On Coding Navigation Paths for In-Memory Navigation in Persistent Object Stores Markus Kirchberg 1 , Alexander Kuckelberg 2 , Klaus-Dieter Schewe 1 , Alexei Tretiakov 1 1 Massey University, Department of Information Systems Private Bag 11 222, Palmerston North, New Zealand 2 RWTH Aachen, Chair of Railway Studies and Transport Economics Mies-von-der-Rohe-Str. 1, 52056 Aachen, Germany Abstract We consider matrix index and navigation index approaches to in-memory navigation of persistent object stores. We demonstrate that both approaches can be re-formulated independently from the used coding technique. We expose the limitations of fixed length coding and the inefficiency of simple continued fractions as a variable length coding technique. The conclusion is that alternative variable length coding techniques, possibly based on known compression algorithms, should be explored. The architecture of persistent object stores should permit plugging in and configuring coding algorithms to optimise the operation of a particular system. 1. Introduction Persistent Object Stores (POSs) provide the core of the physical architecture of a database management system [6]. The term arose in the context of object oriented databases (OODBs), for which it became necessary to consider complex objects with identifiers, values of complex types and references [3, 16]. While the interest in OODBs has faded, persistent object stores are still extremely important, e.g. for realising XML-based databases, for which again complex tree-like values and references have to be supported [1]. The type of access to data in a POS can be classified into three groups: direct: In this case a physical object is retrieved by providing its unique object identifier. This access method is always implicitly present, even in relational databases, where tuple identifiers are used on the physical level. associative: In this case objects are retrieved by providing some key values. This is the most common access method for relational databases, which has led to various methods for hashing and index-based approaches such as B-trees, B + -trees, grid-files, etc. For spatial databases these access methods have been generalised to R-trees and R * -trees [2, 9]. navigational: Navigational access is the only access method that does not have an analogue for relational databases. Roughly speaking navigational access retrieves objects by following the references between objects forward and backward. The major focus in POS research has been on navigational access [5, 10, 12, 13, 14, 19, 20, 21]. The approach by Subieta in [19] to navigational access is based on physical data structures on spiders of rings and thus is rather similar to old techniques used in network database systems.