Du Y.,Tsinghua University |
Du Y.,National Engineering Laboratory for Disaster Backup and Recovery |
Yu H.,Tsinghua University |
Zheng W.,Tsinghua University
Jisuanji Yanjiu yu Fazhan/Computer Research and Development | Year: 2012
Phase-change memory (PCM) is an emerging memory technique. PCM offers many advantages over traditional DRAM, and thus has the potential to be the next generation main memory in computer systems. One of the hurdles for its use is the limited number of writes to storage cells. Furthermore, the non-uniformity of memory accesses in typical workloads makes this situation worse. In such case, wear-leveling is often employed to map the logical address to the physical address and remap them to distribute the writes among all cells to prevent some cells from being worn out sooner than the others. This paper proposes an algebraic-mapping-based matrix wear-leveling method. Our method views the storage cells as a matrix and then levels the writes in rows and columns in a two-dimensional scheme. Every logical address is simultaneously remapped in a column and a row, and thus can be efficiently remapped to any physical address as soon as possible. The method can extend the lifetime of PCM both under normal applications and in malicious attacks, and meanwhile incur very little write overhead.
Liu W.,Communication University of China |
Luo S.-S.,Beijing University of Posts and Telecommunications |
Luo S.-S.,National Engineering Laboratory for Disaster Backup and Recovery |
Wang Y.-B.,Communication University of China
Tien Tzu Hsueh Pao/Acta Electronica Sinica | Year: 2010
Secure two-party vector dominance statistic problem is a problem generalized from the millionaires' problem, which can be used by two party to get the number of ai > bi without leaking further information. A secure two-party vector dominance statistic protocol in semi-honest model is presented based on the additive homomorphic encryption. The correctness, security and complexity of the protocol are analyzed. A secure components sum of two vectors ranking protocol is proposed based on the dominance statistic protocol and the ranking protocol is also applied in the secure minimum spanning trees algorithm.
Liu L.,CAS Institute of Computing Technology |
Xu L.,National Engineering Laboratory for Disaster Backup and Recovery |
Applied Mathematics and Information Sciences | Year: 2011
We present a decentralized resource allocation approach, DRG (Decentralized Response-time Guarantee), that can provide response time guarantees for multiple concurrent workloads sharing a back-end storage system in a distributed manner without assuming any support from the storage itself. This new approach uses several run-time statistics of both workloads and system as indicators of busty and load condition at the backend storage and accepts a control equation periodically to adjust the number of I/O requests which could be issued per workloads to meet the performance goals. Using a real I/O trace, we demonstrate that our approach can simultaneously meet the response-time requirements imposed by an SLO without requiring extensive knowledge of the underlying storage system. © 2011 NSP.
Maohua S.,Beijing University of Posts and Telecommunications |
Shoushan L.,National Engineering Laboratory for Disaster Backup and Recovery |
Lei P.,Beijing University of Posts and Telecommunications |
Zhe J.,Beijing University of Posts and Telecommunications |
Yang X.,Beijing Safe Code Technology Co.
International Journal of Advancements in Computing Technology | Year: 2012
With the rapid growth of the internet and the progress of the society, people need to share more and more information with others. How to ensure the authorized persons to get the information is a big problem now. Secret sharing is a widely used tool to solve this problem. In this paper, the related work and the framework of secret sharing is demonstrated. We also develop the relationship between secret sharing and secure multi-party computation, and introduce how to build the privacy-preserving secret sharing scheme. At the same time, future works of secret sharing are proposed in the end.
Yu H.,Tsinghua University |
Yu H.,National Engineering Laboratory for Disaster Backup and Recovery |
Zhang F.,Tsinghua University |
Wu Y.,Tsinghua University
Future Generation Computer Systems | Year: 2014
Up to now, more and more people use Internet storage services as a new way of sharing. File sharing by a distributed storage system is quite different from a specific sharing application like BitTorrent. And as large file sharing becomes popular, the data transmission rate takes the place of the response delay to be the major factor influencing user experience. We present the design and implementation of a distributed storage system named as Granary in this paper, which provides reliable data storage and sharing service to cyber users. Granary uses a specific DHT(Distributed Hash Table) layer to store file meta-data and employs a raw data storage scheme to scatter large data. We introduce its adaptive DHT recovery algorithm in this paper which assures the availability and consistency of meta-data with small bandwidth consumption and improved throughout. The replication strategies which are used to accelerate file sharing with low bandwidth consumptions are further discussed. Experimental results show that these methods offer a reliable and efficient data storage and sharing speed with network bandwidth costs less than conventional policies. © 2013 Elsevier B.V. All rights reserved.