Sobalvarro PG, Weihl WE. Demand-based coscheduling of parallel jobs on multiprogrammed multiprocessors. Proceedings of the Parallel Job Scheduling Workshop at IPPS (JSSPP) (Lecture Notes in Computer Science, vol. 949). Springer: Berlin, 1995.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....are scheduled simultaneously, thereby giving the computation the illusion of running on a dedicated machine. Interestingly, it has recently been shown that in networks of workstations coscheduling can be achieved with little or no modification to existing multiprocessor operating systems [17, 35]. Unfortunately, for some job mixes, coscheduling is not appropriate. For example, a job mix consisting of one parallel computation and one serial computation cannot be coscheduled efficiently. With process control [36] processors are dynamically partitioned among the running computations so that ....

Patrick G. Sobalvarro and William E. Weihl. Demand-based coscheduling of parallel jobs on multiprogrammed multiprocessors. In Proceedings of the IPPS '95 Workshop on Job Scheduling Strategies for Parallel Processing, pages 106--126, April 1995.

....on the Tera MTA. But that application is for a massively parallel system which largely protects threads from each other. Thus, while the scale of the scheduling problem is great, the number of factors determining how threads interact are few and relatively straight forward. Sobalvarro and Weihl [26], Gupta, et al. 11] and Dusseau, et al. 4] all explore the benefits of coscheduling parallel jobs based on their communication patterns. In fact, a multiprocessor scheduler should solve a similar problem how to coschedule threads on different processors to maximize efficiency in the face ....

P. G. Sobalvarro and W. E. Weihl. Demand-based coscheduling of parallel jobs on multiprogrammed multiprocessors. In IPPS95, pages 63--75, April 1995.

....but run at the lowest priority when local user jobs exist. This method avoids the process migration, but does not guarantee the performance of parallel jobs. Several variations of coscheduling have been proposed to reduce the overhead of coscheduling. For example, Sobalvarro and Weihl [14] propose a demand based coscheduling policy to schedule parallel processes simultaneously only if they communicate. There are two options: dynamic coscheduling and predictive coscheduling. When parallel processes are dynamically coscheduled, an arriving message will cause the targeted process at ....

Sobalvarro, P. G., and Weihl, W. E. Demand-based co-scheduling of parallel jobs on multiprogrammed multiprocessors. Proceedings of the IPPS'95 Workshop on Job Scheduling Strategies for Parallel Processing. 1995, pp. 63--75.

....methods with some simple schedulers, which are described in sections 2.5 2.7. 2. 2 Dynamic and Implicit Coscheduling Dynamic coscheduling (DCS) was proposed as a way to reduce the amount of global synchronization of coscheduling, and thus decrease the system s overhead and improve its sealability [19, 26, 27]. The philosophy behind DCS is demand based coscheduling: coordination is achieved by observing the communication between processes, and not by a master daemon, Communication between processes is used to deduce which processes should be coscheduled and to effect coscheduling. Thus, when a ....

Patrick Sobalvarro and William E. Weihl. Demand-Based Coscheduling of Parallel Jobs on Multiprogrammed Multiprocessors. In Proceedings of the 9th International Parallel Processing Symposium, IPPS'95, Santa Barbara, CA, April 1995.

....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 Sobalvarro and Weihl [83] and by Dusseau, Arpaci, and Culler [15] Taking system load and minimum and maximum parallelism of each job into account as well, still higher throughputs can be sustained [77] Chiang et al. 8] show that use of knowledge of some job characteristics plus permission to use a single preemption ....

P. G. Sobalvarro and W. E. Weihl, "Demand-based coscheduling of parallel jobs on multiprogrammed multiprocessors". In Job Scheduling Strategies for Parallel Processing, D. G. Feitelson and L. Rudolph (eds.), pp. 106--126, Springer-Verlag, 1995. Lecture Notes in Computer Science Vol. 949.

....on the Tera MTA. But that application is for a massively parallel system which largely protects threads from each other. Thus, while the scale of the scheduling problem is great, the number of factors determining how threads interact are few and relatively straight forward. Sobalvarro and Weihl [27], Gupta, et al. 11] and Dusseau, et al. 4] all explore the benefits of coscheduling parallel jobs based on their communication patterns. In fact, a multiprocessor scheduler should solve a similar problem how to coschedule threads on different processors to maximize efficiency in the face ....

