lNTRODUCTIONThe technological evolution of telecommunication networks in recent years has shown that the availability of transmission bandwidth has been growing at a faster pace than the STNitChing bandwidth inside network nodes [4]. This means that in the near future electronic switching will become the perfonnance bottleneck of high-capacity networks. The cost of a Switching device (be it an IP router, an Ethernet switch, or an ATM switch) grows more than linearly with the data rate on switched channels. Today, data rates of 2.5 Gbls are quite commcn, but the technology is ready for 10 Gh/s, and researchers are actively working on 40 Gb/s transmitters and receivers. Further impairments of higb-capacity electronic switching devices are the large required power supply, and the large footprint (i.e., the large volume), which increase the infrastructure CostS.The optical technology 151, [61. [71 has the potential foi-solving some of these limitations, featuring a switching cost that is largely independent from the channel data rate, and smaller power supply and footprint requirements. Thus. while optics today is limited to pin-t-point transmission, it will most likely enter the realm of switching soon. Networks implemeuting switching functions in the optical domain are often called all-opticd networks. More difficult, and far from current technology, is instead the adoption of optical technologies for network control. We refer in this paper to all-optical networks with electronic control of switching devices.Current designs of all-optical networks are based on Op- [IO]. A key functionality for all-optical networking is optical 3R (Regeneration, Rcshaping, Retiming). and a number of components and systems to implement this functionality has recently been studied in research labs I1 11. Optical 3R will enable a decoupling between network design and transmission system engineering, which is fundamental for the success of all-ptical networks. An additional useful functionality for alhptical networking is wavelength conversion [12]. Although subsystems capable of implementing these functions are not available today. we assume that they will be available soon. Packet switching is not well matched with state-f-the-art optical technology, mainly due to the lack of optical memories, and to the lack of optical processing capabilities (i.e., we do not have optical microprocessors). Moreover, optical switching fabrics are typically not capable of being reconfigured in a packet-by-packet fashion (consider that a IWhytes packet lasts approximately 40 ns at 40 Ghls). However, packet switching well matches the dominating IP technology. which today is the main support to user applications. Intermediate solutions between fast circuit switching and packet switching, such as Optical Burst Switching (OBS) [13], are currently studied in optical network?. Long-term network architectures must however address packet switching in the optical domain. This paper focuses on optical packet switching, assuming that most limitations of the opti...