Bodhisattwa Mukherjee and Karsten Schwan. Experiments with a configurable lock for multiprocessors. In Proc. of the Twenty Second International Conference on Parallel Processing, volume 2, pages 205--208, August 1993.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....of applications. The problem can be serious when the optimizations embedded in the operating system are particularly bad for the new application. For instance, the usual paging policy provided by most operating systems is simply not appropriate for database applications [28] Mukherjee and Schwan [20] showed that both spinlocks and blocking locks can provide superior performance under different circumstances. Thus it is important to provide operating system facilities that are appropriate to the application. Micro kernels have addressed this problem by providing a minimal kernel, and ....

....the kernel. These spindles are written in a type safe programming language to ensure that they do not adversely affect kernel operations. Object oriented operating systems allow customization through the use of inheritance, invocation redirection, and meta interfaces. Mukherjee and Schwan et al. [12, 20, 21] control concurrent execution and modification of an object using attributes and object ownership. Choices [6] provides generalized components, called frameworks, which can be replaced with specialized versions using inheritance and dynamic linking. The Spring kernel uses an extensible RPC ....

Bodhisattwa Mukherjee and Karsten Schwan. Experiments with Configurable Locks for Multiprocessors. Report GIT-CC-93/05, College of Computing, Georgia Institute of Technlolgy, Atlanta, GA, January 1993.

....and Tektronix. The problem can be serious when the optimizations embedded in the operating system are particularly bad for the new application. For instance, the usual paging policy provided by most operating systems is simply not appropriate for database applications [30] Mukherjee and Schwan [22] showed that both spinlocks and blocking locks can provide superior performance under different circumstances. Thus it is important to provide operating system facilities that are appropriate to the application. Micro kernels have addressed this problem by providing a minimal kernel, and ....

....the kernel. These spindles are written in a type safe programming language to ensure that they do not adversely affect kernel operations. Object oriented operating systems allow customization through the use of inheritance, invocation redirection, and meta interfaces. Mukherjee and Schwan et al. [14, 22, 23] control concurrent execution and modification of an object using attributes and object ownership. Choices [7] provides generalized components, called frameworks, which can be replaced with specialized versions using inheritance and dynamic linking. The Spring kernel uses an extensible RPC ....

Bodhisattwa Mukherjee and Karsten Schwan. Experiments with Configurable Locks for Multiprocessors. Report GIT-CC-93/05, College of Computing, Georgia Institute of Technlolgy, Atlanta, GA, January 1993.

....implies that locality efficacy may vary as the simulation proceeds and may even vary from domain to domain within a run. While static decisions about the levels at which to exploit molecular locality may provide satisfactory results, dynamic configuration techniques similar to those explored in [MS93] may be useful in obtaining the highest performance for a MD simulations. 4.4 Synchronization Costs Section 3.2.1 presents different approaches to synchronization and particle location exchange, one based on barriers and the other two based on synchronization performed between neighboring ....

Bodhisattwa Mukherjee and Karsten Schwan. Experiments with a configurable lock for multiprocessors. In Proc. of the Twenty Second International Conference on Parallel Processing, volume 2, pages 205--208, August 1993.

....basic measurements demonstrating the performance penalties due to monitoring of lock parameters and reconfigurability are discussed in section 5. Section 6 compares our work with related research and finally, section 7 concludes the paper and presents some future directions. 2 Previous Work In [MS93], we investigate the advantages of dynamic configuration for a specific class of operating system primitives: those used for task or thread synchronization. Using artificial work loads on a NUMA multiprocessor, we demonstrate that: ffl In NUMA machines, spin locks consistently outperform blocking ....

....not currently waiting on a critical section. Furthermore, the experiments show that any locking mechanism offered by an operating system kernel should permit the mixed use of spinning, back off spinning, and blocking as waiting strategies. The design of a reconfigurable lock object is presented in [MS93] offering such multiple waiting strategies. A reconfigurable lock object permits the explicit alteration (statically and or dynamically) of its waiting and scheduling 1 behaviors. Experiments with reconfigurable locks with a workload generator on a 32 node BBN Butterfly multiprocessor (using a ....

