Network Working Group X. Zhu, R. Pan
Internet Draft M. A. Ramalho, S. M. de la Cruz
Intended Status: Informational C. Ganzhorn, P. E. Jones
Expires: March 16, 2015 Cisco Systems
S. DAronco
Ecole Polytechnique Federale de Lausanne
September 12, 2014
NADA: A Unified Congestion Control Scheme for Real-Time Media
draft-zhu-rmcat-nada-04
Abstract
This document describes a scheme named network-assisted dynamic
adaptation (NADA), a novel congestion control approach for
interactive real-time media applications, such as video conferencing.
In the proposed scheme, the sender regulates its sending rate based
on either implicit or explicit congestion signaling, in a unified
approach. The scheme can benefit from explicit congestion
notification (ECN) markings from network nodes. It also maintains
consistent sender behavior in the absence of such markings, by
reacting to queuing delays and packet losses instead.
We present here the overall system architecture, recommended
behaviors at the sender and the receiver, as well as expected network
node operations. Results from extensive simulation studies of the
proposed scheme are available upon request.
Status of this Memo
This Internet-Draft is submitted to IETF in full conformance with the
provisions of BCP 78 and BCP 79.
Internet-Drafts are working documents of the Internet Engineering
Task Force (IETF), its areas, and its working groups. Note that
other groups may also distribute working documents as
Internet-Drafts.
Internet-Drafts are draft documents valid for a maximum of six months
and may be updated, replaced, or obsoleted by other documents at any
time. It is inappropriate to use Internet-Drafts as reference
material or to cite them other than as "work in progress."
The list of current Internet-Drafts can be accessed at
http://www.ietf.org/1id-abstracts.html
Zhu et al. Expires: March 16, 2015 [Page 1]
INTERNET DRAFT NADA September 12, 2014
The list of Internet-Draft Shadow Directories can be accessed at
http://www.ietf.org/shadow.html
Copyright and License Notice
Copyright (c) 2012 IETF Trust and the persons identified as the
document authors. All rights reserved.
This document is subject to BCP 78 and the IETF Trust's Legal
Provisions Relating to IETF Documents
(http://trustee.ietf.org/license-info) in effect on the date of
publication of this document. Please review these documents
carefully, as they describe your rights and restrictions with respect
to this document. Code Components extracted from this document must
include Simplified BSD License text as described in Section 4.e of
the Trust Legal Provisions and are provided without warranty as
described in the Simplified BSD License.
Table of Contents
1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . 4
2. Terminology . . . . . . . . . . . . . . . . . . . . . . . . . . 4
3. System Model . . . . . . . . . . . . . . . . . . . . . . . . . 4
4. Network Node Operations . . . . . . . . . . . . . . . . . . . . 5
4.1 Default behavior of drop tail . . . . . . . . . . . . . . . 5
4.2 ECN marking . . . . . . . . . . . . . . . . . . . . . . . . 5
4.3 PCN marking . . . . . . . . . . . . . . . . . . . . . . . . 6
4.4 Comments and Discussions . . . . . . . . . . . . . . . . . . 7
5. Receiver Behavior . . . . . . . . . . . . . . . . . . . . . . . 7
5.1 Monitoring per-packet statistics . . . . . . . . . . . . . . 7
5.2 Aggregating congestion signals . . . . . . . . . . . . . . . 8
5.3 Sending periodic feedback . . . . . . . . . . . . . . . . . 8
5.4 Discussions on delay metrics . . . . . . . . . . . . . . . . 9
6. Sender Behavior . . . . . . . . . . . . . . . . . . . . . . . . 9
6.1 Video encoder rate control . . . . . . . . . . . . . . . . . 10
6.2 Rate shaping buffer . . . . . . . . . . . . . . . . . . . . 11
6.3 Reference rate calculator . . . . . . . . . . . . . . . . . 11
6.4 Video target rate and sending rate calculator . . . . . . . 12
6.5 Start-up behavior . . . . . . . . . . . . . . . . . . . . . 13
7. Incremental Deployment . . . . . . . . . . . . . . . . . . . . 13
8. Implementation Status . . . . . . . . . . . . . . . . . . . . . 14
9. IANA Considerations . . . . . . . . . . . . . . . . . . . . . . 14
10. References . . . . . . . . . . . . . . . . . . . . . . . . . . 14
10.1 Normative References . . . . . . . . . . . . . . . . . . . 14
10.2 Informative References . . . . . . . . . . . . . . . . . . 14
Zhu et al. Expires: March 16, 2015 [Page 2]
INTERNET DRAFT NADA September 12, 2014
Authors' Addresses . . . . . . . . . . . . . . . . . . . . . . . . 15
Zhu et al. Expires: March 16, 2015 [Page 3]
INTERNET DRAFT NADA September 12, 2014
1. Introduction
Interactive real-time media applications introduce a unique set of
challenges for congestion control. Unlike TCP, the mechanism used for
real-time media needs to adapt fast to instantaneous bandwidth
changes, accommodate fluctuations in the output of video encoder rate
control, and cause low queuing delay over the network. An ideal
scheme should also make effective use of all types of congestion
signals, including packet losses, queuing delay, and explicit
congestion notification (ECN) markings.
