Dror G. Feitelson. Packing Schemes for Gang Scheduling. In Dror G. Feitelson and Larry Rudolph, editors, Job Scheduling Strategies for Parallel Processing, volume 1162 of Lecture Notes in Computer Science, pages 89--110. Springer, 1996.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....when multiprogramming a set of parallel jobs on a multiprocessor machine [22] Coscheduling strives to ensure that all processes belonging to a job are scheduled at the same time. Subsequent work has generalized and refined the coscheduling (now often called gang scheduling) concept [5, 6, 8, 9, 11, 14, 15, 19, 23, 24, 25, 26]. Gang scheduling schemes are a practical result of the multiprocessor scheduling community and have been adapted for inclusion in several production systems including the Intel Paragon [4] CM 5 [3] Meiko CS 2, multiprocessor SGI workstations [1] and the IBM SP2 [17, 18] Work has been done to ....

....reporting mean slowdowns and response times. We provide numerous simulation results to make a case for proper quantum allocation. In addition to considering the DHC algorithm, we consider the effect of different quantum allocations with simpler gang scheduling schemes such as Matrix [22] and LRS [5]. These simpler algorithms are especially relevant since many current production level schedulers use variants of these simpler algorithms. Finally, our work also provides a more exhaustive comparison of DHC with Matrix and LRS by considering more workloads than considered in [5] and comparing DHC ....

[Article contains additional citation context not shown here]

D. Feitelson. Packing schemes for gang scheduling. In Proc. of the IPPS'96 Workshop on Job Scheuling Strageies for Parallel Processing, pages 65--88. Springer LNCS #1162, 1996.

....Some approaches try to estimate the runtime of future jobs based on the past [15, 16] Results of an optimal o#ine schedule (submit time, runtime, and scalability are known in advance) are used as a benchmark for our approach. A combination of time and space sharing is called gang scheduling [4, 5]. For a specific amount of time (gang) space sharing is used to assign (and withdraw) processors to jobs, and also jobs are scheduled simultaneously, which means that jobs only start at the beginning of a gang. Di#erent gangs are executed in timesharing manner. Small jobs may have to wait for the ....

D. G. Feitelson. Packing schemes for gang scheduling. Lecture Notes in Computer Science, 1162:89--101, 1996.

.... Manager or NM (one daemon per compute node) and the Program Launcher or PL (several daemons per compute node) The MM is in charge of resource allocation for jobs (both in space and time) Whenever a new job arrives, the MM queues it and tries to allocate PEs to it (using a buddy tree algorithm [11, 12]) If the scheduling policy allows for multiprogramming (e.g. GS) the PEs are allocated in any time slot that has enough available resources. After a successful allocation, the MM broadcasts a job launch message to all the NMs, and those NMs on nodes that are allocated to the job will launch it ....

Dror G. Feitelson. Packing Schemes for Gang Scheduling. In Dror G. Feitelson and Larry Rudolph, editors, Job Scheduling Strategies for Parallel Processing -- Proceedings of the IPPS'96 Workshop, volume 1162, pages 89--110. Springer, 1996.

.... Serve algorithm in combination with Backfilling ( 17, 13] where each step consists of the method described above. Otherwise a single machine with enough free resource is searched. Upon a successful search the job is started on this machine, which is chosen by using the BestFit strategy [8], immediately. A search failure leads to the check of the parameter lower bound e.g. a value of 8 implies that only jobs requesting more than 8 resources may be split up. Next, the needed number of fragments is calculated by using a machine list ordered by the decreasing number of free resources. ....

D.G. Feitelson. Packing Schemes for Gang Scheduling. Lecture Notes in Computer Science, 1162:89--101, 1996.

....runtime. We therefore used the actual runtime in the simulations. This is equivalent to assuming that the estimates are perfect. In addition, we also use a log of jobs run on the San Diego Supercomputer Center (SDSC) IBM SP2 from May 1998 through April 2000 , and a model proposed by Feitelson [5]. As can be seen from the graphs, these two workloads have a similar distribution of job sizes which has fewer small jobs than the CTC and Jann workloads. The Feitelson model is also unique in having many more short jobs. The SDSC log also has shorter jobs than CTC and Jann, but not as short as in ....

....the distribution of sizes emphasized serial jobs and jobs in the range of sizes from 8 to 31, and serial jobs also received special treatment in terms of making them longer. Note that this is at odds with the original Feitelson model, in which a weak correlation exists between job size and runtime [5]. Simulating the behavior of the EASY and conservative schedulers on this modified Feitelson workload indeed showed behavior similar to that of the Jann workload. However, the magnitude of the effect was smaller. Based on our current understanding of the matter, we can suggest a simpler ....

D. G. Feitelson, "Packing schemes for gang scheduling ". In Job Scheduling Strategies for Parallel Processing, pp. 89--110, Springer-Verlag, 1996. Lect. Notes Comput. Sci. vol. 1162.

