Computer Science Technical Report Archive

USC Computer Science Technical Reports, no. 928 (2012)

(USC DC Other)

USC Computer Science Technical Reports, no. 928 (2012)

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content Resource Allocation for Network Virtualization
through Users and Network Interaction Analysis
Bo-Chun Wang
∗
, Y.C. Tay
†
, Leana Golubchik
∗
∗
Department of Computer Science, University of Southern California
Email:{bochunwa,leana}@usc.edu
†
Department of Computer Science, National University of Singapore
Email: dcstayyc@nus.edu.sg
Abstract—Network virtualization can potentially overcome In-
ternet ossification. This technology lets multiple virtual networks
run on a shared physical infrastructure. A key step lies in
mapping a virtual network request to a resource allocation in
the network substrate.
Previous approaches to this network embedding problem
assumed the request will ask for specific resources, such as
network bandwidth or computing power. However, the end-user
ismoreinterestedinperformance.Thispaperthereforeconsiders
a different request format, namely a request will ask for a certain
quality of service (QoS). The infrastructure provider must then
determine the resource allocation necessary for this QoS. In
particular, the provider must take into account user reaction
to perceived performance and adjust the allocation dynamically.
To this end, we propose an allocation mechanism that is
based on analyzing the interaction between user behavior and
network performance. This approach includes the use of mea-
surements to dynamically adjust the allocation. Our simulation-
based experiments demonstrate that the proposed approach
can satisfy user performance requirements through appropriate
resource allocation. Moreover, our approach can adjust resource
allocations efficiently, based on collected measurements.
Index Terms—Virtual Network Embedding; Resource Alloca-
tion; Network Virtualization; Traffic Equilibrium; Quality of
Service.
I. INTRODUCTION
Although the Internet has been a tremendous success in
providing a variety of (packet-delivery based) services, the
Internet’s rapid growth and deployment have also become
obstacles to modification of its current architecture and adop-
tion of new protocols. Specifically, due to the coexistence of
multiple ISPs, such modifications require agreement from the
ISPs. However, given their conflicting policies and goals, ISP
agreement is hard to achieve.
To overcome the ossification problem of the Internet, net-
work virtualization is viewed as a potential approach [1]. In
a network virtualization environment, several virtual networks
(VNs) with heterogeneous resources can coexist over a shared
physical infrastructure. Service providers can create their
own VNs by leasing resources from infrastructure providers
and offer customized end-to-end services without significant
modifications to the physical infrastructure [2, 3]. That is,
one advantage of network virtualization is that VNs can have
heterogeneous resources to satisfy a variety of customized
requirements. Virtualization can also provide a heterogeneous
experimental environment for researchers to evaluate new
protocols [4–6].
When service providers generate VN requests, they can
specify resources for every virtual node and link. Thus, when
VNembeddingmapsaVNrequestontospecificphysicalnodes
and links in the substrate network, it must result in satisfaction
of constraints on virtual nodes and links.
Since substrate resources shared by VNs are finite, the VN
embedding problem is one of the fundamental challenges in
network virtualization. That is, an efficient VN embedding is
necessary to increase resource utilization. However, the VN
embedding problem has been proven to be NP-hard (in offline
and online scenarios [7, 8]).
A. Current Approaches
Addressingthisproblemhasbeenanactiveareaofresearch,
with a number of heuristics having been proposed, e.g., [9–
15], where some of the works restrict the problem space in
order to enable efficient heuristics and reduce complexity.
These limitations include: (i) considering offline versions of
the problem and assuming all virtual network requests are
knowninadvance[12,13,15];(ii)assumingthatresourcesare
unlimited without the need of admission control [10, 12, 15];
and (iii) focusing on specific topologies [12].
Fan et al. [10] consider two costs, the occupancy cost
and the reconfiguration cost. Their goal is to find the op-
timal reconfiguration policies that can minimize costs. Lu
et al. [12] focus on a family of backbone-star topologies
and aim at finding the best topology. Szeto et al. [13] use
the multi-commodity flow algorithm to solve the problem.
The papers [9, 11, 14] do not restrict the problem space;
all consider the online version of the problem. These works
consider link and node constraints together with admission
control mechanisms.