Based on the above considerations, we present a scheme named network-
assisted dynamic adaptation (NADA). The proposed design benefits from
explicit congestion control signals (e.g., ECN markings) from the
network, and remains compatible in the presence of implicit signals
(delay or loss) only. In addition, it supports weighted bandwidth
sharing among competing video flows.
This documentation describes the overall system architecture,
recommended designs at the sender and receiver, as well as expected
network nodes operations. The signaling mechanism consists of
standard RTP timestamp [RFC3550] and standard RTCP feedback reports.
2. Terminology
The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT",
"SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this
document are to be interpreted as described in RFC 2119 [RFC2119].
3. System Model
The system consists of the following elements:
* Incoming media stream, in the form of consecutive raw video
frames and audio samples;
* Media encoder with rate control capabilities. It takes the
incoming media stream and encodes it to an RTP stream at a
target bit rate R_v. Note that the actual output rate from the
encoder R_o may fluctuate randomly around the target R_v. Also,
the encoder can only change its rate at rather coarse time
intervals, e.g., once every 0.5 seconds.
* RTP sender, responsible for calculating the target bit rate
R_n based on network congestion signals (delay or ECN marking
reports from the receiver), and for regulating the actual
sending rate R_s accordingly. A rate shaping buffer is employed
Zhu et al. Expires: March 16, 2015 [Page 4]
INTERNET DRAFT NADA September 12, 2014
to absorb the instantaneous difference between video encoder
output rate R_v and sending rate R_s. The buffer size L_s,
together with R_n, influences the calculation of actual sending
rate R_s and video encoder target rate R_v. The RTP sender also
generates RTP timestamp in outgoing packets.
* RTP receiver, responsible for measuring and estimating end-to-
end delay based on sender RTP timestamp. In the presence of
packet losses and ECN markings, it also keeps track of packet
loss and ECN marking ratios. It calculates the equivalent delay
x_n that accounts for queuing delay, ECN marking, and packet
losses, as well as the derivative (i.e., slope of change) of
this congestion signal as x'_n. The receiver feeds both
information (x_n and x'_n) back to the sender via periodic RTCP
reports.
* Network node, with several modes of operation. The system can
work with the default behavior of a simple drop tail queue. It
can also benefit from advanced AQM features such as RED-based
ECN marking, and PCN marking using a token bucket algorithm.
In the following, we will elaborate on the respective operations at the
network node, the receiver, and the sender.
4. Network Node Operations
We consider three variations of queue management behavior at the network
node, leading to either implicit or explicit congestion signals.
4.1 Default behavior of drop tail
In conventional network with drop tail or RED queues, congestion is
inferred from the estimation of end-to-end delay and/or packet losses.
Packet drops at the queue are detected at the receiver, and contributes
to the calculation of the equivalent delay x_n. No special action is
required at network node.
4.2 ECN marking
In this mode, the network node randomly marks the ECN field in the IP
packet header following the Random Early Detection (RED) algorithm
[RFC2309]. Calculation of the marking probability involves the following
steps:
Zhu et al. Expires: March 16, 2015 [Page 5]
INTERNET DRAFT NADA September 12, 2014
* upon packet arrival, update smoothed queue size q_avg as:
q_avg = alpha*q + (1-alpha)*q_avg.
The smoothing parameter alpha is a value between 0 and 1. A value of
alpha=1 corresponds to performing no smoothing at all.
