J.M. Barton and N. Bitar. A scalable multi-discipline multiple processor scheduling framework for irix. In Job Scheduling Strategies for Parallel Processing, Lecture Notes in Computer Science 949. Springer-Verlag, June 1995.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....of the system need to be replaced ) Moreover, achieving the maturity and stability required of production software is much harder than building a prototyope. Finally, they need to cater to users and administrators with many different needs, leading to the creation of rather elaborate systems [4,44]. As a result of such concerns, there is much interest in standardizing various software components. In recent years, message passing libraries were standardized through the MPI effort. Similarly, the PSCHED proposal aims at standardizing the interactions among various components involved in ....

J. M. Barton and N. Bitar, "A scalable multi-discipline, multiple-processor scheduling framework for IRIX ". In Job Scheduling Strategies for Parallel Processing, Notes in Computer Science Vol. 949.

.... threads should be scheduled to execute together [32] This idea, now known as gang scheduling, is becoming increasingly popular, and can be found in various forms on commercial machines such as the CM 5 from Thinking Machines [29] the Intel Paragon [25] the SGI multiprocessors running IRIX [1], the Meiko CS 2 [14] the Alliant FX 8 [41] and the MasPar and DAP SIMD arrays. Gang scheduling has also been used in a production system on a BBN Butterfly at LLNL [20] which is now being ported to a new Cray T3D machine, and several other experimental systems [13, 39, 4, 17, 6] At first ....

J. M. Barton and N. Bitar, "A scalable multi-discipline, multiple-processor scheduling framework for IRIX ". In Job Scheduling Strategies for Parallel Processing, D. G. Feitelson and L. Rudolph (eds.), pp. 45--69, Springer-Verlag, 1995. Lecture Notes in Computer Science Vol. 949.

....is coordinated across the PEs, such that all the threads in a job are scheduled and de scheduled at the same time. Gang scheduling is a prominent feature of the Connection Machine CM 5 system [28] and is available on the Intel Paragon [17] the Meiko CS 2, and multiprocessor SGI workstations [2]. It has also been used extensively in a home grown system on a BBN Butterfly at Lawrence Livermore Labs [13] which has recently been ported to their new Cray T3D system. The main drawback of using gang scheduling is the problem of fragmentation. Specifically, it may happen that a number of jobs ....

J. M. Barton and N. Bitar, "A scalable multi-discipline, multiple-processor scheduling framework for IRIX ". In Job Scheduling Strategies for Parallel Processing, D. G. Feitelson and L. Rudolph (eds.), pp. 45--69, Springer-Verlag, 1995. Lecture Notes in Computer Science Vol. 949.

....to data structures is the main thing that needs to be done to parallelize the Unix kernel [37] A lockable shared queue is used in the Hydra system on C.mmp [621, chap. 12] in the Dynix system on the Sequent Balance [570] in Mach [65, 68] in IRIX on Silicon Graphics multiprocessor workstations [370, 48], and in the DASH system which is based on IRIX [357, 106] It is also used in thread packages such as Presto [59, 192] and in microtasking on multiprocessor Cray supercomputers [220] Much progress has been made lately regarding the efficient implementation of locks, especially by way of ....

.... a number of other applications have been scheduled [249, 576, 598] It has therefore been suggested that affinity be temporally delimited: it would only be established if a thread runs on a PE for a minimal amount of time, and it would expire after a certain delay when the thread is de scheduled [48]. A simple implementation of affinity scheduling uses local modifications to threads priorities [576, 106] Each thread has a global priority that depends on its accumulated runtime, like processes in a Unix system. However, when a given PE scans the queue and looks for 13 The spawn system call ....

[Article contains additional citation context not shown here]

J. M. Barton and N. Bitar, "A scalable multi-discipline, multiple-processor scheduling framework for IRIX ". In Job Scheduling Strategies for Parallel Processing, D. G. Feitelson and L. Rudolph (eds.), pp. 45--69, Springer-Verlag, 1995. Lecture Notes in Computer Science Vol. 949. 125

.... a number of other applications have been scheduled [152, 339, 348] It has therefore been suggested that affinity be temporally delimited: it would only be established if a thread runs on a PE for a minimal amount of time, and it would expire after a certain delay when the thread is de scheduled [31]. A simple implementation of affinity scheduling uses local modifications to threads priorities [339, 57] Each thread has a global priority that depends on its accumulated runtime, like processes in a Unix system. However, when a given PE scans the queue and looks for the thread with the ....

.... induces the next multicontext switch [122] A variant of this is used in IRIX on SGI multiprocessor workstations: the PE that selects the first member of a gang from the global queue interrupts other PEs that are running low priority processes so that they will schedule the other gang members [31]. The controller coordinates the context switching by causing an operating system trap on all the relevant processors. The requirement on this trap is that the variability in the exact time that it occurs on the different PEs be small relative to the scheduling time quantum. One possibility is to ....

