Overview

Performance modeling has many applications in understanding and estimating system performance. We focus on analyzing the effects of sharing low level system resources, such as the processor cores or the memory architecture.

  • We will present a paper about sharing processor cores at CCGRID 2015 titled "Analyzing the Impact of CPU Pinning and Partial CPU Loads on Performance and Energy Efficiency". The paper is among four Best Paper Nominees.
  • We have presented a paper about modeling generational garbage collectors at ICPE 2014 titled "On the Limits of Modeling Generational Garbage Collector Performance". The paper won the Best Research Paper Award.
  • We have presented a paper about modeling memory caches at ICPE 2012 titled "On The Accuracy of Cache Sharing Models". The paper won the Best Student Paper Award.

Resource Modeling

Performance modeling must often deal with systems whose performance is influenced by sharing hardware resources such as processor cores or memory caches, or software resources such as virtual machine heap or messaging communication middleware. When shared, these resources can exhibit significant performance related effects:

  • We describe how sharing processor cores in different configurations impacts throughput and efficiency in [1].
  • The work in [2] describes how to use multiple artificial workloads to investigate the cache parameters of a modern processor. A model for predicting cache behavior under sharing is presented in [3], together with a study on modeling accuracy.
  • The accuracy limits when modeling garbage collection frequency using memory allocation traces are explored in [4].
  • A model of the JMS middleware architecture using the QPN formalism is introduced in [5].

Tools

Tool Link Description
RIB Microbenchmark Source Repository A microbenchmark framework for investigating the performance effects of executing multiple interleaving workloads. Support for performance counter access and plotting through R.
RIP Microbenchmark Source Repository A microbenchmark framework for investigating the performance effects of executing multiple concurrent workloads. Support for performance counter access and plotting through R.

Model Validation

Results produced by performance models can be compared to actual measurements to validate the modeling accuracy. When the model must be constructed by hand, such validation can become both expensive and unconvincing. We have introduced a framework for generation of random validation scenarios [6] to address this problem.

Tools

Tool Link Description
Random Program Generator Source Archive QPN Support for RPG A tool for validating performance models by generating random validation applications. C++ and Java applications supported.

Contact

Lubomír Bulej lubomir.bulej<at-sign>d3s.mff.cuni.cz
Peter Libič peter.libic<at-sign>d3s.mff.cuni.cz
Petr Tůma petr.tuma<at-sign>d3s.mff.cuni.cz

References

  • [1]Podzimek A., Bulej L., Chen L. Y., Binder W., Tůma P., Analyzing the Impact of CPU Pinning and Partial CPU Loads on Performance and Energy Efficiency, In proceedings of the 15th IEEE/ACM International Symposium on Cluster, Cloud, and Grid Computing (CCGRID), Shenzhen, Guangdong, China, Best Research Paper Runner Up, IEEE, ISBN: 978-1-4799-8006-2, pp. 1-10, May 2015. PDF, Link, Resources
  • [2]Babka V., Tůma P., Investigating Cache Parameters of x86 Family Processors, In Proceedings of SPEC Benchmark Workshop 2009, Austin, TX, USA, Awarded with Kaivalya Dixit Award. Springer-Verlag, LNCS 5419, ISBN 978-3-540-93798-2, ISSN 0302-9743, pp. 77-96, January 2009. PDF, Link, Link
  • [3]Babka V., Libič P., Martinec T., Tůma P., On The Accuracy of Cache Sharing Models, In proceedings of the 3rd ACM/SPEC International Conference on Performance Engineering (ICPE), Boston, USA, Best Student Paper Award. ACM, ISBN 978-1-4503-1202-8, pp. 21-32, April 2012. PDF, Link
  • [4]Libič P., Bulej L., Horký V., Tůma P., On the Limits of Modeling Generational Garbage Collector Performance, In proceedings of the 5th ACM/SPEC International Conference on Performance Engineering (ICPE), Dublin, Ireland, Best Research Paper Award, ACM, ISBN: 978-1-4503-2733-6, pp. 15-26, March 2014. PDF, Link
  • [5]Martinec T., Marek L., Steinhauser A., Tůma P., Noorshams Q., Rentschler A., Reussner R., Constructing Performance Model of JMS Middleware Platform, In proceedings of the 5th ACM/SPEC International Conference on Performance Engineering (ICPE), Dublin, Ireland, ACM, ISBN: 978-1-4503-2733-6, pp. 123-134, March 2014. PDF, Link
  • [6]Babka V., Tůma P., Bulej L., Validating Model-Driven Performance Predictions On Random Software Systems, In proceedings of the 6th International Conference on the Quality of Software Architectures (QoSA), Prague, Czech Republic, Springer, LNCS 6093, ISBN: 978-3-642-13820-1, ISSN: 0302-9743, pp. 3-19, June 2010. PDF, Link
Logo of Faculty of Mathematics and Physics
  • Phone: +420 951 554 267, +420 951 554 236
  • Email: info<at-sign>d3s.mff.cuni.cz
  •  
  • How to find us?
Modified on 2016-02-11