* calculate marking probability p as:
p = 0, if q < q_lo;
q_avg - q_lo
p = p_max*--------------, if q_lo <= q < q_hi;
q_hi - q_lo
p = 1, if q >= q_hi.
Here, q_lo and q_hi corresponds to the low and high thresholds of queue
occupancy. The maximum parking probability is p_max.
The ECN markings events will contribute to the calculation of an
equivalent delay x_n at the receiver. No changes are required at the
sender.
4.3 PCN marking
As a more advanced feature, we also envision network nodes which support
PCN marking based on virtual queues. In such a case, the marking
probability of the ECN bit in the IP packet header is calculated as
follows:
* upon packet arrival, meter packet against token bucket (r,b);
* update token level b_tk;
* calculate the marking probability as:
p = 0, if b-b_tk < b_lo;
b-b_tk-b_lo
p = p_max* --------------, if b_lo<= b-b_tk <b_hi;
b_hi-b_lo
p = 1, if b-b_tk>=b_hi.
Here, the token bucket lower and upper limits are denoted by b_lo and
b_hi, respectively. The parameter b indicates the size of the token
bucket. The parameter r is chosen as r=gamma*C, where gamma<1 is the
Zhu et al. Expires: March 16, 2015 [Page 6]
INTERNET DRAFT NADA September 12, 2014
target utilization ratio and C designates link capacity. The maximum
marking probability is p_max.
The ECN markings events will contribute to the calculation of an
equivalent delay x_n at the receiver. No changes are required at the
sender. The virtual queuing mechanism from the PCN marking algorithm
will lead to additional benefits such as zero standing queues.
4.4 Comments and Discussions
In all three flavors described above, the network queue operates with
the simple first-in-first-out (FIFO) principle. There is no need to
maintain per-flow state. Such a simple design ensures that the system
can scale easily with large number of video flows and high link
capacity.
The sender behavior stays the same in the presence of all types of
congestion signals: delay, loss, ECN marking due to either RED/ECN or
PCN algorithms. This unified approach allows a graceful transition of
the scheme as the level of congestion in the network shifts dynamically
between different regimes.
5. Receiver Behavior
The receiver periodically monitors end-to-end per-packet statistics in
terms of delay, loss, and/or ECN marking ratios. It then aggregates all
forms of congestion signals in terms of an equivalent delay, and
periodically reports back to the sender.
5.1 Monitoring per-packet statistics
Upon receipt of each packet, the receiver calculates one-way delay as
the difference between sender and receiver timestamps:
x_n = t_r,n - t_s,n.
It also maintains its estimate of baseline delay, d_f, as the minimum
value of previously observed x_n's over a relatively longer period. This
assumes that that sending and receiving clocks are either well-
synchronized, or have a relatively stable offset. In our implementation,
the baseline delay estimation is updated once every 10 minutes.
Correspondingly, the queuing delay experienced by the packet n is
estimated as:
d_n = x_n - d_f.
Zhu et al. Expires: March 16, 2015 [Page 7]
INTERNET DRAFT NADA September 12, 2014
In addition, the receiver keeps track of both packet loss ratios as p_L
via detection of gaps in the packet sequence numbers, and ECN marking
ratios as p_M.
5.2 Aggregating congestion signals
The receiver aggregates all three forms of congestion signal in terms of
an equivalent delay:
x_n = d_n + p_M d_M + p_L d_L, (1)
where d_M is a prescribed fictitious delay value associated with ECN
markings (e.g., d_M = 200 ms), and d_L is a prescribed fictitious delay
value associated with packet losses (e.g., d_L = 1 second). By
introducing a large fictitious delay penalty for ECN marking and packet
losses, the proposed scheme leads to low end-to-end actual delays in the
presence of such events.
While the value of d_M and d_L are fixed and predetermined in our
current design, we also plan to investigate a scheme for automatically
tuning these values based on desired bandwidth sharing behavior in the
presence of other competing loss-based flows (e.g., loss-based TCP).
It should also be noted that in the absence of loss and marking
information, the value of x_n falls back to the observed queuing delay
d_n for packet n. Our algorithm operates in purely delay-based mode.