....will be invoked in order to find another slot for the removed job. If no such slot is found, the job will be put in a new slot. In our tests we have used M = 20 . In effect, the range of each slot is then bounded as in max(slot) min(slot) 20 . 2. 6 Alternate Scheduling Alternate Scheduling [3] can improve gang scheduling. Using this method we look for idle nodes in a slot. If we find some, we will try to execute a job or even several jobs which need these nodes or part of them. When using paired gang scheduling, we should take into account the characteristic of the alternate scheduled ....

Feitelson D. G., Packing schemes for gang scheduling. In Job Scheduling Strategies for Parallel Processing, Springer-Verlag, LNCS Vol. 1162, pp. 89--110, 1996.

....of jobs. It is useful because many parallel supercomputers provide schedulers for this type of jobs, and because this type of workload model can be derived from accounting logs. Indeed, a number of such models have already been derived and used in the evaluation of schedulers for parallel systems [8, 15, 18]. 0 5 10 15 20 25 percentage of SDSC Paragon 0 0.5 1 job size Fig. 4. Representative distribution of job sizes, showing dominance of small jobs and jobs using a power of two processors. Data from SDSC Paragon. Interestingly, the accounting logs from many diverse systems show ....

....5) This has two possible interpretations: either the smaller jobs are development while the larger ones are production runs, or parallelism is used to solve larger problems rather than to achieve speedup on given problems. A possible model for such a correlation has been proposed by Feitelson [8]. The basis for the model is the observation that job runtimes have a very large variability, manifested by a coefficient of variation that is larger than 1. A plausible model for runtimes is therefore a hyperexponential distribution. For example, a two stage hyperexponential can be used; ....

D. G. Feitelson, "Packing schemes for gang scheduling ". In Job Scheduling Strategies for Parallel Processing, D. G. Feitelson and L. Rudolph (eds.), pp. 89--110, Springer-Verlag, 1996. Lect. Notes Comput. Sci. vol. 1162.

.... it possible to allocate each job an appropriate number of processors to make it operate at a near optimal ratio of execution time to efficiency [16] With the knowledge of how many processors each job uses, policies for packing the jobs into frames for gang scheduling are investigated by Feitelson [18]. Feitelson and Rudolph [22] describe a discipline in which processes that communicate frequently are identified, and it is assured that the corresponding threads are all activated at the same time. Similar schemes in which co scheduling is triggered by communication events were described by ....

D. G. Feitelson, "Packing schemes for gang scheduling ". In Job Scheduling Strategies for Parallel Processing, D. G. Feitelson and L. Rudolph (eds.), pp. 89--110, Springer-Verlag, 1996. Lecture Notes in Computer Science Vol. 1162.

.... 18, 6] Other examples include studies of process arrivals and runtimes [12, 37] le systems [36] and video streams [48] In the area of parallel systems, descriptive studies of workloads have only started to appear in recent years [29, 76, 58, 27, 14] There are also some attempts at modeling [10, 28, 21, 41, 23, 54, 15] and on line characterization [34] But where does the data come from There are two main options: use data that is available anyway, or collect data speci cally for the workload model. The latter can be done in two ways: active or passive instrumentation. Importantly, collected data can and ....

....subranges, and de ne a functional dependency of the model parameters on the subrange. For example, the Feitelson model rst selects the size of each job according to the distribution of job sizes, and then selects a runtime from a distribution of runtimes that is conditioned on the selected size [28]. Speci cally, the runtime is selected from a two stage hyperexponential distribution, and the probability for using the exponential with the higher mean is linearly dependent on the size: p(n) 0:95 0:2(n=N) 12 Thus, for small jobs (the job size n is small relative to the machine size N) the ....

D. G. Feitelson, \Packing schemes for gang scheduling". In Job Scheduling Strategies for Parallel Processing, D. G. Feitelson and L. Rudolph (eds.), pp. 89-110, Springer-Verlag, 1996. Lect. Notes Comput. Sci. vol. 1162.

....Distributed Hierarchical Control scheme [5] which uses a buddy system to allocate processors in blocks that are powers of 2. Seven different workloads were used, of which four were models and three were actual workloads from production systems. The models used were those proposed by Feitelson [4], Jann [7] Downey [2] and Lublin [10] The Feitelson workload, when generated with the default random seed, caused problems when trying to modify the load; using a different seed produced a workload that was more manageable. The Downey workload caused severe problems for the gang scheduler ....

D. G. Feitelson, "Packing schemes for gang scheduling ". In Job Scheduling Strategies for Parallel Processing, pp. 89--110, Springer-Verlag, 1996. Lect. Notes Comput. Sci. vol. 1162.

.... 1995 to December 1996) CM5 : The Los Alamos National Lab 1024 node Connection Machine CM 5 (201387 jobs from October 1994 to September 1996) Workload models developed based on these and other traces: Feitelson : a general model based on data from 6 di erent traces, including CTC and Par above [4] (350000 jobs) Jann : a model developed speci cally for the CTC trace [14] 100000 jobs) All these workloads are available on line from the Parallel Workloads Archive [22] Only the rst three logs contain actual user estimates of runtime. In other cases, accurate estimates are assumed (that is, ....