Yu et al.[14] propose a two stage mapping algorithm based
on shortest path and multi-commodity flow algorithm, han-
dling the node mapping in the first stage and the link mapping
in the second stage. In addition, they allow path splitting
and migration. Lischka and Karl [11] consider node and link
mapping in one stage. Their approach is faster than the two-
stage approach, especially for large virtual networks with high
resource consumption.
2
Such works assume that VN requests indicate the exact
amount of required resources (e.g., “the capacity of a link
between two nodes should be 10 Mbps”). Then, heuristics are
designed to achieve objectives defined from the perspective of
infrastructure providers, e.g., balancing load [9, 15], maximiz-
ing revenue [9, 11, 14], and minimizing cost [9–12].
However, from the perspective of service providers, the
mainconcernishavingsufficientresourcestosupportacertain
level service quality. Existing efforts are not concerned with
this, i.e., they do not consider what resources are required to
support a needed quality of service (QoS).
B. Our Alternative Formulation
To this end, this paper proposes an alternative approach
- namely that of considering QoS as a constraint. That is,
when VN requests are made, service providers would use
the minimum required QoS as constraints of that request,
rather than focusing on the amount of resources to request.
It is typically not straightforward to determine the amount of
resources that should be requested in order to satisfy a desired
level of QoS. Thus, we reconsider the problem in this light.
There are three roles in virtual networks: infrastructure
providers, service providers, and customers, where each role
has different goals:
• Infrastructure providers typically focus on balancing the
load, maximizing revenue, and minimizing cost.
• Service providers typically focus on maximizing revenue
(e.g., by supporting as many customers or requests as
possible); they are the intermediaries between customers
and infrastructure providers.
• Customers typically focus on quality of service, such as
downloading time (in file downloading applications) or
bit-rate (in video viewing applications).
Previous efforts largely focus on infrastructure providers. Our
work also includes a focus on service providers as well as
customers. Specifically, in contrast to previous efforts, we
consider and focus on the use of QoS as a constraint in the
VN embedding problem.
In this paper, we consider two types of services, web and
short videos (e.g., YouTube [16]). These two types of traffic
correspond to more than 30% of the Internet’s traffic in North
America [17]. As mentioned above, from the perspective of
service providers, revenue maximization is an important goal.
Several metrics can be used to evaluate revenue. For example,
service providers can measure how many customers they can
support or how many requests (connections) can be completed
over a certain time period. In this paper, we mostly use
connection completion rate as our QoS constraint, i.e., when
generating a VN request, service providers will specify a
desired connection completion rate.
Since connection completion rates are affected by network
conditions and user behavior, service providers should also
consider user-perceived QoS when requesting resources for a
specific connection completion rate. The difficulty with using
user-perceived QoS for provisioning is that users react to QoS,
so their demand becomes a moving target. To model this user
reaction, Tay et al. [18] decompose traffic equilibrium into
user demand and network supply and then apply this model
to web traffic. The user curve and network curve intersect to
determine traffic equilibrium.
In this paper, we adopt their model and extend it to support
video traffic by considering different user behavior charac-
teristics. Moreover, we develop a corresponding mechanism
to estimate traffic equilibrium when different link capacities
are given based on collected measurement. The mechanism
allows service providers to determine the amount of resources
needed tosatisfyQoSconstraints(connection completionrate)
efficiently, especially when user behavior and QoS constraints
may change dynamically.
In addition to connection completion rate, we also consider
QoS performance metrics from the perspective of customers
(e.g., bit-rate). We propose an algorithm which can estimate
the amount of resources by considering other QoS constraints
as well as connection completion rate simultaneously.
C. Our Contribution
Our contributions in this paper can be summarized as
follows. Unlike previous efforts, we focus on the use of QoS
as a constraint in the VN embedding problem. We develop
models and corresponding techniques that allow determination
of resource amounts needed to satisfy QoS constraints. We
extend the model proposed in [18] to video traffic (see Sec.II).
In Sec.III, we develop a mechanism which can estimate
connection completion rate efficiently. Moreover, we propose
analgorithmwhichcanestimatetheamountofresourceswhen
considering multiple QoS constraints simultaneously.