5.3 Sending periodic feedback
Periodically, the receiver sends back the most recent value of x_n in
RTCP messages, to aid the sender in its calculation of target rate. It
also calculates and sends the derivative of x_n as part of the RTCP
message:
x_n - x_(n-k)
x'_n = ---------------. (2)
delta
Here, the interval between current and previous RTCP messages is denoted
as delta, and the corresponding packet indices are n and (n-k),
respectively. Typically, the interval between adjacent RTCP receiver
reports is on the order of sub-seconds (e.g., 100ms).
Zhu et al. Expires: March 16, 2015 [Page 8]
INTERNET DRAFT NADA September 12, 2014
The size of acknowledgement packets are typically on the order of tens
of bytes, and are significantly smaller than average video packet sizes.
Therefore, the bandwidth overhead of the receiver acknowledgement stream
is sufficiently low.
5.4 Discussions on delay metrics
Our current design works with relative OWD as the main indication of
congestion. the value of the relative OWD is obtained by maintaining the
minimum value of observed OWD over a longer time horizon and subtract
that out from the observed absolute OWD value. Such an approach cancels
out the fixed clock difference from the sender and receiver clocks, and
has been widely adopted by other delay-based congestion control
approaches such as LEDBAT [RFC6817]. As discussed in [RFC6817], the time
horizon for tracking the minimum OWD needs to be chosen with care: long
enough for an opportunity to observe the minimum OWD with zero queuing
delay along the path, and sufficiently short so as to timely reflect
"true" changes in minimum OWD introduced by route changes and other rare
events.
The potential drawback in relying on relative OWD as the congestion
signal is that when multiple flows share the same bottleneck, the flow
arriving late at the network experiencing a non-empty queue may
mistakenly account the standing queuing delay as part of the fixed path
propagation delay. This will lead to slightly unfair bandwidth sharing
amongst the flows.
Alternatively, one could move the per-packet statistical handling to the
sender instead, and use RTT in lieu of OWD, assuming that per-packet
ACKs are available. The main drawback of this latter approach, on the
other hand, is that the scheme will be confused by congestion in the
reverse direction.
Note that the adoption of either delay metric (relative OWD vs. RTT)
involves no change in the proposed rate adaptation algorithm at the
sender. Therefore, comparing the pros and cons regarding which delay
metric to adopt can be kept as an orthogonal direction of
investigation.
6. Sender Behavior
Figure 1 provides a more detailed view of the NADA sender. Upon receipt
of an RTCP report from the receiver, the NADA sender updates its
calculation of the reference rate R_n. It further adjusts both the
target rate for the live video encoder R_v and the sending rate R_s over
the network based on the updated value of R_n, as well as the size of
the rate shaping buffer.
Zhu et al. Expires: March 16, 2015 [Page 9]
INTERNET DRAFT NADA September 12, 2014
--------------------
| |
| Reference Rate | <---- RTCP report
| Calculator |
| |
--------------------
|
| R_n
|
--------------------------
| |
| |
\ / \ /
-------------------- -----------------
| | | |
| Video Target | | Sending Rate |
| Rate Calculator | | Calculator |
| | | |
-------------------- -----------------
| /|\ /|\ |
R_v| | | |
| ----------------------- |
| | | R_s
------------ |L_s |
| | | |
| | R_o -------------- \|/
| Encoder |----------> | | | | | --------------->
| | | | | | | video packets
------------ --------------
Rate Shaping Buffer
Figure 1 NADA Sender Structure
The following sections describe these modules in further details, and
explain how they interact with each other.
6.1 Video encoder rate control
The video encoder rate control procedure has the following
characteristics:
* Rate changes can happen only at large intervals, on the order of
seconds.
* Given a target rate R_o, the encoder output rate may randomly
fluctuate around it.
Zhu et al. Expires: March 16, 2015 [Page 10]
INTERNET DRAFT NADA September 12, 2014
* The encoder output rate is further constrained by video content
complexity. The range of the final rate output is [R_min, R_max].
Note that it's content-dependent, and may change over time.
Note that operation of the live video encoder is out of the scope of our
design for a congestion control scheme in NADA. Instead, its behavior is
treated as a black box.
6.2 Rate shaping buffer
A rate shaping buffer is employed to absorb any instantaneous mismatch
between encoder rate output R_o and regulated sending rate R_s. The size
of the buffer evolves from time t-tau to time t as:
L_s(t) = max [0, L_s(t-tau)+(R_o-R_s)*tau].
A large rate shaping buffer contributes to higher end-to-end delay,
which may harm the performance of real-time media communications.
