J. Cent. South Univ. Technol. (2007)03−0399−05
DOI: 10.1007/s11771−007−0078−0
Improved sample filtering method for measuring end-to-end path capacity
LI Wen-wei(黎文伟)
1
, TANG Jun-long(唐俊龙)
2
, ZHANG Da-fang(张大方)
1
, XIE Gao-gang(谢高岗)
3
(1. School of Software, Hunan University, Changsha 410082, China;
2. Department of Physics and Electronic Science, Changsha University of Science and Technology,
Changsha 410082, China;
3. Institute of Computing Technology, Chinese Academy of Sciences, Beijing 100080, China)
Abstract: By analyzing the effect of cross traffic (CT) enforced on packet delay, an improved path capacity measurement method,
pcapminp algorithm, was proposed. With this method, path capacity was measured by filtering probe samples based on measured
minimum packet-pair delay. The measurability of minimum packet-pair delay was also analyzed by simulation. The results show that,
when comparing with pathrate, if the CT load is light, both pcapminp and pathrate have similar accuracy; but in the case of heavy CT
load, pcapminp is more accurate than Pathrate. When CT load reaches 90%, pcapminp algorithm has only 5% measurement error,
which is 10% lower than that of pathrate algorithm. At any CT load levels, the probe cost of pcapminp algorithm is two magnitudes
smaller than that of pathrate, and the measurement duration is one magnitude shorter than that of pathrate algorithm.
Key words: network measurement; path capacity; capacity measurement; path delay
1 Introduction
Measuring the capacity of an end-to-end Internet
path is a crucial operation for many network activities.
With knowledge of the capacity of end-to-end paths in
mind, ISPs can locate network bottlenecks, for optimized
capacity planning and bandwidth allocation, and
multimedia applications can select appropriate servers.
Moreover, the recent research in overlay networks and
application layer multicast can also benefit from such
capacity information in better structuring their overlays
and trees.
For measuring the capacity, a series of probing
packets with specific pattern should be sent into the
target path. Guaranteeing the estimation accuracy is a
fundamental requirement for any path capacity
measurement algorithms or tools. Furthermore, because
the capacity measurement activity may be adopted
widely, it also requires the algorithms or tools to keep
minor probe costs and have minor probe duration, for
practical applicable.
Starting with fulfill all or part of these requirements,
a lot of tools have been developed in the community
[1−7]
.
However, these tools are far away for practical
utilizations, as they have either poor estimation precision
or too large probing costs
[8]
. According to the probing
techniques adopted, these tools can be divided into two
categories: variable packet size(VPS) method and packet
pair/train dispersion(PPTD) method
[9]
. Tools such as
pathchar
[1]
, clink
[1]
and nettimer
[2]
, are based on VPS
method. VPS method may yield significant capacity
underestimation errors if the measured path includes
store-and-forward layer-2 devices
[10]
, which makes it not
practical for path capacity measurement. Consequently,
the current improvements of capacity measurement are
focused on PPTD method.
The PPTD technique was originally proposed by
JACOBSON
[11]
for congestion control, then much work
has been done to improve it, and it was adopted for
measuring delay variation
[12]
and path capacity
[3−7]
. In
PPTD technique, the source sends multiple packet pairs
to the receiver. Each packet pair consists of two packets
of the same size sent back to back. The path capacity is
estimated by observing the difference of the two packets’
receiving time at the receiver. The main challenge in
PPTD technique is to find a robust statistical technique
that can extract an accurate capacity estimate from noisy
measurements. COSTANTINOS et al
[7]
introduced the
most accurate capacity estimation tool currently
available: pathrate. However, pathrate needs too large
probing cost and too long probing duration for estimation,
and if the network has heavy load, it also has large
estimation errors.
In this paper, after analyzing the effects of cross
traffic on the delays of probing packet pair, a minimum
delay based samples filtering method for path capacity
measurement was proposed. The measurability of
minimum packet pair delays was also analyzed by
Foundation item: Projects(60473031, 60673155) supported by the National Natural Science Foundation of China; Project(2005AA121560) supported by the
High-Tech Research and Development Program of China
Received date: 2006−06−28; Accepted date: 2006−09−27
Corresponding author: LI Wen-wei, PhD; Tel: +86-731-8821570-8424; E-mail: liww@hnu.cn