date, local allocation schemes are uncommon. Systems that use local allocation include VMS, Microsoft s NT [Custer 93] and working set [Denning 68] The working set policy combines a simple global allocation policy with an LRU like local allocation policy. The admission policy prevents thrashing and the LRU like page replacement policy minimizes paging activity due to most applications locality of reference. 2.5.3 Degrees of flexibility ....

Peter J. Denning. The Working Set Model for Program Behavior. Communications of the ACM , 11(5):323--333, 1968.

....in their size and organization, and there may be more than one level of cache in the hierarchy. These details are not important to our discussion; we do not address them further. 1. We use this term in a more informal sense than its original definition, which refers to virtual memory pages [Den68]. CPU CACHE MAIN I O REGISTER FILE MEMORY BUS I O BUS Figure 1.1 Typical Memory Hierarchy DEVICES MEMORY Chapter 1: Introduction 3 The next level of the hierarchy is main memory, which can be organized in a variety of ways. The important parameters for our discussion are width, bandwidth, ....

P. Denning, "The Working Set Model for Program Behavior", Communications of the ACM, 11(5):323-333, May 1968.

....emulate page referenced support. This implementation provides an adequate LRU approximation with low overhead, leading to the conclusion that hardware page referenced information is not a requirement for a satisfactory page replacement algorithm. Page replacement is a very old problem [Belady 66, Denning 68] that has received little attention in recent years. This is for good reason. The choice of page replacement algorithm largely doesn t matter, because memory is cheap and most machines have enough memory that paging is not a concern. If a machine clearly doesn t have enough memory to support its ....

Denning, P. J. The Working Set Model for Program Behavior. Communications of the ACM, 11(5):323--333, May 1968.

....and the main memory (since no information was written to the main memory from that time on) Thus, by invalidating the contents of the cache and reloading the processor state at that point, we establish consistency in the data content of the cache and the main memory. 2 Cache models, such as [5, 1, 12], indicated the existence of working sets during the executing time of the program. These assume that the miss ratio within the working set can be considered as a constant and only when the program moves from one working set to another, a burst of cache misses can occur. If we assume a uniform ....

P. Denning, The working set model for program behavior, Communication of the ACM, vol. 11, pp. 323--333, May 1968.

.... characteristics of the application, including dynamic instruction count, number of distinct touched instructions, a parallel execution profile (serial and parallel phases) numberof synchronization barriers and locks, I O traffic, and percentage of memory instructions (by type) # The working set [1] of an application in an execution interval is the number of distinct memory locations accessed in this interval. The working set size is a measure of the application s temporal locality, which affects its cache performance. Working set characterization is important for cache size selection for a ....

P. J. Denning, "Working Set Model for Program Behavior," Commun. ACM, vol. 11, no. 6, pp. 323--333, 1968.

....for eviction is the page not accessed for the longest period of time. In practice, the exact per page time of last reference is usually not available, so most policies only approximate LRU. LRU replacement works well in practice, and has a reasonable theoretical underpinning (the Working Set model [12]) and is simple to implement. However, when a single large program requires more main memory that is present, LRU often suffers poor performance. A well known case for which LRU page replacement performs poorly is so called 2 sequential flooding, in which a program accesses memory in an ....

.... appear for frequently used pages) Gnuplot makes three passes over a single large memory area, but the time required by the second pass differs substantially from the input phase (first pass) and the output phase (third pass) Denning introduced the Working Set model as a model program behavior [12]. Central to the model is the idea of phase transitions, in which programs stop doing one computation and move on to another; during such transitions, one would expect the current working set of pages to go out of use, and another set of pages to come into use. While there is some evidence of ....

Peter J. Denning. The working set model for program behavior. Communications of the ACM, 11(5):323--333, May 1968.

....through a simulator to determine hit rates or average memory access times. Occasionally in the past and more often now, researchers are taking a different approach and attempting to design not just a better cache, but better ways to design and analyze caches through new models or measures [1, 3, 5, 7, 8, 9, 12, 13, 14, 16]. This paper describes an analytical framework for cache design. There are four major components that form the framework, each of which is a contribution on its own. First, the TSpec notation is a more formal way for researchers to communicate with clarity about memory references generated by a ....

P. Denning. The working set model for program behavior. Communications of the ACM 11(5), May, 1968.

....devices. For example, a typical three level hierarchy might consist of a tape juke box with several read write devices, a number of magnetic disks, and some memory [7] A copy of the data resides on the tape juke box on a permanent basis. The magnetic disk space is used to maintain the working set [5] (locality) of the target application, i.e. those objects that are repeatedly accessed during a pre speci ed window of time. The memory maintains pages of the les that might pertain to either traditional data or multimedia data. The reason for expecting HSS is the cost of storage. It is ....

P. J. Denning. The Working Set Model for Program Behavior. Communications of the ACM, 11(5), 1968.

....who offered the handling of I O activity via a technique called virtual memory which also assumes a considerable amount of help from the hardware. The techniques on this line generally fall into two groups: 1) techniques which consider smart virtual memory implementations and replacement policies [2, 11]; and (2) techniques which consider re shaping the data reference patterns in order to exploit the given hardware facilities and system software [23, 32] The latter group then paved the way for automatic program restructuring techniques like loop distribution and page indexing [1] In general, ....

