IP over photons: how not to waste the waist of the hourglass

IP is often referred to as the waist of the hourglass, because of the simple model the Internet provides packet level communication compared to other approaches (e.g., FR, X.25 or B-ISDN), but runs over a complex plethora of links and switched networks, and is used by a firmament of application and transport protocols. The ever increasing demand for capacity is pushing us past the 40 Gbit/s barrier, and towards the terabit link. Dense mode WDM offers 128 wavelengths at tens of gigabytes per wavelength. However re-tuning takes geological epochs compared to IP packet switching, or flow switching, or even QoS routing time frames. The problem resembles that of IP over ATM or other virtual circuits, but this may be misleading: what is needed is a novel approach to a network architecture, that avoids the same design flaws and pitfalls we have been forced into in IP over ATM (including MPLS). This paper briefly describes why these approaches are misled, and attempts to outline an approach based on network edge packet sorting and scheduling, which may allow us to use pure optical core networks without sacrificing the flexibility of what we might call legacy electron-datagram services.