End-to-end asymmetric link capacity estimation?
Ling-Jyh Chen, Tony Sun, Guang Yang, M. Y. Sanadidi, Mario Gerla
Computer Science Department, UCLA, Los Angeles, CA 90095, USA
Abstract. Knowledge of link capacity is important for network design,
management, and utilization. With the increasing popularity of asym-
metric link technologies (such as DSL, 1xRTT, and satellite links), it is
desirable to have a capacity estimation technique, which can simulta-
neously measure forward and backward direction link capacities on an
Internet path. Moreover, this estimation must often be “sender only”,
because of receiver limitations or lack of standards. In this study, we
propose a simple, fast and accurate technique, called AsymProbe, to es-
timate asymmetric link capacities. AsymProbe is a “sender only”, round
trip procedure. It achieves asymmetric link capacity estimation by strate-
gically altering the ratio of probe and acknowledgement packet sizes. Us-
ing simulation and testbed experiments, we validate AsymProbe with a
variety of network configurations. The results show that AsymProbe can
correctly estimate the asymmetric link capacities as long as an appropri-
ate packet size ratio can be employed.
Knowledge of link capacity is particularly important for network design, manage-
ment and utilization. A simple and accurate scheme for capacity measurement
and monitoring is becoming increasingly desirable, especially for emerging tech-
nologies and applications such as overlay, peer-to-peer (P2P), sensor, grid and
mobile networks. A successful capacity estimation solution will need to encom-
pass speed of execution, simplicity, accuracy, and extendibility beyond the limits
of traditional networks, in particular the increasingly popular asymmetric access
methods to the Internet, e.g. DSL, cable modem and satellite links. It is also of-
ten imperative to carry out the estimation in a round trip, “sender only” fashion.
This is because the receiver is not powerful enough to implement the estimation
algorithm. It must, however, participate in the response to probe packets.
Several capacity estimation methods exist, including CapProbe , which is
sender only and fast. However, sender only methods so far have addressed sym-
metric path extimations, ie, the minimum capacity is the same in both directions.
Yet, asymmetric links do exist; moreover, many applications are intrinsically
asymmetric too and thus can benefit from the knowledge of such asymmetries.
?This material is based upon work supported by the National Science Foundation
under Grant No. CNS-0435515.
For example, in multimedia streaming and file downloading, bulk data is trans-
mitted only in the forward direction, consuming much more bandwidth than the
control traffic in the reverse direction. In this case, the knowledge of one-way
capacity on the forward direction link is mandatory, as it is a much better predic-
tor of the streaming or downloading rate, than the blindly measured round-trip
bottleneck capacity, accounting for the cases when the forward link has larger
capacity than the backward link.
Previous approaches on capacity estimation can be divided into two cat-
egories: one-way probing (e.g. Pathrate ) and round-trip probing (e.g. Cap-
Probe ). In , a thorough comparison of modern capacity estimation methods
was presented, where CapProbe was especially singled out as a fast and accurate
capacity estimation mechanism addressing both wired and wireless links. How-
ever, limited by its round-trip nature, CapProbe only works well on symmetric
links. When operating on an asymmetric link, CapProbe measures the narrower
capacity of the two directions. It cannot distinguish the respective capacities of
the forward and backward links.
Even though capacity estimation for asymmetric links can be achieved by
conducting single direction capacity probing (e.g. Pathrate) for the two direc-
tions separately, this estimation strategy is often considered undesirable, as it
imposes unnecessary computation overhead and complexity on the receiver (eg,
the mobile host). Moreover, it requires compatible software and consistent com-
putation methods in both hosts. To simplify the process of estimating asymmet-
ric link capacities, round trip capacity probing is still the most desirable solution.
Still, existing method like CapProbe, lacked such a capacility, modifications are
needed to add support for accurate capacity estimations of asymmetric links.
To this end, in this study, we propose and evaluate a round trip technique
for estimating asymmetric link capacities called AsymProbe. AsymProbe is engi-
neered based on the well proven CapProbe mechanism. Through careful selection
of probe and acknowledgement packet sizes, AsymProbe can successfully provide
simple, fast, and accurate capacity estimates for asymmetric links.