P. G. Sobalvarro and W. E. Weihl. Demand-based coscheduling of parallel jobs on multiprogrammed multiprocessors. In Proceedings of the IPPS 1995.

....scheduling is not a very attractive scalable option for off the shelf clusters, since it requires periodic synchronization across the nodes to coordinate the effort. Longer time quanta to offset this cost can decrease the responsiveness of the system. As a result, there have been recent efforts [5, 19, 15] to design a new class of scheduling mechanisms broadly referred to as dynamic coscheduling which approximate coscheduled execution without explicitly synchronizing the nodes. These techniques use local events (e.g. message arrival) to estimate what is happening at remote nodes, and ....

....difficult to perform an extensive performance evaluation across the entire spectrum. As a result, it is difficult to say which approaches perform best, and under what conditions, and thus many important design and performance questions remain open. Previous studies have used a set of static jobs [19, 5, 15] or a very specific dynamic arrival pattern [24] on a small scale parallel system. While one could use experimentation [15] and simulation [1, 24] to study small scale systems, the suitability of these mechanisms for large scale systems has not been explored, which is one of the motivating factors ....

[Article contains additional citation context not shown here]

P.G. Sobalvarro. Demand-based coscheduling of parallel jobs on multiprogrammed multiprocessors. PhD thesis, Dept. of Elec. Eng. & Comp. Sci., MIT, Cambridge, MA, 1997.

....related to local task control, as performed by the NLS. Depending on the level of inter NLS synchronization, different scheduling variations are possible. Explicit gang scheduling systems always run all the tasks of a job simultaneously [5 10, 17, 21] In contrast, communication driven systems [3, 19, 20] are more loosely coupled, and schedule tasks based on message arrival. Independent of the inter task scheduling approach, we address those cases in which all processes within a task are closely coupled, like the csh example of the first paragraph. In these situations, all processes of a task must ....

P. G. Sobalvarro and W. E. Weihl. Demand-Based Coscheduling of Parallel Jobs on Multiprogrammed Multiprocessors. In IPPS'95 Workshop on Job Scheduling Strategies for Parallel Processing, volume 949 of Lecture Notes in Computer Science, pages 106--126. Springer-Verlag, April 1995.

....users may lower the effectiveness of this method. In a time slicing environment, two issues must be addressed: how to coordinate the simultaneous execution of the processes of a parallel job, and how to manage the interaction between parallel and local user jobs. One alternative is coscheduling [12, 14, 22]. This work was supported by the CICYT under contract TIC98 0433 Coscheduling ensures that no process will wait for a non scheduled process for synchronization communication and will minimize the waiting time at the synchronization points. Thus, coscheduling may be applied to reduce messages ....

....to reduce messages waiting time and make good use of the idle CPU cycles when distributed applications are executed in a cluster or NOW system. Coscheduling decisions are made taking implicit runtime information of the jobs into account, basically execution CPU cycles and communication events [12, 13, 14, 15, 16, 22]. Our framework will be focused on an implicit coscheduling environment, such as scheduling the correspondents the most recent communicated processes in the overall system at the same time, taking into account both high message communication frequency and low penalty introduction into the delayed ....

[Article contains additional citation context not shown here]

P.G. Sobalvarro and W.E. Weihl. "Demand-based Coscheduling of Parallel Jobs on Multiprogrammed Multiprocessors". IPPS'95 Workshop on Job Scheduling Strategies for Parallel Processing, 1995.

....on the Tera MTA. But that application is for a massively parallel system which largely protects threads from each other. Thus, while the scale of the scheduling problem is great, the number of factors determining how threads interact are few and relatively straight forward. Sobalvarro and Weihl [26], Gupta, et al. 12] and Dusseau, et al. 4] all explore the benefits of coscheduling parallel jobs based on their communication patterns. In fact, a multiprocessor scheduler should solve a similar problem how to coschedule threads on different processors to maximize efficiency in the face ....

P. G. Sobalvarro and W. E. Weihl. Demand-based coscheduling of parallel jobs on multiprogrammed multiprocessors. In IPPS95, pages 63--75, Apr. 1995.