J. M. Barton and N. Bitar, "A scalable multi-discipline, multiple-processor scheduling framework for IRIX ". In Job Scheduling Strategies for Parallel Processing, D. G. Feitelson and L. Rudolph (eds.), pp. 45--69, Springer-Verlag, 1995. Lecture Notes in Computer Science Vol. 949.

....returns it to the tail of the queue. As processors are not allocated permanently to jobs, the number of processes in a job may change during runtime without causing any problems. This scheme is especially suitable for shared memory multiprocessors, and indeed it is used on many bus based systems [50,3]. Using a shared queue as described above may suffer from three drawbacks: contention for the queue, frequent migration of processes, and lack of coordinated scheduling. The issue of possible contention has lead to the design of wait free queues, where different processes can access the queue ....

....interact with each other frequently. The only solution is to use gang scheduling. While gang scheduling and a shared queue seem to be in conflict with each other, a scheme that integrates both has been designed in the context of the IRIX operating system for SGI multiprocessor workstations [3]. Finally, it should be noted that this scheme benefits from similarity with runtime systems and thread packages that run within the confines of a single job. 5.4 Moldable Jobs and Adaptive Partitioning As noted above, variable partitioning is a simple but somewhat inefficient scheduling ....

J. M. Barton and N. Bitar, "A scalable multi-discipline, multiple-processor scheduling framework for IRIX ". In Job Scheduling Strategies for Parallel Processing, D. G. Feitelson and L. Rudolph (eds.), pp. 45--69, Springer-Verlag, 1995. Lecture Notes in Computer Science Vol. 949.

....in this queue. Processors pick the first thread from the queue, execute it for a certain time quantum, and then return it to the queue. This approach is especially common on small scale bus based UMA shared memory machines, such as Sequent multiprocessors [46] and SGI multiprocessor workstations [3], and is also used in the Mach operating system [4] The main merit of a global queue is that it provides automatic load sharing. No processor is idle if there is any waiting thread in the system. However, this comes at a price. One problem is contention for the global queue, which grows with the ....

....and maybe to benefit from sustained cache state. This solution is completely general, and works for any programming model. In fact, it decouples the application from the operating system. It is used by some vendors, e.g. the CM 5 from Thinking Machines [26] the IRIX system on SGI multiprocessors [3], the Intel Paragon [23] and the Meiko CS 2, and has also been studied in academia and research prototypes [32, 13, 19, 22] An interesting variant of gang scheduling is based on the observation that coordinated scheduling is only needed if the job s threads interact frequently [16] Therefore ....

J. M. Barton and N. Bitar, "A scalable multi-discipline, multiple-processor scheduling framework for IRIX ". In Job Scheduling Strategies for Parallel Processing, D. G. Feitelson and L. Rudolph (eds.), Springer-Verlag, 1995. Lecture Notes in Computer Science Vol. 949.

....of the system need to be replaced ) Moreover, achieving the maturity and stability required of production software is much harder than building a prototyope. Finally, they need to cater to users and administrators with many different needs, leading to the creation of rather elaborate systems [4, 45]. As a result of such concerns, there is much interest in standardizing various software components. In recent years, message passing libraries were standardized through the MPI effort. Similarly, the PSCHED proposal aims at standardizing the interactions among various components involved in ....

J. M. Barton and N. Bitar, "A scalable multidiscipline, multiple-processor scheduling framework for IRIX ". In Job Scheduling Strategies for Parallel Processing, D. G. Feitelson and L. Rudolph (eds.), pp. 45--69, SpringerVerlag, 1995. Lecture Notes in Computer Science Vol. 949.

....multicomputers. Mechanisms used to achieve efficient performance from UNIX SVR4 on small scale parallel systems are described in [SPY 93, Pea92, CBB 91, CHS91] Mechanisms in the Sun Microsystems versions of UNIX SVR4 are described in [EKB 92] Work on the Silicon Graphics version is described in [BaB95]. These efforts are all focused on achieving sufficient parallelization to avoid synchronization bottlenecks on four or eight processor systems. The largest scale systems constructed to date that provide standard SMP semantics are nonshared memory multicomputers. The Mach SMP kernel has been ....

J.M. Barton and N. Bitar. "A scalable multi-discipline, multiple-processor scheduling framework for IRIX." Workshop on Job Scheduling Strategies for Parallel Processing, pp. 45--69 (Santa Barbara, CA, April 25, 1995). Berlin, Germany: Springer-Verlag, 1995.

No context found.

J.M. Barton and N. Bitar. A scalable multi-discipline multiple processor scheduling framework for irix. In Job Scheduling Strategies for Parallel Processing, Lecture Notes in Computer Science 949. Springer-Verlag, June 1995.

No context found.

J.M. Barton and N. Bitar. A scalable multidiscipline multiple processor scheduling framework for IRIX. Job Scheduling Strategies for Parallel 949, 1995.

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