In Sec.IV, we demonstrate how our approach can be used
either by service providers or by infrastructure providers. Our
approach offers infrastructure providers insight into efficient
resourcemappinginanetworkvirtualizationenvironment.Our
evaluation results demonstrate that our estimation mechanism
is accurate (see Sec.V). Our proposed approach is orthogonal
to (and can be combined with) existing efforts.
II. TRAFFIC EQUILIBRIUM ANALYSIS MODEL
From the perspective of service providers, we use con-
nection completion rate as the main QoS constraint in this
paper. We adopt the traffic equilibrium analysis model for
web traffic proposed in [18] and extend it to video traffic.
For ease and completeness of presentation, we first provide a
summary of background information needed in Sec.II-A; for
clarity of presentation, several figures are reproduced here (the
corresponding figures in [18] are noted accordingly). Then, we
present our extension of the model in Sec.II-B.
A. Background
Tay et al. consider traffic equilibrium as a balance between
an inflow controlled by users and an outflow controlled by
the network (e.g., link capacity, congestion control, etc) [18].
The number of active connections is controlled by users, and
the network condition affects how fast a connection can be
3
wait-abort wait-complete
wait-state
p
retry
q
retry
p
abort
1-p
abort
r
click
r
session
think
r
supply
p
next
q
next
Fig. 1. Surfing session model (corresponds to Fig.3 in [18]).
wait-state
k downloads
in progress
aborted
downloads
completed
downloads
click
rate of unaborted clicks
rclick
rout
pabortrclick
rin=(1-pabort)rclick
Fig. 2. Flow of downloads
(corresponds to Fig.4 in [18]).
k downloads
in progress
rclick
rout
pabortrclick
rin=(1-pabort)rclick
k downloads
in progress
user curve
network curve
Fig. 3. Two submodels: a user curve and a
network curve (corresponds to Fig.5 in [18]).
completed. Since users will react to congestion, the interaction
between users and the network form a loop:
1) TCP’s control mechanism reduces congestion window
due to network congestion.
2) The reduced congestion window causes downloading
time to increase, so users may generate fewer connec-
tions or abort connections.
3) User reaction reduces the number of active connections
and, consequently, TCP increases transfer rate per con-
nection.
4) The increased downloading rate encourages users to
launch more connections, thus causing congestion to
increase, and looping back to 1 (above).
Tay et al. use a surfing session model for equilibrium analysis,
as depicted in Fig.1; it is assumed that the session arrival rate,
r
session
, is independent of network congestion because users
are unaware of network conditions until they arrive.
In each session, a user generates requests by clicking
hyperlinks, buttons, etc. (Typing in a URL is also regarded
as a click.) r
click
is defined as the click rate. Each click may
launch multiple responses, and the traffic sent to the user is
termedadownload.Forsimplicity,weusethetermsdownload
and connection interchangeably in the following.
After a click, a user waits for completing a download
(wait-state is Fig.1). When the user enters the wait-
state, two different cases are possible, wait-abort state
and wait-complete state, i.e., if the downloading time is
too long due to network congestion, the user may decide to
abort the download. The probability of aborting a download
is p
abort
. After the user aborts a download, it may retry again
with probability p
retry
. Otherwise, it quits the session with
probability q
retry
= 1− p
retry
. If the user completes the
download, it enters the think state (when viewing downloaded
content).Then,itmayclickondifferentURLswithprobability
p
next
or finish the session with probability q
next
= 1−p
next
.
Focusing on the wait-state in Fig.1, k is defined as the
average number of ongoing concurrent downloads in the wait-
state (see Fig.2). This k is a measure of network congestion.
The wait-state is decomposed into a user−network model
that includes a user demand curve r
in
and a network supply
curve r
out
(depicted in Fig.3). The user curve represents
the relationship between the unaborted click rate and the
congestion level k, and the network curve describes the
relationship between the rate of completed downloads and k.
The decomposition gives
r
in
=
(1−p
abort
)r
session
1−p
retry
p
abort
−p
next
(1−p
abort
)
(1)
and
r
out
=
k
p
abort
1−p
abort
T
abort
+
S
completed
bTCP
, (2)
where b
TCP
and S
completed
are the average bandwidth pro-
vided by TCP for a download and the average size of a com-
pleted download, respectively. These two equations describe
a pair of user and network curves that determine the traffic
equilibrium where they intersect (see Fig. 5(a)).
