Satellite communications pose serious challenges to transport layer performance, mainly because of long propagation delays (especially in geosynchronous systems) and the possi ble presence of random errors on the satellite link. Solutions that cope with these impairments usually rely upon either the adoption of enhanced versions of transport protocols, or the insertion of intermediate agents, like PEPs (Performance Enhancing Proxies). An alternative approach is to adopt the DTN (Delay/Disruption Tolerant Networking) architecture based on the introduction of the new "bundle" layer in the protocol stack. The aim of the paper is to discuss the advantages and disadvantages of the DTN approach compared to the more conventional solutions mentioned. DTN performance is assessed by considering both fully connected networks and networks where continuous connectivity cannot be guaranteed during data transfer and/or at start-up. Performance is evaluated by using the TATPA testbed (Testbed for Advanced Transport Protocols and Architectures), which is based on a cluster of Linux PCs running the DTNperf application, developed to this end by the authors and now included in the official DTN package. Comparative results show that DTN, coupled with TCP Hybla, outperforms NewReno, offering a performance very close to PEPsal and end- to-end Hybla, which are among the most effective solutions on satellite channels. In addition to these encouraging results, DTN offers an intrinsic greater robustness in intermittent or disruptive environments.