The rest of the paper is organized as follows. In section 2, we survey and
summarize work related to this study. An in-depth description of AsymProbe
follows in Section 3. In section 4, we evaluate the accuracy of AsymProbe in esti-
mating link capacity through series of NS2 simulations. In section 5, we present
results from our testbed experiments to validate the capability of AsymProbe.
Section 6 concludes the paper.
2 Background and Related Work
Previous research on capacity estimation relied on either delay variations among
probe packets as illustrated in pathchar , or dispersion among probe packets
as described in Nettimer  and Pathrate . The analysis in  clearly revealed
that the dispersions distribution can be multi-modal without multi-channels,
and that the strongest mode in the multimodal distribution of the dispersion
may correspond to either (1) the capacity of the path, or (2) a “compressed”
Fig.1. (a) under-estimation caused by “expansion” (b) over-estimation caused by
“compression” (c) the ideal case. (T?: Measured dispersion; Tqueue: Queueing delay)
dispersion, resulting in capacity over-estimation, or (3) to the Average Dispersion
Rate (ADR), which is lower than the capacity.
Other tools such as pchar and clink  use variations of the same idea as
pathchar. Pchar employs regression techniques to determine the slope of the
minimum RTT versus the probing packet size. However, pathchar-like tools have
limitations with respect to the speed of estimation process as shown in .
CapProbe  is a recently proposed capacity estimation technique shown to
be both fast and accurate over a large range of scenarios. When a back-to-back
packet pair is launched into a network, it is always dispersed at the bottleneck
link according to the bottleneck capacity. If such dispersion is preserved until
the pair arrives to destination, it identifies the bottleneck capacity (as shown in
Fig. 1-c). Unfortunately, the dispersion can be either expanded or compressed,
where “expansion” of dispersion leads to under-estimation and “compression”
of dispersion leads to over-estimation of the capacity, as shown in Fig. 1-a,b.
To overcome this problem, CapProbe combines the use of dispersion and
end-to-end delay measurements thus filtering out packet pair samples distorted
by cross traffic. Whenever an incorrect value of capacity is estimated, either the
first or the second packet, or both, have been delayed by cross traffic. In this
case, the sum of the delays of the two packets in the packet pair, called the
delay sum, includes some queuing delay. A delay sum that does not include any
queuing delay introduced by cross traffic is referred to as the minimum delay
sum. The dispersion of such a packet pair sample is not distorted by cross traffic
and reflects the actual capacity. A valid sample can easily be identified since its
delay sum is the minimum among delay sums of all packet pair samples. The
capacity is then estimated by the equation:
where P is the sampling packet size, and T is the dispersion of the sample packet
pair with the minimum delay sum.
The majority of the existing capacity estimation tools, including the ones
discussed above, are inherently round-trip based. They estimate the narrowest
capacity on the round-trip path. These techniques encounter severe constraints
when measuring link capacities of increasingly popular asymmetric links, such
as DSL, cable modem and satellite links where the forward link capacity is very
different from the backward link capacity. In this study, we propose AsymProbe,
a novel scheme that measures asymmetric link capacities in the round trip fash-
ion. Details of this proposed approach will be presented in the following sections;
evaluation of AsymProbe will be discussed in the simulation and experiments
3 Proposed Approach: AsymProbe
In this section, we present AsymProbe, a novel capacity measuring technique
that allows to measures the capacity of either the forward or backward narrow
link on the path. The basic idea of AsymProbe stems from the observation that
the measured dispersion in the original CapProbe can be introduced either in
the forward or backward direction of an asymmetric link. When probing and
acknowledgement packets are of same size, the measured dispersion is good for
estimating the round-trip bottleneck capacity, since the narrowest link along the
round-trip path gives the largest dispersion to the (probing or acknowledgement)
packet pairs. One can then easily estimate this capacity by applying Eq. 1.
Fig. 2 depicts the packet pair interactions in an asymmetric link scenario,
with link capacity C1on the forward direction link and capacity C2on the back-
ward direction link. The probe packets are sent back-to-back with packet size
P1on the forward direction link (from A to B); the acknowledgement packets
are sent immediately upon receipt of probe packets with packet size P2on the
backward direction link (from B to A). Suppose T1and T2represent the respec-
tive dispersions of probe packets and acknowledgement packets when they are
sent back-to-back on the link; from the definition of Eq. 1, T1and T2can then
be derived as T1=P1
Fig.2. Interaction of probe packets in asymmetric link scenarios
Table 1. Estimate asymmetric link capacity by varying packet sizes (ideal case without
cross traffic and any queuing delays)
Probe (P1) and ACK (P2)
T?→ T1= T2
T?→ max(T1,T2) C?
