Spatial Memory Streaming

Somogyi, Stephen; Wenisch, Thomas F.; Ailamaki, Anastassia; Falsafi, Babak; Moshovos, Andreas

doi:10.1145/1150019.1136508

Cited by 95 publications

(78 citation statements)

References 35 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…10 These experiments model a 16-way, distributedshared-memory multiprocessor loosely based on the HP GS1280 and Compaq Piranha designs. The relative comparison contrasts a system with SMS hardware to a base system.…”

Section: Discussionmentioning

confidence: 99%

“…(For complete details of the experiments, system models, and workload configurations, see our latest work. 3,10 ) Simulation without sampling is simply infeasible, requiring CPU years to provide a high probability of accurate results. In contrast, the SimFlex methodology enables highconfidence estimates of application performance in minutes for SPEC CPU2000 and with only a few CPU days for commercial applications.…”

Section: Discussionmentioning

confidence: 99%

See 1 more Smart Citation

SimFlex: Statistical Sampling of Computer System Simulation

et al. 2006

View full text Add to dashboard Cite

Section: Discussionmentioning

confidence: 99%

Section: Discussionmentioning

confidence: 99%

SimFlex: Statistical Sampling of Computer System Simulation

et al. 2006

View full text Add to dashboard Cite

“…A non-disruptive approach to improving DRAM energy efficiency is to increase the fraction of DRAM accesses that hit in the row buffer, thus eliminating redundant page activations. Advanced prefetchers can enforce row buffer hits by predicting spatial footprints of load-triggered memory reads within a page [4,25,44]; however, these proposals carry a high storage overhead (60KB per core) and ignore store-triggered memory reads and LLC writebacks. Eager writeback mechanisms [27,28,45] can improve row buffer locality for writebacks, but only to a limited degree as they schedule writebacks of only a few adjacent cache blocks at a time.…”

Section: Introductionmentioning

confidence: 99%

“…Compared to state-of-the-art spatial prefetching [44] and eager writeback [45] mechanisms, BuMP reduces memory energy per access by 20% and 13%, while improving server throughput by 3% and 9%, respectively.…”

Section: Introductionmentioning

confidence: 99%

BuMP: Bulk Memory Access Prediction and Streaming

Volos¹,

Picorel²,

Falsafi³

et al. 2014

2014 47th Annual IEEE/ACM International Symposium on Microarchitecture

Self Cite

View full text Add to dashboard Cite

Abstract-With the end of Dennard scaling, server power has emerged as the limiting factor in the quest for more capable datacenters. Without the benefit of supply voltage scaling, it is essential to lower the energy per operation to improve server efficiency. As the industry moves to lean-core server processors, the energy bottleneck is shifting toward main memory as a chief source of server energy consumption in modern datacenters. Maximizing the energy efficiency of today's DRAM chips and interfaces requires amortizing the costly DRAM page activations over multiple row buffer accesses.This work introduces Bulk Memory Access Prediction and Streaming, or BuMP. We make the observation that a significant fraction (59-79%) of all memory accesses fall into DRAM pages with high access density, meaning that the majority of their cache blocks will be accessed within a modest time frame of the first access. Accesses to high-density DRAM pages include not only memory reads in response to load instructions, but also reads stemming from store instructions as well as memory writes upon a dirty LLC eviction. The remaining accesses go to low-density pages and virtually unpredictable reference patterns (e.g., hashed key lookups). BuMP employs a low-cost predictor to identify high-density pages and triggers bulk transfer operations upon the first read or write to the page. In doing so, BuMP enforces high row buffer locality where it is profitable, thereby reducing DRAM energy per access by 23%, and improves server throughput by 11% across a wide range of server applications.

show abstract

“…Spatial memory streaming (SMS) is a runtime data prefetcher that extracts spatially-correlated memory access patterns over large regions of memory and uses them to predict future access patterns [14]. SMS streams the predicted data blocks into the level one cache as fast as available bandwidth and resources allow.…”

Section: Prefetchingmentioning

confidence: 99%