Therefore, the sender has a strong incentive to constrain the size of
the shaping buffer. It can either deplete it faster by increasing the
sending rate R_s, or limit its growth by reducing the target rate for
the video encoder rate control R_v.
6.3 Reference rate calculator
The sender initializes the reference rate R_n as R_min. Upon receipt of
a new receiver RTCP reports containing values of x_n and x'_n, it
updates the rate as:
kappa * delta_s
R_n <-- R_n + ---------------- * (theta-(R_n-R_min)*x_hat) (3)
tau_o^2
where
theta = w*(R_max - R_min)*x_ref. (4)
x_hat = x_n + eta*tao_o* x'_n (5)
Zhu et al. Expires: March 16, 2015 [Page 11]
INTERNET DRAFT NADA September 12, 2014
In (3), delta_s refers to the time interval between current and previous
rate updates. Note that delta_s is the same as the RTCP report interval
at the receiver (see delta from (2)) when the backward path is un-
congested.
In (4), R_min and R_max denote the content-dependent rate range the
encoder can produce. The weight of priority level is w. The reference
congestion signal x_ref is chosen so that the maximum rate of R_max can
be achieved when x_hat = w*x_ref.
Proper choice of the scaling parameters eta and kappa in (3) and (5) can
ensure system stability so long as the RTT falls below the upper bound
of tau_o. In our design, tau_o is chosen as 500ms.
The final target rate R_n is clipped within the range of [R_min, R_max].
Note that the sender does not need any explicit knowledge of the
management scheme inside the network. Rather, it reacts to the
aggregation of all forms of congestion indications (delay, loss, and
marking) via the composite congestion signals x_n and x'_n from the
receiver in a coherent manner.
6.4 Video target rate and sending rate calculator
The target rate for the live video encoder is updated based on both the
reference rate R_n and the rate shaping buffer size L_s, as follows:
L_s
R_v = R_n - beta_v * -------. (6)
tau_v
Similarly, the outgoing rate is regulated based on both the reference
rate R_n and the rate shaping buffer size L_s, such that:
L_s
R_s = R_n + beta_s * -------. (7)
tau_v
In (6) and (7), the first term indicates the rate calculated from
network congestion feedback alone. The second term indicates the
influence of the rate shaping buffer. A large rate shaping buffer nudges
the encoder target rate slightly below -- and the sending rate slightly
above -- the reference rate R_n.
Intuitively, the amount of extra rate offset needed to completely drain
Zhu et al. Expires: March 16, 2015 [Page 12]
INTERNET DRAFT NADA September 12, 2014
the rate shaping buffer within the same time frame of encoder rate
adaptation tau_v is given by L_s/tau_v. The scaling parameters beta_v
and beta_s can be tuned to balance between the competing goals of
maintaining a small rate shaping buffer and deviating the system from
the reference rate point.
6.5 Start-up behavior
The rate adaptation algorithm specified by (3)--(5) naturally leads to a
linear rate increase at start-up, when queuing delay stays at zero in
the beginning:
kappa*delta_s
R_n <-- R_n + ----------------* theta (8)
tau_o^2
Given that theta = w*(R_max - R_min)*x_ref, the speed of increase scales
with the value of kappa, weight of priority w, and dynamic range of the
flow (R_max - R_min).
In practice, one may desire a more aggressive ramp-up behavior during
the start-up period, e.g., by doubling the rate upon the receipt of each
new RTCP message which reports on near-zero values of x_n and x'_n.
We note here that design of the start-up behavior can be kept orthogonal
to the design of the steady-state rate adaptation behavior. This topic
is worthy of further investigation separately.
7. Incremental Deployment
One nice property of proposed design is the consistent video end point
behavior irrespective of network node variations. This facilitates
gradual, incremental adoption of the scheme.
To start off with, the proposed encoder congestion control mechanism can
be implemented without any explicit support from the network, and rely
solely on observed one-way delay measurements and packet loss ratios as
implicit congestion signals.
When ECN is enabled at the network nodes with RED-based marking, the
receiver can fold its observations of ECN markings into the calculation
of the equivalent delay. The sender can react to these explicit
congestion signals without any modification.
Ultimately, networks equipped with proactive marking based on token
bucket level metering can reap the additional benefits of zero standing
queues and lower end-to-end delay and work seamlessly with existing
senders and receivers.
