Betty Salzberg

Comparison of access methods for time-evolving data

This paper compares different indexing techniques proposed for supporting efficient access to temporal data. The comparison is based on a collection of important performance criteria, including the space consumed, update processing, and query time for representative queries. The comparison is based on worst-case analysis, hence no assumptions on data distribution or query frequencies are made. When a number of methods have the same asymptotic worst-case behavior, features in the methods that affect average case behavior are discussed. Additional criteria examined are the pagination of an index, the ability to cluster related data together, and the ability to efficiently separate old from current data (so that larger archival storage media such as write-once optical disks can be used). The purpose of the paper is to identify the difficult problems in accessing temporal data and describe how the different methods aim to solve them. A general lower bound for answering basic temporal queries is also introduced.

Access methods for multiversion data

We present an access method designed to provide a single integrated index structure for a versioned timestamped database with a non-deletion policy. Historical data (superceded versions) is stored separately from current data. Our access method is called the <italic>Time-Split B-tree</italic>. It is an index structure based on Malcolm Eastons Write Once B-tree. The Write Once B-tree was developed for data stored entirely on a Write-Once Read-Many or <italic>WORM</italic> optical disk. The Time-Split B-tree differs from the Write Once B-tree in the following ways:<list><item>Current data <italic>must</italic> be stored on an <italic>erasable</italic> random-access device. </item><item>Historical data <italic>may</italic> be stored on <italic>any</italic> random-access device, including WORMs, erasable optical disks, and magnetic disks. The point is to use a faster and more expensive device for the current data and a slower cheaper device for the historical data. </item><item>The splitting policies have been changed to reduce redundancy in the structure—the option of pure key splits as in B<supscrpt>+</supscrpt>-trees and a choice of split times for time-based splits enable this performance enhancement. </item><item>When data is migrated from the current to the historical database, it is consolidated and appended to the end of the historical database, allowing for high space utilization in WORM disk sectors. </item></list>

The hB-tree: a multiattribute indexing method with good guaranteed performance

A new multiattribute index structure called the hB-tree is introduced. It is derived from the K-D-B-tree of Robinson [15] but has additional desirable properties. The hB-tree internode search and growth processes are precisely analogous to the corresponding processes in B-trees [1]. The intranode processes are unique. A k-d tree is used as the structure within nodes for very efficient searching. Node splitting requires that this k-d tree be split. This produces nodes which no longer represent brick-like regions in k-space, but that can be characterized as holey bricks, bricks in which subregions have been extracted. We present results that guarantee hB-tree users decent storage utilization, reasonable size index terms, and good search and insert performance. These results guarantee that the hB-tree copes well with arbitrary distributions of keys.

The performance of a multiversion access method

The Time-Split B-tree is an integrated index structure for a versioned timestamped database. It gradually migrates data from a current database to an historical database, records migrating when nodes split. Records valid at the split time are placed in both an historical node and a current node. This implies some redundancy. Using both analysis and simulation, we characterise the amount of redundancy, the space utilization, and the record addition (insert or update) performance for a spectrum of different rates of insertion versus update. Three splitting policies are studied which alter the conditions under which either time splits or key space splits are performed.

Online event-driven subsequence matching over financial data streams

Subsequence similarity matching in time series databases is an important research area for many applications. This paper presents a new approximate approach for automatic online subsequence similarity matching over massive data streams. With a simultaneous on-line segmentation and pruning algorithm over the incoming stream, the resulting piecewise linear representation of the data stream features high sensitivity and accuracy. The similarity definition is based on a permutation followed by a metric distance function, which provides the similarity search with flexibility, sensitivity and scalability. Also, the metric-based indexing methods can be applied for speed-up. To reduce the system burden, the event-driven similarity search is performed only when there is a potential event. The query sequence is the most recent subsequence of piecewise data representation of the incoming stream which is automatically generated by the system. The retrieved results can be analyzed in different ways according to the requirements of specific applications. This paper discusses an application for future data movement prediction based on statistical information. Experiments on real stock data are performed. The correctness of trend predictions is used to evaluate the performance of subsequence similarity matching.

Access methods

Although main memories are becoming larger, many large databases still cannot fit entirely in main memory and are disk-resident. The fastest magnetic disks today have an average access time of ten milliseconds whereas CPU operations are measured in nanoseconds, making the access of one disk page at least a million times slower than adding two integers in the CPU. Thus although access methods are not a new topic, they are becoming increasingly important. In addition, new application areas are requiring more complex disk access methods. For example, some of the data being stored in large databases is multidimensional. These dimensions may be spatial or may contain a temporal component. Data mining applications include dimensions such as sales date, type of item sold, color, sales area, and so forth. Queries may involve any of the dimensions, so an access method based on only one dimension is not adequate.

Access method concurrency with recovery

Providing high concurrency in B+-trees has been studied extensively. But few efforts have been documented for combining concurrency methods with a recovery scheme that preserves well-formed trees across system crashes. We describe an approach for this that works for a class of index trees that is a generalization of the Blink-tree. A major feature of our method is that it works with a range of different recovery methods. It achieves this by decomposing structure changes in an index tree into a sequence of atomic actions, each one leaving the tree well-formed and each working on a separate level of the tree. All atomic actions on levels of the tree above the leaf level are independent of database transactions, and so are of short duration.

A finite state model for respiratory motion analysis in image guided radiation therapy

Effective image guided radiation treatment of a moving tumour requires adequate information on respiratory motion characteristics. For margin expansion, beam tracking and respiratory gating, the tumour motion must be quantified for pretreatment planning and monitored on-line. We propose a finite state model for respiratory motion analysis that captures our natural understanding of breathing stages. In this model, a regular breathing cycle is represented by three line segments, exhale, end-of-exhale and inhale, while abnormal breathing is represented by an irregular breathing state. In addition, we describe an on-line implementation of this model in one dimension. We found this model can accurately characterize a wide variety of patient breathing patterns. This model was used to describe the respiratory motion for 23 patients with peak-to-peak motion greater than 7 mm. The average root mean square error over all patients was less than 1 mm and no patient has an error worse than 1.5 mm. Our model provides a convenient tool to quantify respiratory motion characteristics, such as patterns of frequency changes and amplitude changes, and can be applied to internal or external motion, including internal tumour position, abdominal surface, diaphragm, spirometry and other surrogates.

Concurrency and recovery for index trees

Abstract. Although many suggestions have been made for concurrency in B

Subsequence matching on structured time series data

^+