M. W. Mutka and M. Livny. Profiling workstations' available capacity for remote execution. Performance, 1987.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....these areas. We also describe the Bulk Synchronous Parallel model which aims to unify various parallel architectures including networks of workstations into one model. 2. 1 Workstation Usage Patterns Mutka et al. analyzed the workstation usage patterns and their availability for remote execution [52]. They monitored a network of workstations consisting of 13 workstations over a period of 5 months. The workstations observed were owned by a variety of users: faculty, systems programmers and graduate students. They obtained the profile of available and nonavailable periods of workstations. A ....

....adapt to the computing environment and analyze their performance. 4.1 Nondedicated Workstations as Transient Processors Processors in a network of workstations (NOW) are often underutilized. Sev eral studies indicate that a large number of workstations in a NOW are idle at any given time [15, 38, 52]. Arpaci et al. [4] report that, although the set of idle machines changes over time, the total number of idle machines stays relatively con stant. Our objective is to use the idle workstations to run additional jobs. There have been several systems that attempt to make use of idle workstations to ....

[Article contains additional citation context not shown here]

M. W. Mutka and M. Livny. Profiling Workstations' Available Capacity for Remote Execution. In Proc. 12th Symposium on Computer Performance, Brussels, Belgium, December 7-9, 1987.

....computation intensive parallel applications on a cluster of workstations. However, the application may need to be restructured by changing the data distribution to avoid frequent communication. 1. 3 Parallel Computing on Nondedicated Workstations Workstations in a cluster are often underutilized [8]. Arpaci et al. [1] report that, although the set of idle machines changes over time, the total number of idle machines stays relatively constant. Our objective is to use the idle workstations in a cluster to run additional parallel jobs. 1.3.1 Nondedicated Workstations as Transient Processors ....

....scalable. 1.6.3 Results Figure 1.6(a) shows a plot of the execution times of maximum independent set problem on transient processors using the A BSP library with ts 40 minutes and t 20 minutes respectively. These values for ts and t are within the range of values reported in earlier works [8]. The measurements were taken on a clus ter of Sun Sparc 5 workstations connected by a 10 Mbps Ethernet. The number of processors available is much larger than the degree of parallelism used in the simulations and therefore, migration to an available processor was always possible. The execution ....

M. W. Mutka and M. Livny. Profiling Workstations' Available Capacity for Remote Execution. In Proc. 12th Symposium on Computer Peryormance, Brussels, Belgium, December 7-9, 1987.

....fully utilize the capacity of the workstation they own, and frustrated users who for long periods of time have computing demands that are beyond the power of their workstations. The throughput of these frustrated users is limited by the power of their workstations. In a paper by Mutka and Livny [16] it was shown that in a computing environment of workstations connected by a local area network, about 70 of the workstations is available for remote execution. Condor is a software package for executing long running, computation intensive jobs on workstations that would otherwise be idle. ....

M.W. Mutka and M. Livny, "Profiling workstations' available capacity for remote execution, " in Proceedings of Performance '87, The 12th IFIP W.G. 7.3 International Symposium on Computer Performance Modeling, Measurement and Evaluation, Brussels, Belgium, 1987, pp. 529---544.

....calls and checkpointing, which are implemented in Condor. Finally, we describe an error found in the new checkpoint mechanism, which occurred only with WAN connections, and how this error was corrected. For more detailed information of the Condor system, we refer the reader to the literature [10, 11, 12, 13, 14, 15, 16]. 2.2.1 The Condor daemons A Condor workstation (machine) runs two Condor daemons, the scheduler daemon Schedd and the starter daemon Startd. One Condor machine is designated to run the Central Manager (CM) which consists of two daemons, the Negotiator and the Collector. The Condor daemons ....

....BM determines the bandwidths between the source machine and the destination machines, and returns them to the CM. 3.4.2 The estimated available run time The communication criteria require an estimation of the period of time an execution machine will be available to run the job. Mutka and Livny [15] observed that when a machine was monitored to be (un)available during one instance of a particular hour, it tended to remain so for the entire hour. They concluded that when a machine becomes available and a job is scheduled on it, there is a high probability that it will run there for at least ....

M.W. Mutka, M. Livny, Profiling workstations' available capacity for remote execution. Performance '87, Proc. of the 12th IFIP WG 7.3 Int'l Symp. on the Computer Performance Modeling, Measurement and Evaluation, Brussels, Belgium, pp. 529-544, 1987.

....time. Workstations are only intermittently idle as a rule. A system which can utilize intermittently idle workstations can make computing on networked workstations very cost effective. In fact, for sequential or semi parallel (i.e. multiple tasks with no inter dependency) jobs, there are systems[2, 14, 18, 11, 25, 28, 44, 48, 52, 53] to utilize idle or under utilize workstations effectively. For parallel computation, there are also systems[2, 11, 25, 44] In the last several years, there has been a proliferation of commercial and research prototype parallel software systems on networks of workstations. Popular systems ....

