Ravi S. Prasad

Bandwidth estimation: metrics, measurement techniques, and tools

In a packet network, the terms bandwidth and throughput often characterize the amount of data that the network can transfer per unit of time. Bandwidth estimation is of interest to users wishing to optimize end-to-end transport performance, overlay network routing, and peer-to-peer file distribution. Techniques for accurate bandwidth estimation are also important for traffic engineering and capacity planning support. Existing bandwidth estimation tools measure one or more of three related metrics: capacity, available bandwidth, and bulk transfer capacity. Currently available bandwidth estimation tools employ a variety of strategies to measure these metrics. In this survey we review the recent bandwidth estimation literature focusing on underlying techniques and methodologies as well as open source bandwidth measurement tools.

Effects of Interrupt Coalescence on Network Measurements

Several high-bandwidth network interfaces use Interrupt Coalescence (IC), i.e., they generate a single interrupt for multiple packets sent or received in a short time interval. IC decreases the per-packet interrupt processing overhead. However, IC also introduces queueing delays and alters the “dispersion” (i.e., interarrival time spacing) of packet pairs or trains. In this work, we first explain how IC works in two popular Gigabit Ethernet controllers. Then, we identify the negative effects of IC on active and passive network measurements. Specifically, we show that IC can affect active bandwidth estimation techniques, causing erroneous measurements. It can also alter the packet interarrivals in passive monitors that use commodity network interfaces. Based on the “signature” that IC leaves on the dispersion and one-way delays of packet trains, we show how to detect IC and how to filter its effects from raw measurements. Finally, we show that IC can be detrimental to TCP self-clocking, causing bursty delivery of ACKs and subsequent bursty transmission of data segments.

Socket Buffer Auto-Sizing for High-Performance Data Transfers

It is often claimed that TCP is not a suitable transport protocol for data intensive Grid applications in high-performance networks. We argue that this is not necessarily the case. Without changing the TCP protocol, congestion control, or implementation, we show that an appropriately tuned TCP bulk transfer can saturate the available bandwidth of a network path. The proposed technique, called SOBAS, is based on automatic socket buffer sizing at the application layer. In non-congested paths, SOBAS limits the socket buffer size based on direct measurements of the received throughput and of the corresponding round-trip time. The key idea is that the send window should be limited, after the transfer has saturated the available bandwidth in the path, so that the transfer does not cause buffer overflows (‘self-induced losses’). A difference with other socket buffer sizing schemes is that SOBAS does not require prior knowledge of the path characteristics, and it can be performed while the transfer is in progress. Experimental results in several high bandwidth-delay product paths show that SOBAS provides consistently a significant throughput increase (20% to 80%) compared to TCP transfers that use the maximum possible socket buffer size. We expect that SOBAS will be mostly useful for applications such as GridFTP in non-congested wide-area networks.

Beyond the Model of Persistent TCP Flows: Open-Loop vs Closed-Loop Arrivals of Non-persistent Flows

It is common for simulation and analytical studies to model Internet traffic as an aggregation of mostly persistent TCP flows. In practice, however, flows follow a heavy- tailed size distribution and their number fluctuates significantly with time. One important issue that has been largely ignored is whether such non-persistent flows arrive in the network in an open-loop (say Poisson) or closed-loop (interactive) manner. This paper focuses on the differences that the TCP flow arrival process introduces in the generated aggregate traffic. We first review the Processor Sharing models for such flow arrival processes as well as the corresponding TCP packet-level models. Then, we focus on the queueing performance that results from each model, and show that the closed-loop model produces lower loss rate and queueing delays than the open-loop model. We explain this difference in terms of the increased traffic variability that the open-loop model produces. The cause of the latter is that the flow arrival rate in the open-loop model does not reduce upon congestion. We also study the transient effect of congestion events on the two models and show that the closed-loop model results in congestion-responsive traffic while the open-loop model does not. Finally, we discuss implications of the differences between the two models in several networking problems.

A Stateless and Light-Weight Bandwidth Management Mechanism for Elastic Traffic

Unbounded growth in the number of active flows can lead to severe service quality degradation to existing flows in a network . To guarantee minimum service quality for individual flows, specially for multimedia traffic, it is necessary to estimate and bound the number of active flows in the network. Prior work on estimating the number of active flows has been difficult without maintaining per-flow state. In this paper, we propose a light weight (not requiring per flow state) mechanism to estimate the number of active flows. This estimate is then used to determine the probability of admitting a new flow into the network with the goal of preventing extreme degradation of throughput to existing flows. The key idea here is that the number of active flows can be inferred from the frequency at which a newly arriving packet is part of the same flow as a randomly selected packet in the buffer. Since this scheme relies on information already available in the buffer, no per-flow state is maintained in the network. This mechanism requires very little per-packet processing, even less than a forwarding table lookup. Simulation results show that the proposed scheme can stabilize an overloaded network by bounding the number of active flows without significantly impairing link utilization. This scheme also gives good performance when buffer sizes are small and when the network has a mix of TCP and UDP traffic.

Congestion responsiveness of internet traffic: (a fresh look at an old problem)

The stability of Internet traffic has been attributed to TCP’s congestion control. In this work, we argue that congestion control for individual transfers is not sufficient to produce stable aggregate traffic. The offered load at a network link is generated from users/applications that generate finite-length transfers or group of transfers (sessions). A traffic aggregate is congestion responsive if the session arrival rate decreases when the offered load exceeds the link capacity. We examine two traffic generation models at the session layer. First, a closed-loop model where each user from a fixed-size population can generate a new session only after the completion of her previous session. Second, an open-loop model where sessions arrive independently of previous sessions. These two models produce traffic with very different congestion responsiveness, even if each connection is controlled by TCP. We introduce two metrics to quantify the congestion responsiveness of a traffic aggregate, the throughput responsiveness and the flow rate responsiveness, and show that the closed-loop model results in congestion responsive traffic, while the open-loop model can lead to persistent overload and congestion collapse.