P.J. Denning. The Working Set Model for Program Behavior. In Comm. of the ACM, Vol. 11, No. 5, May 1968. 29

....this thesis, the relation of this work to previous literature, our approach to analysis, and the complete thesis structure. 1. 1 Problem Origin The benefits of caching, utilized throughout the memory hierarchy in computer systems design, depends on the well known locality of reference principle [18, 19]. In the late 1980 s, a new generation of mainframe I O controllers were introduced [33, 2] which capitalize on an intelligent multiported cached design. When a desired record is not found in the the cache, the cache storage buffers the disk request along with the remaining records on the track ....

Denning, P.J. "The Working Set Model for Program Behavior". Communications of the ACM, 11(5), May 1968.

....Since the load returns only two unique values, addunsigned is called from only two call sites during this run. FDO2 submission Please do not distribute 9 Several researchers have investigated the time based behavior of applications with respect to a single metric. For instance, Denning s [8] concept of a working set is often explained using a plot of pages referenced over time. Jouppi [15] examined the distribution of instruction level parallelism in applications over time, and more recently, Albonesi [1] proposed an adaptive processor architecture based on such observations. Within ....

P. Denning. "The Working Set Model for Program Behavior," Communications of the ACM, May 1968.

....Furthermore, the monotonic decrease of the curve for gap sizes less than 8 implies that the least recently used (LRU) cache replacement algorithm is optimal for cache sizes less than 8. In Sec 4 This can be expressed by saying that our trace is characterized by a small working set window size [6]. tion 3.2, we compare cache replacement algorithms. 3.2 Cache replacement algorithms A hit is said to occur in the cache when a name received by the DSA is already in the cache. A fault or miss is said to occur when a name received by the DSA is not found in the cache. On a cache miss, one of ....

P. J. Denning, "The working set model for program behavior", Comm. ACM, vol. 11, no. 5, May 1968, pp. 323-333.

....property to arrivals on computer networks it has been found that the performance of gateways can be made more efficient [11] This chapter deals with the basic concepts of locality and caching. 2. 1 Locality The principle of locality was first applied in the study of memory referencing behaviour [9], but was subsequently observed in the usage patterns of other resources, such as file systems [28] The general principle of locality [10] has been described as: 12 13 1. During any interval of time the resource demands of the system are distributed non uniformly over the total set of available ....

....is an aspect of locality which deals with this issue. The tendency of the destination addresses to be limited to a small subset of the address space is defined as the destination address concentration of the trace. The working set model, used extensively to characterize program behaviour [9, 13], seems to be a good framework to express the notion of concentration in a reference stream. Let w be the working set window size and W w be the corresponding average working set size. The working set size for a particular trace is the number of different hosts that receive at least one of the w ....

Denning P. J., "Working Set Model for Program Behavior", Communications of the ACM, Vol. 11, No. 5, May 1968, pp. 323-333.

....real indexed multi megabyte cache) Thus, our results suggest that careful mapping by the operating system can get about half the cache miss reduction that a cache size (or associativity) doubling can. 1. Introduction Most general purpose computer systems support both paged virtual memory [DENN68] and caches [SMIT82] Virtual memory caches pages by translating virtual addresses to their corresponding real addresses (often by consulting a page table) This translation from virtual pages to real page frames is usually fully associative: any page frame in the physical main memory can hold ....

P. J. DENNING, "The Working Set Model for Program Behavior," Communications of the ACM, vol. 11, no. 5, May 1968, pp. 323-333.

No context found.

Denning PJ. The working set model for program behavior. Communications of the ACM 1968; 11(5):323--333.

No context found.

P.J. Denning, The working set model for program behavior, Commun. ACM 11 (5) (1968) 323--333.

No context found.

P.J. Denning, "The Working Set Model for Program Behavior", CACM 19(5) pp.285-294 (1976).

No context found.

P.J. Denning, "The working set model for program behavior," Communications of the ACM 11 (1968), pp. 323--333.

No context found.

P. J. Denning. The working set model for program behavior. Commun. ACM, 11(5):323--333, May 1968.

No context found.

P. J. Denning. The working set model for program behavior. Communications of the ACM, 11(5):323--333, May 1968.

No context found.

P. J. Denning. The working set model for program behavior. Communications of the ACM, 11(5):323--333, 1968.

No context found.

P. J. Denning. The Working Set Model for Program Behavior. Communications of the ACM, 11(5), May 1968. 10

No context found.

Denning, P.J. The working set model for program behavior. Comm. ACM 11, 5 (May 1968), 323-333.

No context found.

P. J. Denning. The working set model for program behavior. Communications of the ACM, 11(5):323-333, May 1968.

No context found.

Peter J. Denning. The working set model for program behavior. Communications of the ACM, 11(5):323--333, May 1968.

No context found.

P. Denning, The working set model for program behavior, Comm. Assoc. Comput. Mach. 11 (1968), 323#333.

No context found.

Denning, P.J. The working set model for program behavior. Comm. ACM 1l, 5 (May 1968), 323-333.

No context found.

Denning, P. J., "The Working Set Model for Program Behavior", CACM 11, 5 (May 1968), pp. 323-333.

No context found.

DEN68 P. Denning. "The working set model for program behavior.", Communications of the ACM 11:5, May, 1968.

No context found.

P. J. Denning. The working set model for program behavior. Communications of the ACM, 11(5):323--333, May 1968.

No context found.

P. Denning, "The working set model for program behavior", Communications of the ACM vol. 11 no. 5, May, 1968.

First 50 documents  Next 50

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