....actions to process the information delivered by a control message. Due to the excessive load imposed on the central manager, the gang scheduler does not scale well to a very large system. Another method for achieving coscheduling is a decentralized scheme called dynamic coscheduling (DCS) [22, 27, 28]. In DCS, the coordinated scheduling of processes that constitute a parallel job is performed independently by the local scheduler, with no centralized control. Since there is no xed schedule to follow in DCS, the local scheduler must rely on certain local events to determine when and which ....

....status information among local schedulers is not a scalable solution. An alternative is to use certain implicit local information to infer the status of remote processes. Such implicit information includes response time, message arrival, and scheduling progress [2] Like all the previous work [2, 22, 27, 28], our coscheduler depends upon message arrival to infer status information of remote processes. The message arrival refers to the receipt of a message from a remote node. When a message is received, this implies the sending process is highly likely to be currently scheduled. Therefore, it is ....

P. G. Sobalvarro and W. E. Weihl. Demandbased Coscheduling of Parallel Jobs on Multipr ogrammed Multiprocessors. In Proc. IPPS'95 Workshop on Job Scheduling Strategies for Parallel Processing, pages 63-75, Apr. 1995.

....actions to process the information delivered by a control message. Due to the excessive load imposed on the central manager, the gang scheduler does not scale well to a very large system. Another method for achieving coscheduling is a decentralized scheme called dynamic coscheduling (DCS) [22, 27, 28]. In DCS, the coordinated scheduling of processes that constitute a parallel job is performed independently by the local scheduler, with no centralized control. Since there is no xed schedule to follow in DCS, the local scheduler must rely on certain local events to determine when and which ....

....scheduled. This implies is that upon receiving a message, the local scheduler should schedule the receiving process immediately, if not already scheduled, to coschedule both the sending and receiving processes. A few experimental scheduling systems based on this method have been developed [22, 27]. All of these prototypes are implemented in an NOW environment, where workstations are interconnected through fast switches like Myrinet [21] Interprocess communication is carried out using high performance user level messaging layers that support user space to user space communication [24, 29, ....

[Article contains additional citation context not shown here]

P. G. Sobalvarro. Demand-based Coscheduling of Parallel Jobs on Multiprogrammed Multiprocessors. PhD thesis, Dept. of Electrical Engineering and Compuer Science, Massachusetts Institutute of Technology, 1997.

....and the less efficiently it can utilize non gang, fragmented resources. Our goal is to identify characteristics of an application which relate to its degree of benefit from gang scheduling. Many studies have measured the benefits of gang scheduling relative to uncoordinated scheduling [1, 3, 9, 17]. Our study differs in that we are interested in the marginal benefit of a pure gang scheduled environment when compared to a gang scheduled environment with disruptions. In order to get a quantification tool, we measure the performance of applications under various near gang scheduled ....

....effect. To avoid scheduler induced load imbalances, interprocess fairness is a good criteria to keep in mind when a gang scheduler does alternate scheduling. The failure to do so could be a reason why research has found that random alternate scheduling provides little gain to system throughput [17]. To first order, there are two types of blocking communication. One is the barrier like communication used to check that processes have all reached a checkpoint. The other type is the request reply type message, where one process is trying to obtain information which can be readily provided by ....

[Article contains additional citation context not shown here]

P. G. Sobalvarro and W. E. Weihl. Demand-based Coscheduling of Parallel Jobs on Multiprogrammed Multiprocessors. In Lecture Notes in Computer Science, 949, pages 106--126, Santa Barbara, 1995. Springer Verlag. Workshop on Parallel Job Scheduling, IPPS '95.

....speed have not been matched by improvements in network latencies, limiting Chapter 2. Background 32 the speedups achievable [PBS97] Sobalvarro et al. SPWC98] study coscheduling on workstation clusters. They use a mechanism known as dynamic co scheduling (first studied by Sobalvarro et al. SW95] to achieve strong co scheduling behaviour. Threads that are not running when a message arrives are awakened and run. Transitive message passing relationships can tend to co schedule groups of communicating processes without explicit identification of the communicating groups. This approach has ....

Patrick G. Sobalvarro and William E. Weihl. Demand-based Coscheduling of Parallel Jobs on Multiprogrammed Multiprocessors. In Proceedings of the IPPS '95 Workshop on Job Scheduling Strategies for Parallel Processing, pages 63--75, April 1995.

