Predicting memory-access cost based on data-access patterns

Byna, Suren; Sun, Xian-He; Gropp, William; Thakur, Rajeev

doi:10.1109/clustr.2004.1392630

Cited by 18 publications

(15 citation statements)

References 20 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…However, predicting the optimization parameters with low overhead and automatically applying these optimization strategies is a challenging research issue. We developed models for predicting memory-access cost [2] that can help in dynamically applying optimizations. By combining optimization methods with a memory access model, we have introduced in this paper an approach to optimize memory performance automatically.…”

Section: Discussionmentioning

confidence: 99%

“…Our method first retrieves the data access pattern of a derived datatype from user's definition and verifies whether performance improvement is possible with optimizations for a derived datatype before applying them. If improvement is possible, our optimization method uses an analytical model [2] to predict memory access cost and to find optimization parameters with the lowest access cost. These parameters are passed to templates to pack/unpack noncontiguous data.…”

Section: Optimization Mechanismmentioning

confidence: 99%

“…In our approach, we first select these optimization parameters based on heuristics. To determine if these parameters are optimal, we developed a simple, fast, and accurate memory-access-cost prediction model [2]. This model verifies whether the memory access cost is reduced with the selected parameters.…”

Section: Optimization Mechanismmentioning

confidence: 99%

See 2 more Smart Citations

Automatic Memory Optimizations for Improving MPI Derived Datatype Performance

Byna

Sun

Thakur

et al. 2006

Recent Advances in Parallel Virtual Machine and Message Passing Interface

View full text Add to dashboard Cite

Abstract. MPI derived datatypes allow users to describe noncontiguous memory layout and communicate noncontiguous data with a single communication function. This powerful feature enables an MPI implementation to optimize the transfer of noncontiguous data. In practice, however, many implementations of MPI derived datatypes perform poorly, which makes application developers avoid using this feature. In this paper, we present a technique to automatically select templates that are optimized for memory performance based on the access pattern of derived datatypes. We implement this mechanism in the MPICH2 source code. The performance of our implementation is compared to well-written manual packing/unpacking routines and original MPICH2 implementation. We show that performance for various derived datatypes is significantly improved and comparable to that of optimized manual routines.

show abstract

Section: Discussionmentioning

confidence: 99%

Section: Optimization Mechanismmentioning

confidence: 99%

See 1 more Smart Citation

Automatic Memory Optimizations for Improving MPI Derived Datatype Performance

Byna

Sun

Thakur

et al. 2006

Recent Advances in Parallel Virtual Machine and Message Passing Interface

View full text Add to dashboard Cite

show abstract

“…Loop transformations (e.g., loop unrolling, loop tiling) and array restructuring can improve data access and performance by increasing the locality associated with these memory accesses [4]. Previous work has shown that we can reorder memory accesses to significantly improve application performance [8,11].…”

Section: Related Workmentioning

confidence: 99%

External memory page remapping for embedded multimedia systems

NingKe

KaeliDavid

2007

SIGPLAN Not.

View full text Add to dashboard Cite

As memory speeds and bus capacitances continue to rise, external memory bus power will make up an increasing portion of the total system power budget for system-on-a-chip embedded systems. Both hardware and software approaches can be explored to balance the power/performance tradeoff associated with the external memory.In this paper we present a hardware-based, programmable external memory page remapping mechanism which can significantly improve performance and decrease the power budget due to external memory bus accesses. Our approach was developed by studying common data access patterns present in embedded multimedia applications. In the paper, we evaluate a mechanism that can perform page remapping of external memory. We also develop an efficient algorithm to map application data and instruction memory into external memory pages. We employ graph-coloring techniques to guide the page mapping procedure. The objective is to avoid page misses by remapping conflicting pages to different memory banks (i.e., by assigning them different colors). Our algorithm can significantly reduce the memory page miss rate by 70-80% on average. For a 4-bank SDRAM memory system, we reduced external memory access time by 12.6% . The proposed algorithm can reduce power consumption in majority of the benchmarks, averaged by 13.2% of power reduction. Combining the effects of both power and delay, our algorithm can benefit significantly to the total energy cost.

show abstract

“…They also show that tree traversal applications can generate many random memory accesses resulting in degraded performance, as demonstrated by heap sort. In parallel, Byna et al [2] proposed a memory cost model for looping operations.…”

Section: Related Workmentioning

confidence: 99%