File transfers across optical circuit-switched networks

Recent technological advances allow for the dynamic setup and release of end-to-end circuits consisting of Ethernet segments at the ends mapped on to Ethernet-over-SONET long-distance optical circuits. We call these Dynamically Reconfigurable Ethernet/Ethernet-over-SONET (DREEoS) circuits. For file transfers across end hosts that can be connected by a DREEoS circuit, we propose that, in most cases, the sending end host should first attempt to set up a DREEoS circuit, and if rejected, fall back to the TCP/IP path. If the DREEoS circuit setup is successful, the end host will enjoy a much shorter file transfer delay than with the TCP/IP path. For example, a 1GB file transfer on a TCP/IP path with a round-trip time of 50ms, link rate of 1Gbps, and a loss probability of 0.0001 takes 395.7sec, while on a DREEoS circuit with the same link rate, the transfer time is 8.08sec. The availability of the fallback TCP/IP path allows DREEoS service to be introduced gradually into optical networks. At low loads, the network can be operated at high call blocking probabilities to achieve high utilization. As loads increase, the network can be engineered to retain high utilization while simultaneously offering low call blocking probabilities. An important component of this proposal is hardware acceleration of signaling protocol implementations. This results in low call setup delays allowing for the DREEoS option to be attempted even for small file sizes. We compare mean delay incurred with a DREEoS circuit attempt against the mean delay incurred with directly choosing the TCP/IP path for different values of call blocking probability in the circuit-switched network, probability of packet loss in the IP network, round-trip times, link rates, etc. 1. Background and problem statement There is a growing interest in improving current protocols or developing new ones to increase the effective throughput of file transfers on the Internet [1-5]. Of particular interest is the effective throughput of large file transfers, e.g., petabyte ( ) and exabyte ( ) sized file transfers created in Particle Physics, Earth Observation, Bioinformatics, and Radio Astronomy studies [5]. In the general Internet setting, this problem should be solved for file transfers between end hosts connected by any heterogeneous end-to-end path. Solutions for such settings are being developed by others [3][4]. We consider a slight variant of this problem. Consider the case when the two communicating end hosts are located on the same network. Then the question is should these two hosts continue using TCP/IP for their communication, or should they use protocols tailor-made for intra-network communication? The original design tenet of TCP/IP protocols was to use the same protocols irrespective of the end-to-end path to simplify protocol implementations [6]. However, this sacrifices performance. For example, consider the case of two communicating end hosts connected by a direct high-speed link. Given TCP does not examine the nature of the end-to-end path before data transfer, it will run the usual congestion control algorithms such as Slow Start. Even without packet losses, the Slow Start congestion algorithm will lead to poor performance especially in high-speed long-distance environments. For example, if round-trip propagation delay on a 1Gbps link is 50ms, then a 100MB file transfer experiences an effective throughput of 0.54Gbps because of Slow Start. As link speeds increase to 10Gbps, effective throughput will be small even in lower propagation-delay environments. The specific intra-network setting we consider is a wide-area optical circuit-switched network. The problem statement of this work is to develop protocols for fast file transfers across an optical circuit-switched network. This is not a pure academic exercise. Instead, a few recent technological advances have made implementation of this concept quite feasible. These advances include deployment of (i) optical fiber to enterprises, (ii) Multi-Service Provisioning Platforms (MSPPs) in enterprises, and (iii) Ethernet over SONET (EoS) capabilities in MSPPs. Before we describe these advances, we clarify our problem definition with an illustration. Fig. 1 shows the Internet as a global network of IP routers that interconnects different types of networks grouped into sets. Set 1 is a set of high-speed optical circuit-switched networks consisting of SONET/SDH/WDM1 crossconnects, Add/Drop Multiplexers (ADMs), etc. Some of these networks could be all-optical, with optical links and all1. SONET/SDH/WDM: Synchronous Optical NETwork/Synchronous Digital Hierarchy/Wavelength Division Multiplexing 10 15 10 18