Skip to main content
SpringerLink
Log in

Identifying the optimal level of parallelism in transactional memory applications

  • Published:
Computing Aims and scope Submit manuscript

Abstract

In this paper we investigate the issue of automatically identifying the “natural” degree of parallelism of an application using software transactional memory (STM), i.e., the workload-specific multiprogramming level that maximizes application’s performance. We discuss the importance of adapting the concurrency level in two different scenarios, a shared-memory and a distributed STM infrastructure. We propose and evaluate two alternative self-tuning methodologies, explicitly tailored for the considered scenarios. In shared-memory STM, we show that lightweight, black-box approaches relying solely on on-line exploration can be extremely effective. For distributed STMs , we introduce a novel hybrid approach that combines model-driven performance forecasting techniques and on-line exploration in order to take the best of the two techniques, namely enhancing robustness despite model’s inaccuracies, and maximizing convergence speed towards optimum solutions.

This is a preview of subscription content, log in via an institution to check access.

Access this article

Log in via an institution

Price excludes VAT (USA)
Tax calculation will be finalised during checkout.

Instant access to the full article PDF.

Institutional subscriptions

Fig. 1
Fig. 2
Fig. 3
Fig. 4
Fig. 5
Fig. 6
Fig. 7
Fig. 8
Fig. 9
Fig. 10

Similar content being viewed by others

Notes

  1. One should note at this point that none of the benchmarks we experimented with (i.e., STAMP applications and various micro-benchmarks) exhibits multiple maxima when observing throughput as a function of the number of threads, up to the hardware limit of our 48-core test machine.

  2. This algorithm, although modifying the number of threads frequently, already achieves good performance results. Our intent was to avoid using fixed thresholds in order to provide a generic solution while keeping the approach simple. A more elaborate algorithm may detect portions of the execution where it is possible to keep the number of threads constant and would achieve even better performance.

  3. A large collection of other experimental results can be found in a companion research report [25].

  4. Note that state transfer is required upon elastic scaling both of replicated DTSMs (which are considered in this paper) and non-replicated ones. Even without replication, in fact, state transfer is needed to send a portion of the data-set to joining nodes in case of scale-up of the system and to preserve the data set upon shrinking the platform’s size, redistributing data from the leaving nodes to the remaining ones.

  5. The original TPC-C benchmark is designed to operate on a relational database, hence we developed a porting running directly on top of a transactional key-value store such as Infinispan (code available here: http://github.com/cloudtm).

References

  1. Abouzour M, Salem K, Bumbulis P (2010) Automatic tuning of the multiprogramming level in Sybase SQL Anywhere. In: Proc. of ICDE workshops

  2. Cao Minh C, Chung J, Kozyrakis C, Olukotun K (2008) STAMP: Stanford transactional applications for multi-processing. In: Proc. of IISWC

  3. Couceiro M, Romano P, Carvalho N, Rodrigues L (2009) D2stm: dependable distributed software transactional memory. In: Proc. of PRDC

  4. Di Sanzo P, Ciciani B, Palmieri R, Quaglia F, Romano P (2012) On the analytical modeling of concurrency control algorithms for software transactional memories: the case of commit-time-locking. Performance Evaluation

  5. Di Sanzo P, Ciciani B, Quaglia F, Romano P (2008) A performance model of multi-version concurrency control. In: Proc. of MASCOTS

  6. Didona D, Felber P, Harmanci D, Romano P, Schenker J (2013) Identifying the optimal level of parallelism in transactional memory applications. In: Proc. of NETYS

  7. Didona D, Romano P, Peluso S, Quaglia F (2012) Transactional auto scaler: elastic scaling of in-memory transactional data grids. In: Proc. of ICAC

  8. Dragojevic A, Guerraoui R (2010) Predicting the scalability of an stm: a pragmatic approach. In: TRANSACT

  9. Elnikety S, Dropsho S, Cecchet E, Zwaenepoel W (2009) Predicting replicated database scalability from standalone database profiling. In: Proc. of EuroSys

  10. Ghanbari S, Soundararajan G, Chen J, Amza C (2007) Adaptive learning of metric correlations for temperature-aware database provisioning. In: Proc. of ICAC

  11. Harmanci D, Gramoli V, Felber P, Fetzer C (2010) Extensible transactional memory testbed. Journal of Parallel and Distributed Computing, Special Issue (Transactional Memory) 70(10):1053–1067

    Google Scholar 

  12. Harris T, Larus JR, Rajwar R (2010) Transactional memory, synthesis. Lectures on computer architecture, 2nd edn. Morgan & Claypool Publisher, San Rafael