Zhu et al. Expires: March 16, 2015 [Page 13]
INTERNET DRAFT NADA September 12, 2014
8. Implementation Status
The proposed NADA scheme has been implemented in the ns-2 simulation
platform [ns2]. Extensive simulation evaluations of an earlier version
of the draft are documented in [Zhu-PV13]. Evaluation results of current
draft over several test cases in [I-D.draft-sarker-rmcat-eval-test] have
been presented at the recent IETF meeting [IETF-90].
The scheme has also been implemented in Linux and has been evaluated in
a lab setting also described in [IETF-90]. Evaluation results of NADA in
single-flow and multi-flow scenarios from this testbed will be disclosed
soon.
9. IANA Considerations
There are no actions for IANA.
10. References
10.1 Normative References
[RFC2119] Bradner, S., "Key words for use in RFCs to Indicate
Requirement Levels", BCP 14, RFC 2119, March 1997.
[RFC3550] Schulzrinne, H., Casner, S., Frederick, R., and V.
Jacobson, "RTP: A Transport Protocol for Real-Time
Applications", STD 64, RFC 3550, July 2003.
10.2 Informative References
[RFC3168] Ramakrishnan, K., Floyd, S., and D. Black, "The Addition
of Explicit Congestion Notification (ECN) to IP",
RFC 3168, September 2001.
[RFC2309] Braden, B., Clark, D., Crowcroft, J., Davie, B., Deering,
S., Estrin, D., Floyd, S., Jacobson, V., Minshall, G.,
Partridge, C., Peterson, L., Ramakrishnan, K., Shenker,
S., Wroclawski, J., and L. Zhang, "Recommendations on
Queue Management and Congestion Avoidance in the
Internet", RFC 2309, April 1998.
[RFC6187] S. Shalunov, G. Hazel, J. Iyengar, and M. Kuehlewind, "Low
Extra Delay Background Transport (LEDBAT)", RFC 6817,
December 2012
[ns2] "The Network Simulator - ns-2", http://www.isi.edu/nsnam/ns/
Zhu et al. Expires: March 16, 2015 [Page 14]
INTERNET DRAFT NADA September 12, 2014
[Zhu-PV13] Zhu, X. and Pan, R., "NADA: A Unified Congestion Control
Scheme for Low-Latency Interactive Video", in Proc. IEEE
International Packet Video Workshop (PV'13). San Jose, CA,
USA. December 2013.
[I-D.draft-sarker-rmcat-eval-test] Sarker, Z., Singh, V., Zhu, X.,
and Ramalho, M., "Test Cases for Evaluating RMCAT
Proposals", draft-sarker-rmcat-eval-test-01 (work in
progress), June 2014.
[IETF-90] Zhu, X. et al., "NADA Update: Algorithm, Implementa6on, and
Test Case Evalua6on Results", presented at IETF 90,
https://tools.ietf.org/agenda/90/slides/slides-90-rmcat-
6.pdf
Authors' Addresses
Xiaoqing Zhu
Cisco Systems,
12515 Research Blvd.,
Austin, TX 78759, USA
Email: xiaoqzhu@cisco.com
Rong Pan
Cisco Systems
510 McCarthy Blvd,
Milpitas, CA 95134, USA
Email: ropan@cisco.com
Michael A. Ramalho
6310 Watercrest Way Unit 203
Lakewood Ranch, FL, 34202, USA
Email: mramalho@cisco.com
Sergio Mena de la Cruz
EPFL, Quartier de l'Innovation
Batiment E
Ecublens, Vaud 1015, Switzerland
Email: semena@cisco.com
Charles Ganzhorn
7900 International Drive
International Plaza, Suite 400
Bloomington, MN 55425, USA
Email: cganzhor@cisco.com
Zhu et al. Expires: March 16, 2015 [Page 15]
INTERNET DRAFT NADA September 12, 2014
Paul E. Jones
7025 Kit Creek Rd.
Research Triangle Park, NC 27709, USA
Email: paulej@packetizer.com
Stefano D'Aronco
EPFL STI IEL LTS4
ELD 220 (Batiment ELD), Station 11
CH-1015 Lausanne, Switzerland
Email: stefano.daronco@epfl.ch
Zhu et al. Expires: March 16, 2015 [Page 16]