....Their advantages over massively parallel computers are wide availability and cost effectiveness. First, unlike supercomputers installed in a few institutions, these machines are widely available; many institutions have hundreds of high performance workstations which are unused most of the time[25, 52]. Second, they are already paid for and are connected via communication networks; no additional cost is required for parallel processing. Finally, they can rival supercomputers with their aggregate computing power and main memory. However, most machines are private (they are usually sitting on ....

[Article contains additional citation context not shown here]

M.W. Mutka and M. Livny. Profiling workstations' available capacity for remote execution. In Performance '87, pages 529--544. Elsevier Science Publishers B.V., 1988.

....good candidate for task placement. Our method is more precise and we found that having one rule (such as waiting five minutes) for all workstations does not work well. For the workstations in their study they found that their idleness behavior was different from one another. Our results agree. In [3] Mutka and Livny studied workstation availability patterns and fit the available period length distribution to a three stage hyperexponential distribution. They find that the lengths of idle periods tend to be correlated. We find that the idle period length distributions have important features ....

....unpartitioned case does, and the other two predict more ten minute idle periods. 8.4 Using the Length of the Previous Idle Period Another way to try to improve predictions of the remaining length of the current idle period is to use the length of the previous idle period. Previous researchers [3] found the lengths of successive idle periods to be 11 Past and future predictions (1 month) 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0 50 100 150 200 minutes into idle period probability past future Past and future predictions (1 month) 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 ....

M. Mutka and M. Livny. Profiling workstations' available capacity for remote execution. In Performance '87. pg. 529-544.

....personal computers as a single computational resource. Marketing slogans like the network is the computer , invented in the 80s by SUN Microsystems, are now becoming reality. The performance of workstations increases at a dramatic pace, currently doubling in speed every 12 to 18 months. Studies [28] have shown, however, that more than half of the workstations are unused, even during peak hours. One possible use of the idle CPU cycles would be to run parallel applications on networks of workstations that are not used otherwise. Compared to special purpose high performance systems, networks of ....

M.W. Mutka, M. Livny. Profiling workstations' available capacity for remote execution. Performance'87, Procs. 12th IFIP WG 7.3 Symposium on Computer Performance, Brussels Belgium (Dec. 1987).

....that 50 70 work stations out of 350 were idle during the day, and over 100 were idle at night. Mutka and Livny studied the use of MicroVAX II work stations and found over a 5 month period that work stations were available for outside use according to conservative criteria about 70 of the time [ML87a] The machine utilization rates continue to drop. In 1990 Litzkow and Livny [LL90, p. 100] reported that 70 of their machines were idle. In 1991 Douglis and Ousterhout [DO91, p. 759] stated that 66 78 of their machines were idle. Also in 1991 Krueger and Chawla [KC91, p. 342] reported that ....

Mutka and Livny. Profiling workstations' available capacity for remote execution. Performance, pages 529--544, 1987.

....were used to execute parallel ATM network simulators [12] 15] With the advent of recent technology, workstations have become a powerful and yet inexpensive computational resource. In addition, workstations in many academic and commercial organizations are highly avaliable and under utilized [9]. In this paper, a parallelism in which simulation runs are executed by replicated copies of parallel simulation on workstation clusters in parallel is proposed. This approach aims at combining the benefits of the parallel simulation approach and the RSS approach, which are reduction in the ....

M.W. Mutka and M. Livny. Profiling workstations' available capacity for remote execution. In 12th IFIP WG &.3 Symposium on Computer Performance, Brussels, Belgium, 1987.

....1.6.3 Results Table 1.1(a) shows the execution times of maximum independent set problem on transient processors using the adaptive replication scheme with t a = 40 minutes and t n = 20 minutes respectively. These values for t a and t n are within the range of values reported in earlier works (Mutka and Livny, 1987). The measurements were taken on a network of Sun Sparc 5 workstations connected by a 10 Mbps Ethernet. The number of processors available is much larger 12 than the degree of parallelism used in the simulations and therefore, migration to an available processor was always possible. The ....

Mutka, M. W. and Livny, M. (1987). Profiling Workstations' Available Capacity for Remote Execution. In Proc. 12th Symposium on Computer Performance, Brussels, Belgium.

....for clusters of multi user workstations (as is the situation at KSLA) A study of Livny and Melman [Livny82] showed that for a network of autonomous nodes, there is a large probability that a least one node is idle while tasks are queued at some other node. In another study, Mutka and Livny [Mutka87] analyzed the usage patterns of a group of single user workstations in a network. It showed that the nodes were in use approximately 30 of the available time. For a network of multi user workstations the probability that a node is idle is assumed to be smaller, as is the figure for ....