Analyzing how a flash crowd or a failed link affects the
equilibrium thus reduces to examining how the user curve
or network curve is affected; this breaks the feedback look
illustrated earlier for TCP.
B. Extension
The work presented in [18] only considers web traffic.
We now extend the model to short videos. YouTube[16], for
example provides video-sharing services and allows users to
upload videos of up to 15 minutes in length.
In web services, a user aborts a download mostly because
it is slow. In video sharing services, however, a user may
abort a download for other reasons. Gill et al. have studied the
YouTube system and found that approximately 24% of video
downloads were interrupted [19]. They argued that there were
two main reasons: (1) the same as for web services, i.e., poor
performance due to slow downloading rate; (2) poor content
quality(e.g.,videocontentisuninteresting,orvideoresolution
is low), in which case, users aborted connections before the
video ended.
Because user behavior is different in video sharing services,
4
the model described above cannot be applied directly, and
the two cases described above need to be considered. We
define p
rate
as the probability that downloading rate of a
connectionistooslowandp
quality
astheprobabilitythatusers
abort a connection because of poor content quality even if
downloadingrateisfast.Then,p
abort
invideosharingservices
can be calculated as:
p
abort
=p
rate
+(1−p
rate
)p
quality
(3)
This equation gives the user curve a shape that is different
from the one for web traffic (compare Fig. 5(a) and (b)).
We assume p
retry
is the same for both cases (rate and
quality) when users abort a connection. The equation for the
user curve in web services (Eq.1) still works in video sharing
services. The only difference is that p
abort
in video sharing
services is calculated using Eq.3.
We also define T
rate
as the average time users wait before
aborting a connection because of slow downloading rate, and
T
quality
as the average time users wait before aborting a
connection because of poor content quality. By Little’s Law,
the value of k is
k =[p
rate
T
rate
+(1−p
rate
)p
quality
T
quality
+(1−p
abort
)t
completed
]r
click
Then, the equation for r
click
in video sharing services is:
r
click
=
k
prateTrate+(1−prate)p
quality
T
quality
+(1−p
abort
)t
completed
and the equation for the network curve, r
out
, is:
r
out
= (1−p
abort
)r
click
(4)
=
k
prate
1−p
abort
T
rate
+
(1−prate)p
quality
1−p
abort
T
quality
+
S
completed
bTCP
Again, this equation gives the network curve a shape that is
different from that for web services (see Fig. 5).
The above assumes that users will watch the entire video, if
they do not abort a video download. Since the length of more
than 75% of videos in YouTube is within 300 seconds [20],
we believe this assumption is reasonable.
A more important assumption is that users will react to
network congestion. Therefore, the traffic considered in this
paper is not elephantine traffic. It is reasonable to assume
that users will react to network congestion when they are
monitoring progress. However, when users download a large
file, such monitoring may not occur (they may take a break,
do something else, etc).
Tay et al. have studied such non-reactive elephantine con-
nections and found that such traffic may cause a loss of equi-
librium and induce a performance collapse [18]. We therefore
assumethereisaseparatemechanism(e.g.,admissioncontrol)
for dealing with elephantine flows, and focus only on reactive
connections in this paper.
III. PERFORMANCE ESTIMATION MECHANISM
In this paper, we consider service providers as well as
customers.WeuseconnectioncompletionrateasaQoSmetric
from the perspective of service providers. In Section III-A,
we develop a mechanism to estimate the amount of resources
neededtosatisfysuchaQoSconstraint.Then,inSectionIII-B,
we propose an algorithm to estimate the amount of resources
needed when multiple QoS constraints are requested.
A. Connection Completion Rate
To calculate the connection completion rate, service
providers or infrastructure providers would need to know
user behavior characteristics and network conditions. If such
information is not available (e.g., initially), traffic conditions
need to be measured. However, since network equilibrium
changes with time, it is not efficient to measure traffic condi-
tioncontinuously. For example, userarrival rateinthe daytime
is typically higher than at night. Thus, to maintain the same
QoS, infrastructure providers should allocate more bandwidth
inthedaytime.Moreover,serviceprovidersmaymakerequests
for higher QoS.
