TridentFS是一個混合式檔案系統同時管理三種不同類型儲存裝置，包括了NVRAM(NVRAM, Non-Volatile Memory)、固態硬碟(SSD, Solid State Disk) 和磁碟。所有新建立的資料都會被先寫至NVRAM中，當NVRAM空間用盡，TridentFS就會將NVRAM中的資料搬到固態硬碟或磁碟已清出可用空間。若資料搬移的效率不佳，則會使得需要可用空間的資料存取被阻擋而影響整個檔案系統的效能。我們提出了一個平行處理溢出資料的方法，使固態硬碟和磁碟能夠平行處理溢出的資料，除此之外也考慮了資料和儲存裝置的存取特性，來提高清出NVRAM可用空間的效能。根據實驗結果可以得知與目前TridentFS的方法相比最高能改善85%的效能。與由固態硬碟和磁碟所組成的RAID0相比，我們最高改善了494.9%。

TridentFS is a hybrid file system which manages three kinds of storage, i.e. Non-Volatile Memory (NVRAM), Solid State Disk(SSD) and disk. In TridentFS, all new-created data will be stored in NVRAM first. If the NVRAM space is depleted, for NVRAM space reclamation, cold data in NVRAM is evicted to SSD or disk. Speed of NVRAM eviction has a significant impact to whole file system performance because file operations could be blocked for NVRAM space. In this thesis, we propose a parallel eviction approach for TridentFS. For performance improvement, our approach enables well parallel distribution of the evicted data to SSD and disk. In addition, the access pattern of the evicted data and the performance characteristics of the storages are considered. According to the experiment results, when comparing to the eviction approach in existing TridentFS, the outperformance of our approach is up to 85%. When comparing to RAID0 consists of SSD and disk, our approach achieves improvement by up to 494.9%.

[1] H.T. Lin (2010), "Design and implementation of a large file allocation method to improve disk performance for hybrid storage systems", Masters Dissertation, Department of Computer Science and Information Engineering, National Cheng Kung University.
[2] T.Y. Hung (2011), "Techniques of evicting data from NVRAM in the TridentFS hybrid storage based file system", Masters Dissertation, Department of Computer Science and Information Engineering, National Cheng Kung University.
[3] S.W. Tsai (2011), "Management of non-volatile random access memory in a hybrid storage system", Masters Dissertation, Department of Computer Science and Information Engineering, National Cheng Kung University.
[4] L. W. Dowdy & D. V. Foster (1982), "Comparative models of the file assignment problem". ACM Computing Surveys, 14(2), pp.287-313.
[5] T. Xie & Y. Sun (2009), "A file assignment strategy independent of workload characteristic assumptions", ACM Transactions on Storage, 5(3), pp.10.
[6] R. L. Graham (1969), "Bounds on multiprocessing timing anomalies", SIAM Journal on Applied Mathematics, 17(2), pp.416-429.
[7] L. W. Lee, P. Scheuermann & R. Vingralek (2000), "File assignment in parallel I/O systems with minimal variance of service time", IEEE Transactions on Computer, 49(2), pp.127-140.
[8] Y. Zhu, Y. Yu, W. Y. Wang, S. S. Tan & T. C. Low (2010), "A balanced allocation strategy for file assignment in parallel I/O systems", IEEE International Conference on Networking, Architecture and Storage, pp. 257-266.
[9] G. Weikum, P. Zabback & P. Scheuermann (1991), "Dynamic file allocation in disk arrays", ACM International Conference on Management of data, 20(2), pp. 406-415.
[10] J. Tjioe, R. Widjaja, A. Lee & T. Xie (2009), "DORA: A dynamic file assignment strategy with replication", IEEE International Conference on Parallel Processing, pp. 148-155.
[11] R. MCDOUGALL (2004), "Filebench : A prototype model based workload for file systems", work in progress.
http://solarisinternals.com/si/tools/filebench/filebench_nasconf.pdf
[12] J. Katcher (1997), "Postmark: A new file system benchmark", Technical Report TR3022, Network Appliance.
[13] K. M. Evans & G. H. Kuenning (2002), "A study of irregularities in file-size distributions", International Symposium on Performance Evaluation of Computer and Telecommunication Systems.
[14] F. Chen, D. A. Koufaty & X. Zhang (2011), "Hystor: making the best use of solid state drives in high performance storage systems", ACM International Conference on Supercomputing, pp. 22-32.
[15] H. Payer, M. A. Sanvido, Z. Z. Bandic & C. M. Kirsch (2009), "Combo drive: Optimizing cost and performance in a heterogeneous storage device", Workshop on Integrating Solid-state Memory into the Storage Hierarchy, 1(1), pp. 1-8.
[16] D. Lee & K. Koh (2009), "PDC-NH: Popular data concentration on NAND flash and hard disk drive", IEEE/ACM International Conference on Grid Computing, pp. 196-200.
[17] T. Xie & D. Madathil (2008), "SAIL: Self-adaptive file reallocation on hybrid disk arrays", Springer Berlin Heidelbergon High Performance Computing, pp. 529-540.
[18] Y. LIU, et al. (2010), "Raf: A random access first cache management to improve SSD-based disk cache", IEEE International Conference on Networking, Architecture and Storage, pp. 492-500.
[19] S.H. KIM, et al. (2009), "HeteroDrive: Re-shaping the storage access pattern of oltp workload using ssd", Workshop on Software Support for Portable Storage, pp.13-17.
[20] B. Marsh, F. Douglis & P. Krishnan (1994), "Flash memory file caching for mobile computers", IEEE International Conference on System Sciences, pp. 451-460.
[21] F. Chen, S. Jiang & X. Zhang (2006), "SmartSaver: Turning flash drive into a disk energy saver for mobile computers", International Symposium on Low Power Electronics and Design, pp. 412-417.
[22] ATTO Windows Disk Benchmark 2.02.