M.W. Mutka and M. Livny, "Profiling workstations' available capacity for remote execution", Performance '87, Proceedings of the 12th IFIP WG 7.3 symposium on computer performance, Brussels, December 1987.

....the probability of load balancing success in distributed systems [32] Subsequently, these findings suggested that load balancing is likely to benefit job response times. Several studies have also been performed on network of workstations to assess, in practice, the potential of idle workstations [27, 23, 28], and these studies for example found that even during daytime only 30 of the total network wide computation capacity is used. Over the past decade, dynamic load sharing and balancing algorithms have been an active area of research [33] Distributed controlled dynamic load balancing algorithms ....

M W Mutka and M Livny, "Profiling Workstations' Available Capacity for Remote Execution," Proceedings of the 12th IFIP WG 7.3 Symposium on Computer Performance, December 1987.

....to obtain results on system performance in best and worst case. There is only few literature reporting on measurement studies of parallel systems under real workload. See for example [VanV 94] where the workload on an iPSC 860 system CHAPTER 2. FRAMEWORK AND METHODOLOGY 24 is investigated, or [Mutk 88] where the load on a network of workstations is observed. Artificial workload An artificial workload mimics the quantitative and or functional behavior of real workloads. In system measurements an executable model has to be used. Consisting of a set of operations, that are either typical for a ....

M. W. Mutka and M. Livny. "Profiling Workstations' Available Capacity For Remote Execution ". In: P.-J. Coutrois and G. Latouche, Eds., Performance'87, pp. 529--544, North-Holland, 1988.

....numerical approximations were used. Of the three dynamic policies, MTP performs best, but the di#erence with MRTP is rather small. Another example of a dynamic central policy is implemented in the condor system developed at the University of Wisconsin, Madison, by Litzkow, Livny, and Mutka [60, 67, 66, 65]. condor is a batch queueing system implementing a central load sharing policy, and it runs on unix TM workstations. The policy is to match idle workstations and jobs waiting for service. In condor, fair access to idle workstations is controlled by the up down algorithm. The jobs submitted for ....

M.W. Mutka and M. Livny. Profiling Workstations' Available Capacity for Remote Execution. In P.-J. Courtois and G. Latouche, editors, Performance '87, pages 529-- 544. North-Holland, 1987.

No context found.

M.W. Mutka and M. Livny, Profiling workstations' available capacity for remote execution, Performance '87, Proc. of the 12th IFIP WG 7.3 Int'l Symp. on Computer Performance Modeling, Measurement and Evaluation, Brussels, Belgium (1987) 529-544.

No context found.

Mutka, M. and Livny, M. "Profiling Workstations' Available Capacity For Remote Execution ". Proceedings of Performance-87, The 12th IFIP W.G. 7.3 International Symposium on Computer Performance Modeling, Measurement and Evaluation. Brussels, Belgium, December 1987.

No context found.

M. W. Mutka and M. Livny. Profiling workstations' available capacity for remote execution. Performance, 1987.

No context found.

M. W. Mutka and M. Livny. Profiling workstations' available capacity for remote execution. Performance, 1987.

No context found.

Mutka, M.W., Livny, M.: Profiling Workstations' Available Capacity for Remote Execution. 12th IFIP WG7.3 International Symposium on Computer Performance (Performance '87) (1987) 529--544

No context found.

M. Mutka and M. Livny. Profiling workstations' available capacity for remote execution. In Proceedings of Performance '87: Computer Performance Modelling, Measurement, and Evaluation, 12th IFIP WG 7.3 International Symposium, December 1987.

No context found.

M. W. Mutka and M. Livny. Profiling workstations' available capacity for remote execution. Performance, 1987.

No context found.

M. Mutka and M. Living, Profiling Workstations' Available Capacity for Remote Execution, Performance'87, Procs. of the 12th IFTP WG7.3 Symposium on Computer Performance, Brussels, December, 1987.

No context found.

M.W. Mutka and M. Livny, "Profiling Workstations' Available Capacity for Remote Execution," in Performance '87, Proc. of the 12th IFIP WG 7.3 Int'l Symp. on Computer Performance Modelling, Measurement and Evaluation, Brussels, Belgium, 1987, pp. 529-544.

No context found.

Mutka, M. and Livny, M. "Profiling Workstations' Available Capacity For Remote Execution ". Proceedings of Performance-87, The 12th IFIP W.G. 7.3 International Symposium on Computer Performance Modeling, Measurement and Evaluation. Brussels, Belgium, December 1987.

No context found.

Mutka, M.W. and Livny, M. "Profiling Workstations' Available Capacity for Remote Execution", Performance '87, pp 529-544, Elsevier Science Publishers B.V. (North Holland), 1988.

First 50 documents

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