    Google Scholar 

  13. Heindl A, Pokam G, Adl-Tabatabai AR (2009) An analytic model of optimistic software transactional memory. In: Proc. of ISPASS

  14. Heiss HU, Wagner R (1991) Adaptive load control in transaction processing systems. In: Proc. of VLDB

  15. Herlihy M, Moss JEB (1993) Transactional memory: architectural support for lock-free data structures. In: Proc. of ISCA

  16. Jiménez-Peris R, Patiño-Martínez M, Alonso G (2002) Non-intrusive, parallel recovery of replicated data. In: Proc. of SRDS

  17. Marchioni F, Surtani M (2012) Infinispan Data Grid Platform. Packt Publishing, Birmingham

    Google Scholar 

  18. Mohammad A, Mikel L, Christos K, Kim J, Chris K, Ian W (2008) Robust adaptation to available parallelism in transactional memory applications. HIPEAC J

  19. Quinlan JR Rulequest Cubist. http://www.rulequest.com/cubist-info.html. Accessed Nov 2013

  20. Quinlan JR (1993) C.45: programs for machine learning. Morgan Kaufmann, Burlington

    Google Scholar 

  21. Raghavan N, Vitenberg R (2011) Balancing the communication load of state transfer in replicated systems. In: Proc. of SRDS

  22. (2011) Red Hat/JBoss: JBoss Infinispan. http://www.jboss.org/infinispan. Accessed Nov 2013

  23. Reimer N, Haenssgen S, Tichy WF (1996) Dynamically adapting the degree of parallelism with reflexive programs. In: Proc. of IRREGULAR

  24. Rughetti D, Di Sanzo P, Ciciani B, Quaglia F (2012) Machine learning-based self-adjusting concurrency in software transactional memory systems. In: Proc. of MASCOTS

  25. Schenker J (2012) Optimistic synchronization and the natural degree of parallelism of concurrent applications, MSc Thesis

  26. Schroeder B, Harchol-Balter M, Iyengar A, Nahum E, Wierman A (2006) How to determine a good multi-programming level for external scheduling. In: Proc. of ICDE

  27. Singh R, Sharma U, Cecchet E, Shenoy P (2010) Autonomic mix-aware provisioning for non-stationary data center workloads. In: Proc. of ICAC. Accessed Nov 2013

  28. TPC Council: TPC-C Benchmark. http://www.tpc.org/tpcc. Accessed Nov 2013

  29. Yoo RM, Lee HHS (2008) Adaptive transaction scheduling for transactional memory systems. In: Proc. of SPAA

  30. Yu PS, Dias DM, Lavenberg SS (1993) On the analytical modeling of database concurrency control. ACM J 40:831–872

    Google Scholar 

  31. Zhang Q, Cherkasova L, Smirni E (2007) A regression-based analytic model for dynamic resource provisioning of multi-tier applications. In: Proc. of ICAC

Download references

Acknowledgments

This work has been partially supported by the projects “Cloud-TM” and “ParaDIME” (co-financed by the European Commission through the contracts no. 257784 and 318693), project specSTM (PTDC/EIA-EIA/122785/2010), the COST Action Euro-TM (IC1001) and by FCT (INESC-ID multiannual funding) through the PEst-OE/EEI/LA0021/2013 Program Funds.

Author information

Authors and Affiliations

  1. Instituto Superior Técnico/INESC-ID, Lisbon, Portugal

    Diego Didona & Paolo Romano

  2. University of Neuchâtel, Neuchâtel, Switzerland

    Pascal Felber, Derin Harmanci & Jörg Schenker

Authors
  1. Diego Didona
    View author publications

    You can also search for this author in PubMed Google Scholar

  2. Pascal Felber
    View author publications

    You can also search for this author in PubMed Google Scholar

  3. Derin Harmanci
    View author publications

    You can also search for this author in PubMed Google Scholar

  4. Paolo Romano
    View author publications

    You can also search for this author in PubMed Google Scholar

  5. Jörg Schenker
    View author publications

    You can also search for this author in PubMed Google Scholar

Corresponding author

Correspondence to Diego Didona.

Additional information

A shorter version of this article [6] appeared in Proc. of International Conference on Networked Systems, 2013.

Rights and permissions

Reprints and permissions

About this article

Cite this article

Didona, D., Felber, P., Harmanci, D. et al. Identifying the optimal level of parallelism in transactional memory applications. Computing 97, 939–959 (2015). https://doi.org/10.1007/s00607-013-0376-3

Download citation

  • Received:

  • Accepted:

  • Published:

  • Issue Date:

  • DOI: https://doi.org/10.1007/s00607-013-0376-3

Keywords

Mathematics Subject Classification

Search

Navigation