D. G. Feitelson, \Packing schemes for gang scheduling". In Job Scheduling Strategies for Parallel Processing, D. G. Feitelson and L. Rudolph (eds.), pp. 89-110, Springer-Verlag, 1996. Lect. Notes Comput. Sci. vol. 1162.

....of runtime. We therefore used the actual runtime in the simulations. This is equivalent to assuming that the estimates are perfect. In addition, we also use a log of jobs run on the San Diego Supercomputer Center (SDSC) IBM SP2 from May 1998 through April 2000, and a model proposed by Feitelson [2]. As can be seen from the graphs, this model is decidedly distinct from the other three. Specifically, it has much fewer small jobs, but many more short jobs. The SDSC log has a similar size distribution to CTC and Jann, but shorter jobs. However, they are not as short as in the Feitelson model. ....

....the distribution of sizes emphasized serial jobs and jobs in the range of sizes from 8 to 31, and serial jobs also received special treatment in terms of making them longer. Note that this is at odds with the original Feitelson model, in which a weak correlation exists between job size and runtime [2]. Simulating the behavior of the EASY and conservative schedulers on this modified Feitelson workload indeed showed behavior similar to that of the Jann workload. However, the magnitude of the effect was smaller. Based on our current understanding of the matter, we can suggest a simpler ....

D. G. Feitelson, "Packing schemes for gang scheduling ". In Job Scheduling Strategies for Parallel Processing, pp. 89--110, Springer-Verlag, 1996. Lect. Notes Comput. Sci. vol. 1162.

No context found.

Dror G. Feitelson. Packing Schemes for Gang Scheduling. In Dror G. Feitelson and Larry Rudolph, editors, Job Scheduling Strategies for Parallel Processing, volume 1162 of Lecture Notes in Computer Science, pages 89--110. Springer, 1996.

No context found.

D.G. Feitelson. Packing Schemes for Gang Scheduling. Lecture Notes in Computer Science, 1162:89-- 101, 1996.

No context found.

Feitelson DG. Packing schemes for gang scheduling. Proceedings of the Job Scheduling Strategies for Parallel Processing (Lecture Notes in Computer Science, vol. 1162). Springer: Berlin, 1996.

No context found.

D. G. Feitelson, "Packing schemes for gang scheduling," in Job Scheduling Strategies for Parallel Processing, D. G. Feitelson and L. Rudolph, Eds. Springer-Verlag, 1996, vol. 1162, pp. 89--110, Lecture Notes in Computer Science.

No context found.

D. G. Feitelson, "Packing Schemes for Gang Scheduling", Job Scheduling Strategies for Parallel Processing, pp. 89-110, Springer-Verlag, 1996. Lectures Notes in Computer Science, vol. 1162.

No context found.

D. G. Feitelson, Packing schemes for gang scheduling, In Job Scheduling Strategies for Parallel Processing, D. G. Feitelson and L. Rudolph (eds.), Lecture Notes Computer Science, Vol. 1162, Springer-Verlag, 1996, pp.89-110.

No context found.

D. G. Feitelson, Packing schemes for gang scheduling, In Job Scheduling Strategies for Parallel Processing, D. G. Feitelson and L. Rudolph (eds.), Lecture Notes Computer Science, Vol. 1162, Springer-Verlag, 1996, pp.89-110.

No context found.

Dror G. Feitelson. Packing schemes for gang scheduling. In Dror G. Feitelson and Larry Rudolph, editors, Job Scheduling Strategies for Parallel Processing, pages 89--110. Springer-Verlag, 1996. Lect. Notes Comput. Sci. vol. 1162.

No context found.

D. G. Feitelson. Packing schemes for gang scheduling. In Job Scheduling Strategies for Parallel Processing, IPPS'96 Workshop, volume 1162 of Lecture Notes in Computer Science, pages 89--110, Berlin, March 1996. Springer-Verlag.

No context found.

Dror G. Feitelson. Packing schemes for gang scheduling. In Dror G. Feitelson and Larry Rudolph, editors, Job Scheduling Strategies for Parallel Processing, pages 89--110. Springer-Verlag, 1996. Lect. Notes Comput. Sci. vol. 1162.

No context found.

D. G. Feitelson, "Packing Schemes for Gang Scheduling", Job Scheduling Strategies for Parallel Processing, Lecture Notes in Computer Science, vol. 1162, pp. 89-110, Springer-Verlag, 1996.

No context found.

D. G. Feitelson. Packing schemes for gang scheduling. In Job Scheduling Strategies for Parallel Processing, Lecture Notes in Computer Science vol. 1162, D. G. Feitelson and L. Rudolph (eds.), pp. 89-110, Springer-Verlag, 1996.

No context found.

Dror Feitelson. Packing schemes for gang scheduling. In Job Scheduling Strategies for Parallel Processing, Lecture Notes in Computer Science vol. 1162, Dror Feitelson and Larry Rudolph (eds.), pp. 89-110, Springer-Verlag, 1996.

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