The dispersion measured at the end host A, denoted as T?, is the dispersion
between back-to-back acknowledgement packets. Suppose T1> T2, this means
that measured dispersion T?equates to T1. We assume that host B immediately
acknowledges the probe packets without incurring additional queuing delay, else
the min sum condition would be violated and the pair discarded. On the other
hand, suppose T1< T2, then T?reflects T2instead, i.e. the dispersion generated
on the backward direction link prevails. Therefore,
T?= max(T1,T2) (2)
By varying the packet size ratio between the probe and the ACK packets,
and observing the forward link capacity estimate (C?
backward link capacity estimation (C?
the correct capacity estimations for both directions of the link. For instance,
suppose C1> C2and the initial packet size P1= P2, it can be concluded that
T?measured equates to T2. Therefore, C?
C1. The estimated capacity is the round-trip bottleneck link capacity on the
asymmetric link (the minimum value of C1 and C2), which is exactly is what
CapProbe estimates as presented in .
However, by increasing P1gradually, C?
creases and approaches C1gradually. When P1increased toP2×C1
to C1and C?
(i.e. T1> T2, since
asymmetric link capacities. Table 1 below details this relationship.
Based on the relationship presented in the table, the AsymProbe algorithm
consists of four phases, of which the first three phases are Probing phases, and
the last is the Decision phase. Two packet sizes are used in the probing phases:
Pmaxand Pmin, which are chosen carefully by taking network and system issues
into account. In the first probing phase, P1and P2are both set to Pmax. Thus
we estimate the bottleneck capacity, Clow, of the round trip path. In phase 2
and 3, (P1,P2) are set first to (Pmax,Pmin) and then to (Pmin,Pmax) in order to
1, which is
T?) and the
2, which isP2
T?), the source node can obtain
C1. From the discussion above, the end-to-end dispersion
T? = C2 and C?
T? = C2 <
2remains equivalent to C2, but C?
2converges to C2. Conversely, after P1increased to larger thanP2×C1
T? < C2. This simple relationship between the estimated capacity (C?
2) and the varying packet size can be harvested for the accurate estimation of
C2), T?will reflect T1. As a result, C?
T? = C1and
Fig.3. AsymProbe Algorithm (The Decision Phase)
estimate the forward and backward link capacities respectively. We use C[i]
and C[i] to denote the estimation results of C?
In the fourth phase, namely the Decision phase, a decision algorithm is per-
formed to determine the estimation results of both direction links from all C[i]
and C[i] as shown in Fig. 3. However, it should also be mentioned that the
capability of AsymProbe in determining the larger capacity is mathematically
bounded by the maximum ratio of packet sizes between probe and acknowledge-
ment packets, i.e. the max of
of all 3 phases are equal, and we know a priori that the link is asymmetric, then,
the packet size ratio is not sufficient large to provide an accurate capacity esti-
mation of the larger link. Therefore, AsymProbe is unable to estimate the actual
capacity in the direction with higher speed, but will indicate that such condi-
tion has occurred and report a “lower-bound” (i.e.Pmax
instead. In this case, one-way capacity estimation tools (e.g. one way version of
CapProbe or Pathrate) can be applied to accurately measure the capacity in
this direction - if this solution is feasible within the scope of the application. In
section 5.4, we discuss another extension of AsymProbe to this problem.
