Andreas Bühmann

Value complete, column complete, predicate complete

Caching is a proven remedy to enhance scalability and availability of software systems as well as to reduce latency of user requests. In contrast to Web caching where single Web objects are accessed and kept ready somewhere in caches in the user-to-server path, database caching uses full-fledged database management systems as caches, close to application servers at the edge of the Web, to adaptively maintain sets of records from a remote database and to evaluate queries on them. We analyze a new class of approaches to database caching where the extensions of query predicates that are to be evaluated are constructed by constraints in the cache. Starting from the key concept of value completeness, we explore the application of cache constraints and their implications on query evaluation correctness and on controllable cache loading called cache safeness. Furthermore, we identify simple rules for the design of cache groups and their optimization before discussing the use of single cache groups and cache group federations. Finally, we argue that predicate completeness can be used to develop new variants of constraint-based database caching.

Database Caching – Towards a Cost Model for Populating Cache Groups

Web caching keeps single Web objects ready somewhere in caches in the user-to-server path, whereas database caching uses full-fledged database management systems as caches to adaptively maintain sets of records from a remote database and to evaluate queries on them. Using so-called cache groups, we introduce the new concept of constraint-based database caching. These cache groups are constructed from parameterized cache constraints, and their use is based on the key concepts of value and domain completeness. We show how cache constraints affect the correctness of query evaluations in the cache and which optimizations they allow. Cache groups supporting practical applications must exhibit controllable load behavior for which we identify necessary conditions. For such safe cache groups, the cost trade-off for record loading and predicate evaluation saving has to be observed during their design. Therefore, we analyze their load overhead and propose a population estimation algorithm to be used for a cache group advisor.

A middleware-based approach to database caching

Database caching supports declarative query processing close to the application. Using a full-fledged DBMS as cache manager, it enables the evaluation of specific project-select-join queries in the cache. In this paper, we propose significant improvements and optimizations – as compared to the well-known DBCache approach – that make our caching concept truly adaptive. Furthermore, we describe an adaptive constraint-based cache system (ACCache) relying on middleware components as a DBMS-independent realization of this approach.

Datenbank-Caching - Eine systematische Analyse möglicher Verfahren

Zusammenfassung.Caching ist ein bewährtes Mittel, um die Skalier- und Verfügbarkeit von Systemen zu steigern sowie die Latenzzeit für Benutzeranforderungen zu verkürzen. Im Gegensatz zum Web-Caching, bei dem einzelne Web-Objekte irgendwo längs ihres Aufrufpfades in der Proxy-Kette vorgehalten werden, setzt Datenbank-Caching „ausgewachsene“ Datenbanksysteme als Caches ein, um dort Satzmengen entfernter Datenbanken möglichst adaptiv verwalten und Anfragen darauf auswerten zu können. Verfahren dazu reichen von separat verwalteten materialisierten Sichten über überlappende, aber replikationsfrei gespeicherte Sichten bis hin zu Cache-Groups, in denen parametrisierte Cache-Constraints den Cache-Inhalt spezifizieren. Wir untersuchen anschaulich die verschiedenen Ansätze und ermitteln daraus eine Klassifikation, die den Lösungsraum zu enthüllen hilft. Im Mittelpunkt steht das Konzept der Prädikatsvollständigkeit: Ein Datenbank-Cache verwaltet vollständige Extensionen von Prädikaten, was ihm ermöglicht, Schlüsse über beantwortbare Anfragen zu ziehen.Abstract.Caching is a proven means of enhancing scalability and availability of software systems as well as of reducing latency of user requests. In contrast to Web caching, where single Web objects are kept ready somewhere along their invocation path in the proxy chain, database caching uses full-fledged database management systems as caches to adaptively maintain sets of records from a remote database and to evaluate queries on them. This procedure can be achieved by a spectrum of techniques ranging from separate materialized views via overlapping, but „replication-free” stored views up to cache groups where parameterized cache constraints specify the contents of the cache. We explore different approaches in an illustrative way and, from the insights gained, we derive a classification scheme which helps to discover the solution space. The key concept for advanced techniques is predicate completeness: A database cache maintains complete extensions of predicates, which enables the system to draw conclusions about locally answerable queries.

Examining the Performance of a Constraint-Based Database Cache

Constraint-based database caching aims at correctly answering SQL query predicates from a local cache database by exploiting constraints that have previously been used in selecting sets of records to be cached from a remote database. In this paper, we take our first steps in looking at performance aspects of our prototype adaptive constraint-based cache (ACCache), which is realized in a middleware manner on top of regular databases. Within our measurement setup, the initial focus is on two central ACCache functions: query processing and cache loading. To demonstrate their time behavior and interaction, we have chosen a scenario based on the TPC-W specification. We conclude with a discussion of our first measurement results.

Ein Schritt zurück ist kein Rückschritt

ZusammenfassungEin Schlüssel zur Erhöhung der Qualität von Web-Anwendungen ist Caching. Während Web-Caching Dokumentfragmente bereithält, die zunehmend aus Datenbank-Daten generiert werden, richtet sich Datenbank-Caching auf die redundante Speicherung dieser Daten selbst. Eine adaptiv verwaltete Teilmenge der Backend-Daten ermöglicht im Cache durch Vollständigkeitseigenschaften die korrekte Auswertung von Anfragen.In Cache Groups für Gleichheitsprädikate, einer Ausprägung des Constraint-basierten Datenbank-Caching, kann die Auswertbarkeit einer Anfrage durch einfache Sondierungsanfragen auf dem Cache-Inhalt entschieden werden. Wir präsentieren ein neues Sondierungsverfahren, das den Cache flexibler und für eine größere Anzahl von Anfragetypen nutzbar macht; dazu gehören unter dem Begriff des negativen Caching auch Anfragen mit leerem Ergebnis.Wir untersuchen, ob sich das neue Sondierungsverfahren weiter verallgemeinern lässt, welche Alternativen sich für seine Umsetzung in konkreten Cache Groups bieten und wie sich der bisherige Ansatz darin einordnet. Das neue Verfahren macht außerdem eine Schwäche in der bisherigen Struktur von Cache Groups deutlich, die durch die Einführung von Kontrolltabellen behoben werden kann, und kann zur Wartung des Cache-Inhalts beitragen. AbstractA key to increasing the quality of web applications is caching. While web caching holds document fragments ready, which are increasingly generated from database data, database caching focuses on the redundant storage of these data themselves. By means of completeness properties, an adaptively managed subset of the backend data permits the correct evaluation of queries in the cache.In cache groups for equality predicates (that is, in an instance of constraint-based database caching) the evaluability of a query can be determined through simple probe queries on the cache contents. We present a new probing procedure that makes the cache usable for more types of queries in a more flexible way; embraced by the term of negative caching, this even includes queries with empty results.We investigate whether the new probing procedure can further be generalized, which alternatives exist for its realization in concrete cache groups, and how the previous approach fits into these. Additionally, the new procedure reveals a shortcoming in the current structure of cache groups, which can be remedied by introducing control tables, and aids in maintaining the cache contents.