One possible approach for infrastructure providers to cater
to changing demand of customers and service providers is to
increase bandwidth gradually and then check whether these
assignedresourcescansatisfytheQoSrequirements.Although
straightforward, this method is inefficient and slow. We there-
fore develop a mechanism that can estimate the amount of
necessary bandwidth based on existing measurement data.
We note that each VN can run its own services. Because
user behavior differs from service to service, we assume that
the data used for estimation purposes must be collected from
thesametypeofservice,e.g.,datacollectedfromwebservices
should not be applied to video services. Since, in their basic
construction, the models presented above are similar, we use
the model for web traffic as an example to demonstrate the
proposed mechanism.
Recall that the traffic equilibrium is a balance between an
inflow controlled by users (Eq.1) and an outflow due to the
network (Eq.2). We can classify parameters in Eq.1 and Eq.2
into three categories based on what causes them to change:
• user: p
abort
, r
session
, p
retry
, p
next
, T
abort
, k
• network: b
TCP
, k
• application: S
completed
Because data are collected from the same service, we can
assume that S
completed
and T
abort
are unchanged. Since users
do not know network conditions before they arrive, we can
assume that r
session
remains constant over a time window.
The analysis of collected traces in [21] indicates that p
retry
is
fairly constant and that p
next
can be represented as a function
of p
abort
. Therefore, our proposed mechanism focuses on the
relationship between p
abort
, k, and b
TCP
.
The first step is to determine the relationship betweenk and
b
TCP
. b
TCP
is the average bandwidth provided by TCP for a
download, and k is the number of concurrent connections in
measured data. When k is 1, b
TCP
will be the throughput of
5
0
50
100
150
200
0 10 20 30 40 50 60 70 80
b
TCP
(kbps)
k
(a) b
TCP
vs. k
0
0.001
0.002
0.003
0.004
0.005
0.006
0.007
0.008
0.009
0.01
0 10 20 30 40 50 60 70 80
p
abort
k
(b) p
abort
vs. k
0
0.001
0.002
0.003
0.004
0.005
0.006
0.007
0.008
0.009
0.01
20 40 60 80 100 120 140 160 180
p
abort
b
TCP
(kbps)
(c) p
abort
vs. b
TCP
Fig. 4. Relationship between average TCP connection bandwidth b
TCP
, average number of concurrent downloads k, and p
abort
the connection. Altman et al. [22] and Barakat et al. [23] have
studied TCP throughput in single-hop and multiple-hop paths,
respectively. Wecan calculate TCPthroughput ofaconnection
based on the path type.
However, the TCP throughput calculated by [22] or [23]
is an ideal value. Reaching the ideal value requires a large
enough maximum congestion window size. If the maximum
congestion window size is small, we can estimate TCP
throughput by using
Wm
RTT
, where W
m
is the maximum
congestion window size and RTT is round trip time [24]. We
define thrp
ideal
as the TCP throughput calculated by models
proposed in [22] and [23]. We can then have the following
equation of TCP throughput, thrp:
thrp =min(thrp
ideal
,
W
m
RTT
)
When k increases, the value of b
TCP
is dominated by the
capacity C of the virtual link. We therefore have
b
TCP
=



thrp , if thrp∗k≤C
C
k
, if thrp∗k >C
(5)
Fig.4(a) illustrates Eq.5, using data from ns2 simulations [25].
In this example, C is 2 Mbps, and users abort a connection if
the connection cannot be completed within 20 seconds.
Users will react to network congestion: When b
TCP
de-
creases, p
abort
will increase. It is difficult to determine their
relationship directly. Since Eq. 5 already relates b
TCP
to k,
we can relate p
abort
to b
TCP
through k instead. There are
tools (e.g., [21]) that can extract information from traffic
measurements for expressing p
abort
in terms of k. Fig.4(b)
illustrates such a relationship with simulation data.
Given these relationships (b
TCP
-k and p
abort
-k), we can
determine the value of p
abort
for different b
TCP
. Fig.4(c),
derived from Figures 4(a) and 4(b), is an example illustration
of the relationship between p
abort
and b
TCP
. We can think
of this relationship as representing how users react to network
congestion. When a connection can be completed with a rea-
sonable downloading time, users will not abort the connection.
