IEEE TRANSACTIONS ON COMPUTERS, VOL. X, NO. X, AUGUST 2013 1 Scheme to Measure Packet Processing Time of a Remote Host through Estimation of End-Link Capacity Khondaker M. Salehin, Roberto Rojas-Cessa, Chuan-bi Lin, Ziqian Dong, and Taweesak Kijkanjanarat Abstract—As transmission speeds increase faster than processing speeds, the packet processing time (PPT) of a host is becoming more significant in the measurement of different network parameters in which packet processing by the host is involved. The PPT of a host is the time elapsed between the arrival of a packet at the data-link layer and the time the packet is processed at the application layer (RFCs 2679 and 2681). To measure the PPT of a host, stamping the times when these two events occur is needed. However, time stamping at the data-link layer may require placing a specialized packet-capture card and the host under test in the same local network. This makes it complex to measure the PPT of remote end hosts. In this paper, we propose a scheme to measure the PPT of an end host connected over a single- or multiple-hop path and without requiring time stamping at the data-link layer. The proposed scheme is based on measuring the capacity of the link connected to the host under test. The scheme was tested on an experimental testbed and in the Internet, over a U.S. inter-state path and an international path between Taiwan and the U.S. We show that the proposed scheme consistently measures PPT of a host. Index Terms—Active measurement, intra-probe gap, link capacity, one-way delay, packet processing time, interrupt coalescence. ✦ 1 I NTRODUCTION P ACKET processing time (PPT) of a host is the time elapsed between the arrival of a packet in the host’s input queue of the network interface card, NIC, (i.e., the data-link layer of the TCP/IP suite) and the time the packet is processed at the application layer [1], [2]. As link rates increase faster than processing speeds [3]–[6], the role of PPT becomes more important in the measurement of different network parameters. One-way delay (OWD) in a local area network (LAN) is an example of a parameter that PPT can significantly impact [7]. Figure 1 illustrates the OWD of packet P over an end- to-end path, between two end hosts, the source (src) and the destination (dst) hosts. The figure shows the different layers of the TCP/IP protocol stack that P traverses at both end hosts. According to the RFC 2679 [1], the actual OWD is the wire time that the packet experiences in the trip from src to dst. The wire time includes the transmission time (t t ), the queueing delay (tq ), and the propagation time (tp), or OWD = t t + tq + tp, as Figure 1 shows. However, as time stamping of packet creation at src (PPTsrc) and packet receiving at dst (PPT dst ) takes place at the application layer, the coarsely measured OWD may include these PPTs, as an • K.M. Salehin and R. Rojas-Cessa (Corresponding Author) are with the Department of Electrical and Computer Engineering, New Jersey Institute of Technology, Newark, NJ 07102, USA. E-mail: {kms29, rojas}@njit.edu • C. Lin is with the Department of Information and Communication Engi- neering, Chaoyang University of Technology, Wufeng District, Taichung, 41349, Taiwan. Email: cblin@cyut.edu.tw • Z. Dong is with the Department of Electrical and Computer Engineering, New York Institute of Technology, New York, NY 10023, USA. Email: ziqian.dong@nyit.edu • T. Kijkanjanarat is with the Department of Electrical and Computer Engineering, Thammasat University, (Rangsit Campus), Pathumtani, Thailand. Email: taweesak@engr.tu.ac.th. apparent OWD (OWD’) of packet P between src and dst, or: OWD ′ = PPTsrc + t t + tq + tp + PPT dst (1) Moreover, because of the low transmission rates of legacy systems, PPT has been considered so far negligible (i.e., PPTsrc = PPT dst ≃ 0). Layer 4 Layer 3 Layer 2 Layer 4 Layer 3 Layer 2 Physical Link (Layer 1) src dst P PPTsrc PPTdst P P OWD = tt + tq + tp Fig. 1. End-to-end one-way delay (OWD) of packet P over a single-hop path. As data rates increase, the contribution of PPT on OWD increases, and the error in the measurement of OWD in high-speed LANs can be large if PPTs are neglected. For example, the measurement of OWD between end hosts connected over a 100-Mb/s link using 1500- and 40-byte packets would have errors of 2.5 and 9%, respectively, for PPTsrc = PPT dst =2 µs, an average tq = 40 µs [8], and tp = 0.5 µs, considering a 100-m Fast-Ethernet cable. In these calculations, error = | (OWD−OWD ′ ) OWD |×100 %. This error increases to 108% when the queuing delay is relieved (i.e., tq ≃ 0 µs [8]–[13]) for a 40-byte packet, which constitutes 50% or more of the IP traffic [14], [15]. In a similar scenario, this error is 16% on a 1-Gg/s link (as tp = 25 µs for a 5-km optical cable in Gigabit Ethernet [16]). Therefore, PPT must be considered for an accurate measurement of OWD. Similarly, knowledge of the PPT of servers can be used in financial-trading data centers for identifying which servers