2in the i-th phase,
P1. Specifically, if the capacity estimates
Pmin×Clow) of the capacity
In this section, we present simulation results that evaluate the accuracy of ca-
pacity estimation of AsymProbe on paths with asymmetric links. AsymProbe is
implemented in the NS-2 simulator . Fig. 4 depicts the simulation topology
that represents a commonly seen scenario nowadays with an asymmetric DSL
link. All links are symmetric 100Mbps Ethernet links except the one between
Fig.4. Last-hop ADSL scenario. The link capacities are 100Mbps for all links, except
the asymmetric DSL link between D and E (D → E : 128Kbps;E → D : 1.5Mbps)
Table 2. Simulation results of AsymProbe in last-hop DSL scenarios (Unit: Kbps)
AsymProbe from A AsymProbe from B CapProbe
A → B
B → A
B → A
A → B
A ⇔ B
FTP (B → C)
FTP (C → B)
128 1500 1505 128.057128
128.0621500 1500128 128
Poisson (B → C, rate=300Kbps)
Poisson (B → C, rate=750Kbps)
Poisson (B → C, rate=1500Kbps)
Poisson (C → B, rate=25.6Kbps)
Poisson (C → B, rate=64Kbps)
Poisson (C → B, rate=128Kbps)
1281500 1500128 128
1281500 1500128 128
127.988 15001483.143128.039 128
node D and E, which is an asymmetric DSL link with 1.5Mbps downlink capac-
ity (from E to D) and 128Kbps uplink capacity (from D to E). Nodes to the
left of node D (namely A and C) belong to a home networks, while nodes to the
right of node E are on the Internet.
The AsymProbe estimation is performed on the path between node A and B.
In addition to the AsymProbe flow, various types of cross traffic were generated
on the DSL link to test AsymProbe robustness. The cross traffic types used were
FTP and Poisson based UDP traffic of different rates. For the Internet segment,
long range dependent (LRD) traffic is created between node E and F in both
directions. The LRD traffic is composed of 16 Pareto flows with alpha = 1.9 ,
and the overall rate of LRD traffic is 60Mbps in each direction.
The maximum and minimum AsymProbe packet sizes, Pmaxand Pmin, are
set to 1500 bytes and 100 bytes respectively. For the various cross traffic configu-
rations described in Table. 2, AsymProbe is independently initiated from both A
and B; results obtained from AsymProbe are then compared against CapProbe
as summarized in Table 2.
Fig.5. Testbed for NIST Net experiments
From the results shown in Table 2, AsymProbe is able to estimate the correct
link capacity in both directions for all test cases; whereas CapProbe can only es-
timate the bottleneck link capacity of the round-trip path. Moreover, simulation
results also show that AsymProbe works when placed on either the end-client
(node A) or the Internet server (node B). The results are consistent in both
It is also worth mentioning that since the link capacity ratio of the simu-
lated scenario is 1.5Mbps/128Kbps, it is smaller than the packet size ratioPmax
AsymProbe is thus able to measure the correct link capacities. However, if we
decrease the packet size ratio (e.g. increasing Pmin in order to avoid the fine
time resolution problem as described in ) and obtain a packet size ratio that
is larger than the link capacity ratio, AsymProbe will only estimate the correct
capacity of the narrower link and output the other direction link as a lower
bound estimation, defined as
estimation tools can be launched.
Pmin× Clow. In such case, one-way link capacity
In this section, we present testbed and Internet experimental results to further
evaluate AsymProbe. We first perform a set of experiments on a “controlled”
testbed to calibrate and verify the correctness of the AsymProbe scheme and its
Linux implementation. We then move to Internet measurements for an evaluation
in the diverse and realistic scenario.
The testbed experiments are performed in the configuration shown in Fig. 5.
The NISTNet emulator  is used to set up the asymmetric bottleneck link of
various capacities. A backlogged file transfer session is generated from the FTP
server to the client as cross traffic. This FTP connection shares the bottleneck
link with an AsymProbe connection that traverses from host A to B.
For reasons we will discuss shortly, we choose 1500 bytes and 500 bytes as
the maximum and minimum packet sizes in this set of experiments, respectively.
Thus we have
and Table 4, in which we may see that when the forward/backward (Table 3)
500= 3. We present the experiment results in Table 3
Table 3. NIST Net results on High/Low
asymmetric links (Unit: Mbps)
1.063 1.063 0.985
F: Forward Link; B: Backward Link
Table 4. NIST Net results on Low/High
asymmetric links (Unit: Mbps)
F: Forward Link; B: Backward Link
or backward/forward (Table 4) capacity ratio is below 3, AsymProbe measures
both forward and backward capacities very accurately. When the ratio increases
beyond 3, only a lower bound can be obtained in the direction with the larger
5.2 Internet Experiments
In addition to the controlled testbed experiments, we also perform a set of In-
ternet measurements to evaluate AsymProbe in a more diverse and realistic
scenario. In this set of experiments, again we have
the asymmetric links we have found, provided by DSL1and Cable2companies,
all have a higher down-link/up-link capacity ratio than 3. As presented in Table
5, AsymProbe captures the up-link capacities accurately, while only obtaining
lower bounds for the down-links.