.... However, the performance of ne grained communication jobs can be orders of magnitude worse than with explicit coscheduling because the scheduling is not coordinated across processors [5] An intermediate approach developed at UC Berkeley and MIT in recent years is implicit or dynamic coscheduling [1, 15]. With implicit coscheduling, each local scheduler makes independent decisions that dynamically coordinate the scheduling actions of cooperating processes across processors. These actions are based on local events that occur naturally within communicating applications. For example, on message ....

....and scheduling overhead. By delaying the communication, we allow for the global scheduling of the communication pattern. 2) A strobing mechanism performs a total exchange of information at the end of each time slice so that multiprocessor machines may move away from isolated scheduling algorithms [1, 15] (where processors make decisions based solely on their local status and a limited view of the remote status) to more outward looking or global scheduling algorithms. 3) Finally, we propose to use non blocking and one sided communication primitives to decouple communication and synchronization in ....

Patrick Sobalvarro and William E. Weihl. DemandBased Coscheduling of Parallel Jobs on Multiprogrammed Multiprocessors. In Proceedings of the 9th International Parallel Processing Symposium, IPPS '95, Santa Barbara, CA, April 1995.

....they are added to parallel processing as overhead every time, they are considered negligible because they account for 0.4 of 200 ms. This is also considered negligible when compared with the time lag of synchronization. 5 Related Research Implicit co scheduling[3] and dynamic co scheduling[9, 10] are resemble to our moderate co scheduling. Implicit co scheduling and dynamic co scheduling are demand based co scheduling, that is, they relay on a local scheduler to run the parallel process when it receive a message. They suppose that the scheduler quickly dispatches parallel processes after ....

P.G. Sobalvarro and W.E. Weihl. Demand-based coscheduling of parallel jobs on multiprogrammed multiprocessors. IPPS'95 Workshop on Jpb Scheduling Strategies for Parallel Processing (Lecture Notes in Computer Science 949), pages 106-- 126, 1995.

....methods with some simple schedulers, which are described in sections 2.5 2.7. 2. 2 Dynamic and Implicit Coscheduling Dynamic coscheduling (DCS) was proposed as a way to reduce the amount of global synchronization of coscheduling, and thus decrease the system s overhead and improve its scalability [19, 26, 27]. The philosophy behind DCS is demand based coscheduling: u coordination is achieved by observing the communication between processes, and not by a master daemon, u Communication between processes is used to deduce which processes should be coscheduled and to e ect coscheduling. Thus, when a ....

Patrick Sobalvarro and William E. Weihl. Demand-Based Coscheduling of Parallel Jobs on Multiprogrammed Multiprocessors. In Proceedings of the 9th International Parallel Processing Symposium, IPPS'95, Santa Barbara, CA, April 1995.

....a multiprogrammed environment. On the other hand, many researchers have considered the problem of scheduling a parallel application in a parallel or distributed system. The goal here is essentially to reduce communication delay by ensuring that cooperating processes are scheduled at the same time [22, 25, 6, 28]. Ousterhout [22] shows that coscheduling is desirable and describes di erent algorithms for accomplishing it. Ousterhout s work assumes that scheduling decisions are made centrally, by a single operating system running across all processors. Sobalvarro et al. 25] and Dusseau et al. 6] assume ....

....at the same time [22, 25, 6, 28] Ousterhout [22] shows that coscheduling is desirable and describes di erent algorithms for accomplishing it. Ousterhout s work assumes that scheduling decisions are made centrally, by a single operating system running across all processors. Sobalvarro et al. [25] and Dusseau et al. 6] assume a loosely coupled system and make scheduling decisions based on communication patterns. Tucker et al. 28] propose keeping the number of runnable processes of a single application equal to the number of available processors. We argue that neither pure load ....

[Article contains additional citation context not shown here]

P. G. Sobalvarro. Demand-based Coscheduling of Parallel Jobs on Multiprogrammed Machines. PhD thesis, M.I.T., January 1997.

