Inline bandwidth measurement techniques for gigabit networks

Abstract

We introduce an inline measurement method that can overcome difficulties in measurement task in high-speed network environment, such as short packet transmission intervals and Interrupt Coalescence function deployed in the high-speed Network Interface Cards. The method adjusts the number of packets involved in a packet burst of an active TCP connection, and utilises the inter-intervals of the bursts of the corresponding ACK packets for bandwidth measurement. Experiment results show that the proposed inline measurement method can measure the bandwidth in the network paths at least 1-Gbps or higher.

Overhead in Available Bandwidth Estimation Tools: Evaluation and Analysis

“…In the allocation of time-two methods are known; the first continuously checks the CPU clock and sends the packet when the clock reaches a certain time; for example, in Linux-X86 systems access to the hardware clock requires approximately 1.9µs. Also, an operation such as write() to call the system requires on average 2µs (for CPU's Pentium), therefore a Linux system can only send packets with higher intervals 2 + 1.9 = 3.9µs [27], [66]. The second method is to register in the program sending one Interrupt Service Routine packets (ISR) of interrupt clock hardware, operating systems such as 4.4BSD and Linux 2.6 kernels, spend between 1s to 0.0001s on each call.…”

“…The complexity of the problem has led several authors to make erroneous considerations on the behavior of their estimation tools [21][22][23][24]. The inability of the software running on application layer to prioritize the probing packet delivery process, introduces additional errors since the transmission times of the packets does not correspond to the theoretical value determined by the tool [25][26][27]. Although new ABET have managed to significantly reduce the probing traffic, it has still a significant impact given the need for repetitive testing to obtain a single estimation [4], [28].…”

A new form of DOS attack in a cloud and its avoidance mechanism

“…Interrupt coalescence is necessary to reduce the CPU load, but it introduces significant timing error that makes it difficult to estimate available bandwidth accurately. For a full discussion on the timing issues, we refer interested readers to [18] [13].…”

“…To cope with the timing inaccuracy at high speed, recent proposals for bandwidth estimation, such as PBProbe [3] and ICIM-abw [18], send a burst of packets in place of a single probe packet. The idea is to enlarge the probe packet beyond the 1500B packet limit.…”

Available Bandwidth Estimation Tools Metrics, Approaches and Performance

“…more used and important but concentrate on new elements like Narrow link, cross-traffic, tight link and add concepts like bulk transfer capacity (BTC), among others. Studies such as [30], [10], [31], [32], [33], [9], [34], [35], [36], [20], [37], [38], [39], [40], [41], concentrate on analyzing the techniques developed, because each author, based on one of the two approaches, creates a technique to optimize the variables of the av_bw metric, such as estimation time, prediction, and relative error. When developing a technique, it is implemented, evaluated and compared with studies such as [42], [11], [12], [43], [44], [45], [46], [47], [22], [48], [20], [49], [16], [50], [51], [13], [52], [53], [54], [55], show the comparative performance between two or more tools evaluated in simulated environments such as using NS-2 or NS-3, and in real network testbed Evaluate protocols, control certain network parameters, see Table 3.…”