500= 3. However,
From the simulation and experiment results above, AsymProbe is capable of
estimating asymmetric link capacities, as long as the capacity ratio of the forward
and backward links is within the range of the packet size ratio of the employed
probe and acknowledgement packets. In order to increase the estimation range
of AsymProbe, the packet size ratio should be as large as possible. However, this
ratio is bound by implementation.
1DSL 1 is provided by Verizon: http://www.verizon.com; DSL 2 is provided by Hinet:
2Cable Modem is provided by Comcast: http://www.comcast.com
Table 5. Internet results on asymmetric link
Claimed Capacity Estimated Capacity
DSL 1 1.5 Mbps 128 Kbps
1 ≥ 379 Kbps 132 Kbps
2 ≥ 382 Kbps 132 Kbps
3 ≥ 380 Kbps 132 Kbps
1 ≥ 1.49 Mbps 565 Kbps
2 ≥ 1.53 Mbps 567 Kbps
3 ≥ 1.51 Mbps 558 Kbps
1 ≥ 721 Kbps 247 Kbps
2 ≥ 730 Kbps 255 Kbps
3 ≥ 723 Kbps 248 Kbps
DSL 2 3 Mbps 512 Kbps
Cable Modem3 Mbps 256 Kbps
Specifically, the maximum size of the employed packets must be bounded
by the Maximum Transmission Unit (MTU), which is the largest size of an IP
datagram allowed to transmit on the path without fragmentation. The size of
MTU may vary greatly in different system configurations. However, practically
it is set to 1500 bytes in most networks. Packets larger than MTU will be seg-
mented into smaller fragments for transmission and then reassembled on the
receiving host. Therefore, using packets larger than MTU is not appropriate for
CapProbe-based capacity estimation techniques, since the dispersion measure-
ment no longer reflects the bottleneck capacity.
On the other hand, the minimum size of AsymProbe packets is also bounded
in accordance with the supported time resolution on the estimating host. This
is due to the fact that a packet pair with a smaller packet size will result in
a smaller inter-packet dispersion, which in turn requires a finer time resolution
to be measured accurately. Assume the capacity of the narrow link is C and
the probing packet size is P, the dispersion time (and also the clock granularity
needed for accurate estimation) that needs to be measured is T = P/C. Table
6 shows the required clock granularities that are needed for different probing
packet sizes and narrow link capacities.
Table 6. Required time resolution for accurate estimation
Narrow Link Capacity
1 Gbps 100 Mbps 10 Mbps 1 Mbps
100 bytes0.0008 ms 0.008 ms 0.08 ms 0.8 ms
500 bytes0.004 ms 0.04 ms0.4 ms 4 ms
1000 bytes 0.008 ms0.08 ms 0.8 ms 8 ms
1500 bytes 0.012 ms0.12 ms1.2 ms 12 ms
It is clear that the time resolution of an end host relies on the hardware
speed and the operating system. A system with fast processors and I/O inter-
faces can provide a finer time resolution. Additionally,  also shows that the
accuracy of CapProbe-based capacity estimation is tightly related to the runtime
execution mode. With kernel mode implementations, the capacity estimation is
faster and more accurate than with user mode implementations. Kernel mode
implementations also provide better time resolutions. Therefore, kernel mode
implementations can use smaller packets for capacity estimation than the user
In the presented testbed experiments, the employed packet sizes are bounded
with 1500 bytes as the maximum and 500 bytes as the minimum. The value of
1500 bytes is determined by the MTU on the path, whereas the value of 500
bytes is the minimum packet size which can measure the dispersion accurately
with the provided machine time resolution. Thus it is only capable of estimating
an asymmetric link with capacity ratio up to 1500 : 500, namely 3 : 1. For those
links with even higher “asymmetric ratios” (e.g. 1.5Mbps/128Kbps DSL links
or 400Kbps/64Kbps satellite links), it is necessary to increase the packet size
ratio by either increasing the maximum packet size or decreasing the minimum
packet size. In such cases, using a faster machine or switching from user mode
to kernel mode can help.
