In this paper, we present a design and implementation of FD-Buffer, a cost-based adaptive buffer manager for flash memory devices. Due to flash memory’s unique hardware features, it has an inherent read-write asymmetry: writes involve expensive erase operations, which usually makes them much slower than reads. To address this read-write asymmetry, we revisit buffer management and consider the average I/O cost per page access as the main cost metric, as opposed to the traditional miss rate. While there have been a number of buffer management algorithms that take the read-write asymmetry into consideration, most algorithms fail to effectively adapt to the runtime workload or different degrees of asymmetry. In this paper, we develop a new replacement algorithm in which we separate clean and dirty pages into two pools. The size ratio of the two pools is automatically adapted based on the read-write asymmetry and the runtime workload. We evaluate the FD-Buffer with trace-driven experiments on real flash memory devices. Our trace-driven evaluation results show that our algorithm achieves 4.0-33.4 percent improvement of I/O performance on flash memory, compared to state-of-the-art flash-aware replacement policies.

Flash memory-based disks (or simply flash disks) have been widely used in today's computer systems. With their continuously increasing capacity and dropping price, it is envisioned that some database systems will operate on flash disks in the near future. However, the I/O characteristics of flash disks are different from those of magnetic hard disks. Motivated by this, we study the core of query processing in row-based database systems-join processing-on flash storage media. More specifically, we propose a new framework, called DigestJoin, to optimize nonindexed join processing by reducing the intermediate result size and exploiting fast random reads of flash disks. DigestJoin consists of two phases: 1) projecting the join attributes followed by a join on the projected attributes, and 2) fetching the full tuples that satisfy the join to produce the final join results. While the problem of tuple/page fetching with the minimum I/O cost (in the second phase) is intractable, we propose three heuristic page-fetching strategies for flash disks. We have implemented DigestJoin and conducted extensive experiments on a real flash disk. Our evaluation results based on TPC-H data sets show that DigestJoin clearly outperforms the traditional sort-merge join and hash join under a wide range of system configurations.

Owing to recent advances in semiconductor technologies, flash disks have been a competitive alternative to traditional magnetic disks as external storage media. In this paper, we study how transaction recovery can be efficiently supported in database management systems (dbmss) running on slc flash disks. Inspired by the classical shadow-paging approach, we propose a new commit scheme, called flagcommit, to exploit the unique characteristics of flash disks such as fast random read access, out-place updating, and partial page programming. To minimize the need of writing log records, we embed the transaction status into flash pages through a chain of commit flags. Based on flagcommit, we develop two recovery protocols, namely commit-based flag commit (cfc) and abort-based flag commit (afc), to meet different performance needs. They are flexible to support no-force buffer management and fine-grained concurrency control. Our performance evaluation based on the tpc-c benchmark shows that both cfc and afc outperform the state-of-the-art recovery protocols.

We design and implement FD-Buffer, a buffer manager for database systems running on flash-based disks. Unlike magnetic disks, flash media has an inherent read-write asymmetry: writes involve expensive erase operations and as a result are usually much slower than reads. Therefore, we address this asymmetry in FD-Buffer. Specifically, we use the average I/O cost per page access as opposed to the traditional miss rate as the performance metric for a buffer. We develop a new replacement policy in which we separate clean and dirty pages into two pools. The size ratio of the two pools is automatically adapted to the read-write asymmetry and the runtime workload. We evaluate FD-Buffer with trace-driven experiments on real flash disks. Our evaluation results show that our algorithm achieves up to 33% improvement on the overall performance on commodity flash disks, in comparison with the state-of-the-art flash-aware replacement policy.

This article examines a new problem of k -anonymity with respect to a reference dataset in privacy-aware location data publishing: given a user dataset and a sensitive event dataset, we want to generalize the user dataset such that by joining it with the event dataset through location, each event is covered by at least k users. Existing k -anonymity algorithms generalize every k user locations to the same vague value, regardless of the events. Therefore, they tend to overprotect against the privacy compromise and make the published data less useful. In this article, we propose a new generalization paradigm called local enlargement , as opposed to conventional hierarchy- or partition-based generalization. Local enlargement guarantees that user locations are enlarged just enough to cover all events k times, and thus maximize the usefulness of the published data. We develop an O ( H n )-approximate algorithm under the local enlargement paradigm, where n is the maximum number of events a user could possibly cover and H n is the Harmonic number of n . With strong pruning techniques and mathematical analysis, we show that it runs efficiently and that the generalized user locations are up to several orders of magnitude smaller than those by the existing algorithms. In addition, it is robust enough to protect against various privacy attacks.

Flash disks have been an emerging secondary storage media. In particular, there have been portable devices, multimedia players and laptop computers that are configured with no magnetic disks but flash disks.It is envisioned that some RDBMSs will operate on flash disks in the near future. However, the I/O characteristics of flash disks are different from those of magnetic disks. Thus, in this paper,we study the core of query processing in RDBMSs - join processing - on flash disks. Specifically, we propose a new join method, called DigestJoin, to exploit fast random reads of flash disks. DigestJoin consists of two phases: (1) projecting the join attributes followed by a join on the projected attributes; and (2)fetching the full tuples that satisfy the join to produce the final join results. While the problem of tuple/page fetching with minimum I/O cost (in the second phase) is intractable, we propose three heuristic fetching strategies. We have implemented DigestJoin on a real flash disk for performance evaluation.Experiments on TPC-H datasets show that DigestJoin clearly outperforms the traditional sort-merge join under various system configurations.

With the rapid increasing capacity of flash chips, flash-aware indexing techniques are highly desirable for flash devices. The unique features of flash memory, such as the erase-before-write constraint and the asymmetric read/write cost, severely deteriorate the performance of the traditional B+-tree algorithm. In this paper, we propose a new indexing method, called lazy-update B+-tree, to overcome the limitations of flash memory. The basic idea is to defer the time of committing update requests to the B+-tree by buffering them in a segment of main memory. They are later committed in groups so that each write operation can be amortized by a bunch of update requests. We identify a victim selection problem for the lazy-update B+-tree and develop two heuristic-based commit policies to address the problem. Simulation results show that the proposed lazy-update method, along with a well-designed commit policy, greatly improves the update performance of the traditional B+-tree while preserving the query efficiency.