An environment-aware transport mechanism for the mobile workforce Sascha Kümmel, Alexander Schill, Thomas Ziegert Dresden University of Technology, Department of Computer Science, Institute for Operating Systems, Databases, and Computer Networks D-01062 Dresden, Germany, Tel.: +49 351 463 8063, Fax.: +49 351 463 8251, email: {kuemmel, schill, ziegert}@ibdr.inf.tu-dresden.de Abstract Common transport systems lack appropriate mechanisms with regard to the problems of mobile computing systems (e.g. temporary inaccessibility, transient network addresses of mobile hosts, along with varying quality of service parameters of physical network connections). Therefore, a need for new environment aware transport mechanisms exists. This paper discusses specific features of GISMO’s (Generic Infrastructure Support for Mobile Objects) infrastructure; in particular, a mobile queuing service - which distributes data in an environment aware manner. We motivate the need for new transport mechanisms, present our concept, as well as some implementation details and an overview of our first experiences with a prototype implementation. Introduction Distributed applications partially running on mobile devices impose several difficulties on system design 1 . The main consequences of mobility and wireless link usage affecting the information interchange are: volatile addresses, intermittent connectivity, frequent phases of no connectivity and high delays commonly resulting in high costs. Conventional transport protocols are not designed with respect to temporary inaccessibility, transient network addresses ((Perkins, 96) tackle this problem for IPv4 Networks) of mobile hosts, along with varying quality of service parameters of physical network connections. Resulting from these intrinsic characteristics of mobile environments the necessity for new "environment aware" transport mechanisms arose. These should incorporate procedures which realize the following features: deferred data transfer by intermediate storing on persistent media, dynamic routing and addressing supported by special locating mechanisms and topology analyzers, detection of QoS parameters and appropriate data volume adaptation prior to the transfer. A QoS- driven adaptation can be achieved by varying the data packet size depending on the average error rate and by data conversion or compression. The semi-parallel transfer of data packets (using multiple send and receive threads) in case of long delay connections obviously decreases the transmission time of RPC/TCP-based transmissions. Some considerable efforts to solve the problem of disconnected operations and temporary inaccessibility are discussed in (Satyanarayanan, 93), (Satyanarayanan, 94), (Huston, 95) and (Kuenning, 94). These solutions mainly focus on file systems, based on caching, operation logging and reintegration. A recoverable queuing service for distributed transaction processing as a solution for reliable operation handling in case of inaccessibility is described in (Dietzen, 92) and (Transarc, 94). These conventional queuing systems do not actually deal with service mobility and resulting reconfiguration as well as they do not cover the problem of QoS adaptation. Mechanisms for the improvement of TCP/IP performance over wireless networks can be found in (Balakrishnan, 95), (Yavatkar, 94). (Bakre, 95) enhance the conventional SUN-RPC-mechanism to support mobility issues. System architecture In this section we will explain the main concepts behind our environment-aware transport system. We call this dedicated component Queuing Service (QS). It is part of a generic system support infrastructure for distributed mobile computing applications (GISMO, see (Schill, 96)). The Queuing Service realizes the transport of arbitrary data units in any size. The atomic data unit the Queuing Service handles is called data body. Data 1 For an exhaustive problem description refer to (Imielinski, 1994)