I am going to present the following paper: "Adaptive Overlapped Declustering: A Highly Available Data-Placement Method Balancing Access Load and Space Utilization" by Akitsugu Watanabe and Haruo Yokota, from ICDE'05.

Federated database systems are a useful tool for businesses and researchers around the world. These systems allow data from multiple remote data sources to be logically combined into one unified local data source. Using this system, queries that would traditionally require query fragments to be submitted to multiple sites can be performed by submitting one query to a central site. This central site can make use of data stored at the different remote sources as though the central site were simply an application requesting data.

These so-called global queries must be optimized, but many additional factors combine to make global query optimization complicated. Beyond the problems of local query optimization, additional costs, including the cost of communication, the cost of remote site optimization, etc., must be factored into cost models. Currently, the performance of global queries in iAnywhere Adaptive Server Anywhere is much worse than the performance of local queries.

This talk presents some background material on federated database systems including information regarding benchmarking performed at iAnywhere Solutions, Inc. during a cooperative work term. A discussion of the results of this benchmarking as well as some initial recommendations for improving the performance of global queries will be presented. Some future work is also outlined.

I will present S. Bansal and D. Modha, "CAR: Clock with Adaptive Replacement", Proceedings of the 3nd USENIX Symposium on File and Storage Technologies, March, 2004 (available as http://www.almaden.ibm.com/cs/people/dmodha/clockfast.pdf). It's a follow on to N. Megiddo and D. S. Modha, "ARC: A self-tuning, low overhead replacement cache", Proceedings of the 2nd USENIX Conference on File and Storage Technologies, March 2003.

Most large software systems today use a three-tier architecture, with simple Web clients, the bulk of the business logic in a middle layer, and a back end layer generally using a standard RDBMS. For ECM applications, such as those built by Open Text, the facilities provided by those RDBMSes are problematic in a number of areas. General solutions to some of these problems could be useful to many other applications as well. This talk will review those ECM application needs, the characteristics of the RDBMSes that do not support those needs well, and some possible directions that could be taken to address those issues.

I'll overview the intra-query parallelism capabilities that have recently been added to SQL Anywhere. Most of the focus will be on the execution model. If there's time, I'll also talk about the costing and optimization.

I'll discuss this paper which discusses how much impact search engines have on the popularity (and therefore the ranking) of pages on the Web. From the abstract: "Recent studies show that a majority of Web page accesses are referred by search engines. In this paper we study the widespread use of Web search engines and its impact on the ecology of the Web. In particular, we study how much impact search engines have on the popularity evolution of Web pages. For example, given that search engines return currently "popular" pages at the top of search results, are we somehow penalizing newly created pages that are not very well known yet? Are popular pages getting even more popular and new pages completely ignored?"

We report on initial research on the concurrency control issue of compiled database applications. We focus on decoupling concurrency control from any functionality relating to recovery. Because it is the possibility of transaction aborts for deadlock resolution that makes the recovery subsystem necessary, we choose the deadlock-free tree locking (TL) scheme for our purpose. With the knowledge of transaction workload, efficacious lock trees for runtime control can be determined at compile-time. Our experimental results show TL produces better throughput than the traditional two-phase locking (2PL) when the transactions are write-only; and for main-memory data, TL performs comparably to 2PL even in workloads with many reads.