Understanding how memory-level parallelism affects the processors performance

As the gap between processor and memory performance increases, performance loss due to long-latency memory accesses become a primary problem. Memory-level parallelism (MLP) improves performance by accessing memory concurrently. An effective system performance analysis model and method is lacked for these researches. Using queuing theory, we establish a system performance analysis model for MLP. Compared with experimental results, the performance trend of the model is accurate. This model can quickly and accurately characterize the features of the system. The average error rate of model is 12%. This model can effectively introduce the initial MLP system design.

[1]  James E. Smith,et al.  A first-order superscalar processor model , 2004, Proceedings. 31st Annual International Symposium on Computer Architecture, 2004..

[2]  Theodore Antonakopoulos,et al.  An Instruction Throughput Model of Superscalar Processors , 2003 .

[3]  Francisco J. Cazorla,et al.  Kilo-instruction processors: overcoming the memory wall , 2005, IEEE Micro.

[4]  Brian Fahs,et al.  Microarchitecture optimizations for exploiting memory-level parallelism , 2004, Proceedings. 31st Annual International Symposium on Computer Architecture, 2004..

[5]  Santosh Nagarakatte,et al.  iCFP: Tolerating All-Level Cache Misses in In-Order Processors , 2010, IEEE Micro.

[6]  Richard E. Matick Comparison of analytic performance models using closed mean-value analysis versus open-queuing theory for estimating cycles per instruction of memory hierarchies , 2003, IBM J. Res. Dev..

[7]  Edward D. Lazowska,et al.  Quantitative System Performance , 1985, Int. CMG Conference.

[8]  T. Puzak,et al.  The optimum pipeline depth for a microprocessor , 2002, Proceedings 29th Annual International Symposium on Computer Architecture.

[9]  Richard E. Kessler,et al.  The Alpha 21264 microprocessor , 1999, IEEE Micro.