Abstract-Dynamic runtimes can simplify parallel programming by automatically managing concurrency and locality without further burdening the programmer. Nevertheless, implementing such runtime systems for large-scale, shared-memory systems can be challenging. This work optimizes Phoenix, a MapReduce runtime for shared-memory multi-cores and multiprocessors, on a quad-chip, 32-core, 256-thread UltraSPARC T2+ system with NUMA characteristics. We show how a multi-layered approach that comprises optimizations on the algorithm, implementation, and OS interaction leads to significant speedup improvements with 256 threads (average of 2.5× higher speedup, maximum of 19×). We also identify the roadblocks that limit the scalability of parallel runtimes on shared-memory systems, which are inherently tied to the OS scalability on large-scale systems.

The effects of dorsal hippocampal lesions on retention of classical trace conditioned responses were examined using the rabbit nictitating membrane preparation. Animals were trained to criteria and then lesioned either in the cortex or in the hippocampus and the cortex. Hippocampal damage had no effect on the retention of responses but produced significantly longer onset latencies. A control group of hippocampal animals acquired conditioned responses (CRs) at least as quickly as the prelesion subjects, and they also exhibited longer response onset latency. A second experiment evaluated the performance of hippocampal lesioned animals in classical trace conditioning with either a low-intensity periorbital shock or corneal air puff as the unconditioned stimulus (UCS). Hippocampal animals successfully acquired CRs under both conditions but exhibited an alteration of response onset which was dependent on the form of the UCS. Hippocampal animals displayed shorter response onset in the air-puff condition and longer response onset in the shock condition. Cortical animals timed responses consistently regardless of the UCS. These findings strongly suggest that the hippocampus modulates temporal characteristics of learned behavior.

This study reexamined the effects of unilateral damage to cerebellar hemispheral lobule VI on the rabbit's conditioned nictitating membrane (NM) response. Extensive unilateral removal of hemispheral lobule VI in 11 rabbits impaired ipsilateral conditioned responses as reflected by reductions of 52% in mean frequency and 53% in mean amplitude during test trials on the first postoperative session. The decreases in the amplitude and frequency of conditioned responses were highly correlated (r = 0.82). The frequency of conditioned responses recovered to control levels but their amplitudes remained reduced such that the correlation between these two measures of responding was no longer significant by the 12th postoperative conditioning session. The decrease in the amplitude of conditioned responses was not accompanied by changes in onset latency or rise time. There was no significant impairment of conditioned responses in surgical controls and animals with only partial damage to hemispheral lobule VI. It was concluded that hemispheral lobule VI plays an important role in the regulation of motor centers in the brainstem so as to facilitate the initiation and optimum execution of the conditioned NM reflex. This cortical regulation of the conditioned NM response may contain learned elements; however, these cannot be resolved with lesion methods, nor has their existence been proven in this or other lesion studies. Nevertheless, the results of this study do demonstrate that the cerebellar cortex cannot be considered as the single locus necessary for NM conditioning.

No abstract

This study examined the role of the mitogen-activated protein kinase (MAPK) family during acquisition of the rabbit's classically conditioned eye-blink response. Eye-blink conditioning produced a significant, bilateral activation of both extracellular signal-regulated protein kinases (ERKs) and p38 MAPK in the anterior cerebellar vermis. There was also a significant bilateral activation of ERKs in the dorsal hippocampus with no change in p38 MAPK. These changes were seen at 2 min after the last conditioning session, were maintained for at least 180 min, and occurred without any change in the protein expression of either ERKs or p38 MAPK. There were no changes in ERKs or p38 MAPK in frontal cortex, in cerebellar hemispheral lobule VI, or in a section of brainstem containing the inferior olive. Moreover, the stress-related protein kinase Jun N-terminal kinase (JNK), another subfamily of MAPKs, was not altered in any of the brain regions examined. Animals receiving explicitly unpaired presentations of a conditioned stimulus and an unconditioned stimulus did not acquire conditioned responses (CRs) and did not demonstrate any changes in ERKs, p38 MAPK, or JNK. The intraventricular injection of SB203580, a selective p38 MAPK inhibitor, significantly retarded CR acquisition and blocked the learning-related increases in p38 MAPK activity in the anterior vermis. PD98059, a selective MAPK kinase inhibitor, had a smaller and only marginally significant effect on CR acquisition, although it did block the learning-related increases in ERK activity in both the hippocampus and anterior vermis. These results indicate that p38 MAPK is activated during associative learning and may play a role in the transcriptional events that lead to memory consolidation.