TridentFS是基於混合性儲存裝置所開發出的檔案系統，整個系統包含三個現今熱門的儲存裝置：非揮發性隨機存取記憶體（NVRAM）、硬碟以及固態硬碟（SSD）。TridentFS依據資料的存取特性，將資料放置到適合的裝置上。具體來說，高速NVRAM儲存具有頻繁存取特性的後資料（metadata）和小型檔案。大容量硬碟儲存具有連續存取特性的大型檔案。當NVRAM容量不敷使用時，SSD被拿來當作NVRAM的延伸空間使用。藉由這種資料區分的方法，整體效能便可以獲得顯著的提升。
為了解決NVRAM容量不足的限制，我們建立了資料溢出的機制並且提出更精確的資料溢出處理方法來減少對SSD的存取量藉以增進系統的整體效能。我們將索引節點（inode）與資料（data）分開以減少假檔案（fake file）所造成的SSD額外存取量。在資料溢出挑選部分，相較於一開始的資料溢出處理方法，我們考慮更多因素：存取的時間、頻率和資料大小，這些都是為了更精確的踢出不必要的資料，並且讓更多的存取留在NVRAM上。
我們另外也提出新的SSD檔案配置設計，來補強在SSD上使用現有檔案系統之目錄搜尋能力上的不足。藉由新的配置設計，我們可以減輕過多的檔案集中存放在單一目錄上。
藉由更精確的資料溢出處理方法，TridentFS在效能上獲得了高達34.12%的改善並且減少了高達26.71%的SSD存取量。藉由新的SSD檔案配置設計，TridentFS的整體效能獲得高達74.53%的改善。

The TridentFS is a hybrid storage file system which combines three popular devices: NVRAM, hard disk drive and SSD. Each device stores different type of data according to the data access patterns. Specifically, the high speed NVRAM stores both frequently-accessed metadata and small files and the hard disk is taken as a high volume storage device for sequentially-accessed large files. The SSD is employed as an extension of NVRAM in case of NVRAM exhaustion. With this data separation, the overall performance can be significantly improved.
To handle the size limitation of NVRAM, we setup an eviction mechanism and a more fine-grained and precise eviction policy is proposed to improve the performance by reducing the SSD I/O. We separate inode from data for reducing the extra I/O caused by fake files. On eviction victim selection, we consider more factors: recency, frequency and size against original eviction policy of TridentFS, to more precisely choose the eviction victims and let most I/O occur in NVRAM.
In addition, the SSD file layout design is proposed to increase the performance of directory entry searching when using existing file system on SSD. By using the design, TridentFS can release excessive file pressure on a single directory of SSD to complement the shortage of directory entry searching on underlying file system.
By using precise eviction policy, TridentFS can demonstrate up to 34.12% improvement in read/write throughput and up to 26.71% reduction of SSD I/O. By using new file layout design on SSD, TridentFS can perform up to 74.53% in time reduction against original TridentFS.

[1] "Everspin Technologies: MR2A16A Series 256K-bit x 16 Asynchronous Magnetoresistive RAM," July 2009.
http://www.futureelectronics.com/
[2] R. Bez, E. Camerlenghi, A. Modelli, and A. Visconti, "Introduction to flash memory," Proceedings of the IEEE, vol. 91, pp. 489-502, 2003.
[3] H. Boeve, C. Bruynseraede, J. Das, K. Dessein, G. Borghs, J. De Boeck, R. C. Sousa, L. V. Melo, and P. P. Freitas, "Technology assessment for the implementation of magnetoresistive elements with semiconductor components in magnetic random access memory (MRAM) architectures," Magnetics, IEEE Transactions on, vol. 35, pp. 2820-2825, 1999.
[4] I. H. Doh, J. Choi, D. Lee, and S. H. Noh, "Exploiting non-volatile RAM to enhance flash file system performance," presented at the Proceedings of the 7th ACM & IEEE international conference on Embedded software, Salzburg, Austria, 2007.
[5] N. K. Edel, D. Tuteja, E. L. Miller, and S. A. Brandt, "MRAMFS: A Compressing File System for Non-Volatile RAM," presented at the Proceedings of the The IEEE Computer Society's 12th Annual International Symposium on Modeling, Analysis, and Simulation of Computer and Telecommunications Systems, 2004.
[6] K. M. Evans and G. H. Kuenning, "A study of irregularities in file-size distributions," in In Proceedings of the 2002 International Symposium on Performance Evaluation of Computer and Telecommunication Systems (SPECTS), 2002.
[7] J. Katcher, "PostMark: A new file system benchmark," Tech. Rep. TR3022. Network Appliance, Inc, 1997.
[8] J. K. Kim, H. G. Lee, S. Choi, and K. I. Bahng, "A PRAM and NAND flash hybrid architecture for high-performance embedded storage subsystems," presented at the Proceedings of the 8th ACM international conference on Embedded software, Atlanta, GA, USA, 2008.
[9] H.-T. Lin, "Design and Implementation of a Large File Allocation Method to Improve Disk Performance for Hybrid Storage Systems," Department of Computer Science and Information Engineering, National Cheng Kung University, 2010.
[10] R. MCDOUGALL, "Filebench: A prototype model based workload for file systems, work in progress " 2004.
http://solarisinternals.com/si/tools/filebench/filebench_nasconf.pdf
[11] E. L. Miller, S. A. Brandt, and D. D. E. Long, "HeRMES: High-Performance Reliable MRAM-Enabled Storage," presented at the Proceedings of the Eighth Workshop on Hot Topics in Operating Systems, 2001.
[12] Y. Park, S.-H. Lim, C. Lee, and K. H. Park, "PFFS: a scalable flash memory file system for the hybrid architecture of phase-change RAM and NAND flash," presented at the Proceedings of the 2008 ACM symposium on Applied computing, Fortaleza, Ceara, Brazil, 2008.
[13] D. Roberts, T. Kgil, and T. Mudge, "Integrating NAND flash devices onto servers," Commun. ACM, vol. 52, pp. 98-103, 2009.
[14] D. Roselli, J. R. Lorch, and T. E. Anderson, "A comparison of file system workloads," presented at the Proceedings of the annual conference on USENIX Annual Technical Conference, San Diego, California, 2000.
[15] A.-I. A. Wang, G. Kuenning, P. Reiher, and G. Popek, "The Conquest file system: Better performance through a disk/persistent-RAM hybrid design," Trans. Storage, vol. 2, pp. 309-348, 2006.