Presentation is loading. Please wait.

Presentation is loading. Please wait.

STATIC CACHE PARTITIONING ROBUSTNESS ANALYSIS FOR EMBEDDED ON-CHIP MULTI- PROCESSORS Anca M. Molnos, Marc J.M. Heijligers, Jos T.J. van Eijndhoven NXP.

Similar presentations


Presentation on theme: "STATIC CACHE PARTITIONING ROBUSTNESS ANALYSIS FOR EMBEDDED ON-CHIP MULTI- PROCESSORS Anca M. Molnos, Marc J.M. Heijligers, Jos T.J. van Eijndhoven NXP."— Presentation transcript:

1 STATIC CACHE PARTITIONING ROBUSTNESS ANALYSIS FOR EMBEDDED ON-CHIP MULTI- PROCESSORS Anca M. Molnos, Marc J.M. Heijligers, Jos T.J. van Eijndhoven NXP Semiconductors, HTC Sorin D. Cotofana Technical University of Delft

2 Outline Introduction Cache Partitioning Robustness Evaluation Methods Experiment Results Conclusion

3 Introduction Media applications are characterized by high requirements with respect to –computation –memory bandwidth Hierarchical caches are common ways to alleviates data availability problem. Robustness is one of the main required properties in media application domain.

4 Associativity Based Partitioning Each task gets a number of ways (columns) from every set of the cache. One task can flush out only its own cache ways in case of cache miss. Reducing the number of misses, speeding up the application. Number of allocable resources is restricted to the number of ways in a set.

5 Set Based Partitioning Each task gets a different amount of sets from the cache. Equivalent to address space partitioning. The number of resources allocable is large. Implementation is more ‘intrusive” into cache organization.

6 Internal Robustness Internal variations in tasks performance are due to switching of tasks. L2 sensitivity function is used for internal robustness analysis. Task sensitivity function: Application sensitivity function:

7 External Robustness (1/2) External variations are due to different input data set.

8 External Robustness (2/2) Stability matrix is used for external robustness analysis Application’s stability for particular input: Overall application stability:

9 Experiment Environments Multiprocessor platform: –4 Trimedia processors –Private instructions and data caches –Unified 4 ways associativity L2 cache Media workload: –MediaBench

10 Experiments Results Internal Robustness

11 Experiments Results Internal Robustness

12 Experiments Results External Robustness

13 Conclusion Static cache partitioning is quite robust: –Average sensitivity of 4% –Average stability 92% Future work – dynamic cache partitioning.


Download ppt "STATIC CACHE PARTITIONING ROBUSTNESS ANALYSIS FOR EMBEDDED ON-CHIP MULTI- PROCESSORS Anca M. Molnos, Marc J.M. Heijligers, Jos T.J. van Eijndhoven NXP."

Similar presentations


Ads by Google