[Article contains additional citation context not shown here]

Bodhisattwa Mukherjee and Karsten Schwan. Experiments with a configurable lock for multiprocessors. Technical Report GIT-CC-93/05, College of Computing, Georgia Institute of Technology, 1993.

....setting values to these attributes (like tuning an AM radio tuner) an implementation is tuned for an application. Such adaptation can be either synchronous (where monitoring and adaptation are performed in line and synchronously with the execution of certain operations on the software component [MS93b, MS93c] or asynchronous (where monitoring and adaptation are performed by additional execution threads dedicated to these tasks[EGSM94, Muk94] The adaptive scheduler component presented in this paper uses an asynchronous scheme. In our past research, we have used a similar scheme to implement ....

Bodhisattwa Mukherjee and Karsten Schwan. Experiments with a configurable lock for multiprocessors. In Proc. of the twenty second International Conference on Parallel Processing, volume 2, pages 205--208, August 1993.

....by changing selected scheduler attributes during program execution. 2 Previous Results We have already investigated threads level configuration for a specific construct reconfigurable locks 2 used for the synchronization of multiple processes on the BBN Butterfly NUMA multiprocessor[MS93a, MS93b]. Using artificial workloads[MS93a] and a representative multiprocessor application a parallel branch and bound program on a 32 node BBN Butterfly multiprocessor[Muk91] we have demonstrated that: ffl Off line configuration is important for use of alternative thread schedulers, such as for ....

....during program execution. 2 Previous Results We have already investigated threads level configuration for a specific construct reconfigurable locks 2 used for the synchronization of multiple processes on the BBN Butterfly NUMA multiprocessor[MS93a, MS93b] Using artificial workloads[MS93a] and a representative multiprocessor application a parallel branch and bound program on a 32 node BBN Butterfly multiprocessor[Muk91] we have demonstrated that: ffl Off line configuration is important for use of alternative thread schedulers, such as for realtime vs. non real time ....

Bodhisattwa Mukherjee and Karsten Schwan. Experiments with a configurable lock for multiprocessors. In To appear in the proceedings of International Conference on Parallel Processing, August 1993. Also available as TR# GIT-CC-93/05.

....Locks: A read write lock allows either multiple readers or a single writer to enter a critical section at the same time. The waiting processes may either spin or block depending on whether the lock is implemented as a spinning read write lock or a blocking read write lock. Configurable Locks: In [176], Mukherjee and Schwan study the effects of application and hardware characteristics on multiprocessor locks, and propose a structure for configurable locks. Such locks allow applications to dynamically alter the waiting (spin, block or both) mechanism and the request handling mechanism (how the ....

....creation and invocation, typical application programs see only the objects and classes defined in their code and offered by KTK. Synchronization. In addition to primitive spin and blocking locks, KTK supports configurable lock objects which can be configured to suit application s requirements [176, 177]. Such locks contain a set of implementation dependent attributes which can be dynamically altered to result in a continuous spectrum of lock configurations ranging from busy waiting to blocking (Some useful configurations are: combined locks, advisory locks, priority locks, handoff locks, and ....

Bodhisattwa Mukherjee and Karsten Schwan. Experiments with a configurable lock for multiprocessors. In Proc. of the twenty secondth International Conference on Parallel Processing, volume 2, pages 205--208, August 1993.

....the application programs built with KTK. H.W. Policies memory management thread control sync invocations attributes invocation control scheduling Configurable Threads KTK Applications Figure 1: Structure of the Kernel Toolkit (KTK) Configurable threads is the partially machine dependent component[20] that implements the basic abstractions used by the remainder of KTK: execution threads, virtual memory regions, synchronization primitives, monitoring support for capture of parallel program and KTK state, and a limited number of basic attributes for the configuration of threads level ....

Bodhisattwa Mukherjee and Karsten Schwan. Experiments with a configurable lock for multiprocessors. In Proc. of the International Conference on Parallel Processing, volume 2, pages 205--208, Aug. 1993.

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