If all of the above procedures do not work, one can resort to one way capacity
estimation as mentioned in Section 3. This, however, requires full implementa-
tion on the receiver. If the receiver does not cooperate, a possible solution is
to use a packet “train” probing concept as suggested by other researchers .
The intent is to replicate the effect of a “long” probing packet without paying
the penalty of reassembly at the host. To illustrate the technique, consider for
example the situation of an asymmetric satellite link with 15Mbps downlink and
128Kpbs uplink capacity. The server in the Internet must determine the down-
link speed to deliver the proper content to a mobile user. The downlink capacity
estimation can be achieved by transmitting a train of 10 consecutive 1500 byte
packets, with a Probe leader and Probe trailer. As before, the two Probe pack-
ets each trigger a 100 byte packet probe response from the receiver. The train
dispersion in the forward link is preserved in the ACK dispersion measured by
the sender after the round trip and provides the desired estimate. Basically, this
scheme is an extension of AsymProbe, where the source experiments with trains
of increasing length until success. As pointed out in , the longer the train, the
less dominant the mode corresponding to the forward narrow capacity. Conse-
quently, the less frequent the train samples where no delay/interference occurred
along the path and thus the less accurate the measurements. The measurement
however, provides a conservative (lower bound) estimate of the narrow forward
capacity, which can be progressively improved as more and more samples are
collected. By the way, the “average” capacity measurement (as opposed to min
sum measurement) was shown to converge for large train length N to a value
between the narrow link and the residual link bandwidth. As expected, the lower
the utilization, the faster the min sum measurement convergence .
In this paper, we studied asymmetric link capacity estimation and proposed an
extension of CapProbe, namely AsymProbe, to estimate asymmetric link capac-
ities. By strategically altering the ratio of probe and acknowledgement packet
sizes, AsymProbe can simultaneously measure the link capacities of both forward
and backward direction links. Through simulation and testbed experiments, we
validated the accuracy and capabilities of our proposed approach.
The unique advantage of AsymProbe is the ability to measure capacities from
the server using a round trip method that does not require the cooperation of
the receiver (which may have limited processing power or may be altogether
unaware of AsymProbe). Moreover, the technique is extremely fast, thus it is
suitable for mobile receivers that experience rapidly varying, often asymmetric
Internet connectivity. The simplicity, accuracy, and speed of AsymProbe make
it ideal in real deployments where online and timely capacity estimation is re-
quired. Better service can be provided by estimating both forward/backward
direction path capacities. Typical applications feature the efficient transfer of
multimedia files over rapidly varying Internet paths (which may include wire-
less segments). Popular examples are P2P streaming and file sharing, overlay
network structuring, and intelligent vertical handoff decision.
1. C. Dovrolis, P. Ramanathan, and D. Moore. What do packet dispersion techniques
measure? In Proc. of IEEE Infocom 2001.
2. A. B. Downey. Using Pathchar to Estimate Internet Link Characteristics. In Proc.
of ACM SIGCOMM 1999.
3. V. Jacobson. Pathchar: A tool to infer characteristics of Internet paths.
4. R. Kapoor, L.-J. Chen, L. Lao, M. Gerla, M. Y. Sanadidi. CapProbe: A Simple and
Accurate Capacity Estimation Technique. In Proc. of ACM SIGCOMM 2004.
5. R. Kapoor, L.-J. Chen, M. Y. Sanadidi, M. Gerla. Accuracy of Link Capacity Es-
timates using Passive and Active Approaches with CapProbe. In Proc. of ISCC
6. K. Lai and M. Baker. Measuring Bandwidth. In Proc. of IEEE INFOCOM 1999.
7. M.S. Taqqu, W. Willinger, R. Sherman. Proof of a fundamental result in self-similar
traffic modeling. SIGCOMM Computer Communications Review, 27: 5-23, 1997.
8. Network Simulator (NS-2). www.mash.cs.berkeley.edu/ns/
9. NIST Net. http://snad.ncsl.nist.gov/itg/nistnet/