Conference Paper

On the Interaction Between Internet Applications and TCP.

DOI: 10.1007/978-3-540-72990-7_83 Conference: Managing Traffic Performance in Converged Networks, 20th International Teletraffic Congress, ITC20 2007, Ottawa, Canada, June 17-21, 2007, Proceedings
Source: DBLP


We focus in this paper on passive traffic measurement tech- niques that collect traces of TCP packets and analyze them to derive, for example, round-trip times or aggregate metrics such as average through- put. The seminal work of Zhang (1) has shown that for more than 50% of the TCP connections observed, it is not the network bandwidth that limits the throughput but rather the application or mechanisms such as TCP slow start or too small a receiver window. Certain types of analysis of the network characteristics are meaningful only when performed on TCP traffic that experiences minimal interference by the application. To eliminate such interference, we propose a generic method that partitions the packets of a TCP connection in bulk data transfer and in applica- tion limited periods: The packets of a bulk data transfer period (BTP) experience minimal interference from the application, while the packets of an application limited period (ALP) experience interference from the application that prevents TCP from fully utilizing the network resources because the application does not produce data fast enough. As a proof of concept, we apply our algorithm to public Internet traffic traces and show that unless the effects of the application are filtered out, studying the end-to-end path and traffic characteristics from a network point of view can produce biased results.

Full-text preview

Available from:
  • Source
    • "An isolate th threshold is used for BTP identification. As stated in [1] we distinguish a BTP from an STP by the number of packets in the period (should be more or equal to 130 for the former). "
    [Show abstract] [Hide abstract]
    ABSTRACT: ORCE is an acronym for Online Root Cause Esti-mator for TCP. We present general aspects and the workflow of the ORCE project, describe implementation of the ORCE tool and report preliminary experimentation results. The ORCE tool is used on the sender side of a TCP connection to identify causes that limit the TCP throughput of this connection. ORCE is based on an existing algorithm for root cause analysis [3] but extends the functionality of the tool towards online measurements of the presently active TCP connections. After calculating all necessary metrics and scores ORCE displays the results in a simple interface indicating the chief reasons for throughput constraints for each measured connection. We have tested the tool with a few applications, such as scp and wget.
    Full-text · Article · Apr 2009
  • Source
    • "On the other hand, when the TCP sender needs to wait for the application on top, we call that period an ALP. We presented in [17] an algorithm that identifies the BTPs and ALPs. Once BTPs have been identified, the root cause analysis toolkit analyzes them for TCP and IP layer throughput limitations, i.e. inferring the root causes for the BTPs, which will be the focus of this paper. "
    [Show abstract] [Hide abstract]
    ABSTRACT: TCP is the most widely used protocol for data transfer over the Internet and for most applications the performance metric of interest is throughput. Identifying the reasons for the throughput limitation of an observed connection is a complex task. We present a set of techniques, called the TCP root cause analysis toolkit, that allow users to determine from a passively captured packet trace the primary cause for the throughput limitation. We present the details of the toolkit and its validation and apply the toolkit to carry out a case study of ADSL traffic.
    Full-text · Article · Jun 2008 · Computer Networks
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: Performance of short TCP transfers, e.g., Web browsing, has a direct impact on the way users perceive the health of their Internet access. It is a common belief that TCP performs better with large than with short transfers, as the latters are more likely to time-out and their duration is dominated by the RTT. In this paper, we revisit the performance of short TCP transfers. We highlight the interplay between TCP and the application on top. We show that while losses can have a detrimental impact on short TCP transfers, the application significantly affects the transfer time of almost all short - and even long - flows in a variety of way. Indeed, the application can induce extremely large tear-down times and it can also slow the rate of actual TCP transfers or affect the ability of TCP to recover using Fast Retransmit/Fast Recovery. We illustrate our findings using several traces from realistic networks including DSL, wireless hotspot and a research lab traffic.
    Preview · Conference Paper · May 2009