....idle periods that are long relative to processor reallocation overheads. Feitelson and Rudolph [8] take a similar approach to ours, proposing to dynamically gather information about communicating sets of processes in an attempt to relax the constraints of co scheduling. Sobalvarro and Weihl [25] also propose several ways to use runtime identification of sets of communicating processes to relax the constraints of coscheduling. 3 Measuring Job Efficiency and Speedup at Runtime 3.1 Measuring Efficiency and Speedup The basic parallel job characteristics we wish to exploit are efficiency ....

P. B. Sobalvarro and W. E. Weihl. Demand-based Coscheduling of Parallel Jobs on Multiprogrammed Multiprocessors. In Proceedings of the IPPS'95 Workshop on Job Scheduling Strategies for Parallel Processing, pages 106--126, Apr. 1995.

....avenue for achieving coordinated parallel scheduling in an environment that coexists with autonomous node schedulers. 1 Introduction Coordinated scheduling of parallel jobs across the nodes of a multiprocessor is well known to produce benefits in both system and individual job efficiency [12, 18, 4, 5, 17, 3]. Without coordinated scheduling, the processes constituting a parallel job suffer high communication latencies because of processor thrashing [12] While multiprocessor systems typically address these problems with a mix of batch, gang, and timesharing scheduling (based on kernel scheduler ....

....to be had in using existing commercial operating systems, so we restrict ourselves here to approaches that involve augmentation of existing operating system infrastructure. The approach to coordinated scheduling that we use is a form of demandbased coscheduling called dynamic coscheduling (DCS) [17, 16], which achieves coordination by observing the communication between threads. This is a bottomup, emergent scheduling approach that exploits the key observation that only those threads which are communicating need be coscheduled. This approach can achieve coscheduling without changes to the ....

[Article contains additional citation context not shown here]

Patrick G. Sobalvarro and William E. Weihl. Demand-based coscheduling of parallel jobs on multiprogrammed multiprocessors. In Proceedings of the Parallel Job Scheduling Workshop at IPPS '95, 1995. Available from http://www.psg.lcs.mit.edu/~pgs/papers/jsw-for-springer.ps. Also appears in Springer-Verlag Lecture Notes in Computer Science, Vol. 949.

....to be had in using existing commercial operating systems, so we restrict ourselves here to approaches that involve augmentation of existing operating system infrastructure. The approach to coordinated scheduling that we use is a form of demandbased coscheduling called dynamic coscheduling (DCS) [17, 16], which achieves coordination by observing the communication between threads. This is a bottomup, emergent scheduling approach that exploits the key observation that only those threads which are communicating need be coscheduled. This approach can achieve coscheduling without changes to the ....

....consist of thread collections, explicitly indicating which threads should be coscheduled. A variety of implicit schemes which do not require explicit programmer annotation have been explored. On distributed memory systems, the need for coscheduling has typically been associated with communication [17, 6, 3, 16]. Feitelson s Runtime Activity Working Set Identification (RAWSI) monitors the communication between processes or threads 3 to determine their rate of communication. Working sets of processes (which require coscheduling) are identified based on their rate of communication. RAWSI collects the ....

[Article contains additional citation context not shown here]

Patrick G. Sobalvarro. Demand-based Coscheduling of Parallel Jobs on Multiprogrammed Multiprocessors. PhD thesis, Massachusetts Institute of Technology, 1997. MIT/LCS/TR-710.

No context found.

Sobalvarro PG, Weihl WE. Demand-based coscheduling of parallel jobs on multiprogrammed multiprocessors. Proceedings of the Parallel Job Scheduling Workshop at IPPS (JSSPP) (Lecture Notes in Computer Science, vol. 949). Springer: Berlin, 1995.

No context found.

Patrick G. Sobalvarro and William E. Weihl. Demand-based Coscheduling of Parallel Jobs on Multiprogrammed Multiprocessors. In Proceedings of the IPPS '95 Workshop on Job Scheduling Strategies for Parallel Processing, pages 63--75, April 1995.

No context found.

Patrick G. Sobalvarro. Demand-based Coscheduling of Parallel Jobs on Multiprogrammed Multiprocessors. PhD thesis, Massachusetts Institute of Technology, January 1997.

No context found.

P. G. Sobalvarro and W. E. Weihl. Demand-based coscheduling of parallel jobs on multiprogrammed multiprocessors. In D. G. Feitelson and L. Rudolph, editors, Job Scheduling Strategies for Parallel Processing, pages 106--126. SpringerVerlag, 1995.

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