However, when network congestion becomes worse and b
TCP
becomes small, users may become impatient when waiting for
connection completion. From Fig.4(c), we can see that p
abort
increases dramatically when b
TCP
decreases.
After determining the relationship between b
TCP
and
p
abort
, we can use it to estimate QoS offline, based on existing
measurement data. For example, when service providers ask
for higher QoS, infrastructure providers can use our mech-
anism to estimate the corresponding amount of resources
needed in an “offline manner”. They can increase the capacity
of the virtual link from C to C
′
and calculate the corre-
sponding b
TCP
and p
abort
. They can then plot the user and
network curves, using Eq.1 and Eq.2; the intersection point
(Fig.5) gives the connection completion rate. Next, they can
change thevalue ofC iteratively, untiltheyfind thenewtraffic
equilibrium that satisfies the QoS requirement. Resources can
then be assigned to service providers based on estimated
results. This approach allows infrastructure providers to avoid
having to take more of an “online approach” (in contrast to
the “offline approach” described above), i.e., they can avoid
having to increase resources gradually and measure traffic
continuously.
B. Multiple QoS Constraints
Above, we gave an approach for estimating the amount of
resourcesneededtosatisfyconnectioncompletionraterequire-
ments. However, from the perspective of customers, quality of
service is what typically matters, such as downloading time
(in file downloading applications) or bit-rate (in video viewing
applications). In order to attract customers, service providers
need to consider multiple QoS constraints simultaneously.
As an example, we use short video services and consider
two QoS constraints, connection completion rate and bit-
rate, from the perspective of service provider and customers,
respectively.Wedesignanalgorithmtocalculatetheamountof
resourcesneededtosatisfymultipleQoSconstraints(assuming
a feasible solution exists).
The equation of the network curve for short video, r
out
, is
derived in Eq.4. Initially, we assume there is no user reactions
and all probabilities in Eq.4 are 0. In such a case, the equation
of the network curve is r
out
=
k
S
completed
b
TCP
. Then, we replace
b
TCP
by
C
k
whereC isthe capacity ofa link,and the equation
of the network curve is r
out
=
C
S
completed
.
If the connection completion rate requested by service
providers is r
rate
, the amount of resources needed should be
at least C
min
where C
min
=r
rate
∗S
completed
. Then, we use
Algo.1 to calculate the amount of resources C
required
that is
needed tosatisfymultiple QoS constraints(file throughput and
6
Algorithm 1 Calculation of the minimum amount of
bandwidth needed to satisfy multiple QoS constraints.
1: QoS Requirement 1 (R1): connection completion rate
2: QoS Requirement 2 (R2): bit-rate
3: //R1 and R2 are evaluated by intersecting the
4: //user curve (Eq.1) and network curve (Eq.2); see Fig. 5
5: //assumes there is a feasible solution for bandwidth
6: C
low
←C
min
7: C
high
← 2∗C
min
8: while true do
9: if R1(C
low
) AND R2(C
low
) then
10: return C
low
11: else if R1(C
high
) AND R2(C
high
) then
12: if
C
high
−C
low
C
low
≤δ then
13: return C
high
14: end if
15: C
temp
←
C
high
+C
low
2
16: if R1(C
temp
) AND R2(C
temp
) then
17: C
high
←C
temp
18: else
19: C
low
←C
temp
20: end if
21: else
22: C
low
←C
high
23: C
high
← 2∗C
high
24: end if
25: end while
video bit rate). The idea here is to set bounds C
low
and C
high
for link capacity, and iteratively adjust them to converge on a
value C that satisfies the QoS requirements.
In Algo.1, Eq.1 and Eq.4 are used to plot the user and net-
workcurves;theintersectionpoint(Fig.5)givestheconnection
completion rate to evaluate R1, while the equilibrium k value
determines b
TCP
(Eq.5) for evaluating R2.
IV. VIRTUAL NETWORK EMBEDDING
WepresentedthetrafficequilibriumanalysismodelinSec.II
and proposed our estimation mechanism for the amount of
resources in Sec.III. Our approach can be used by service
providers or by infrastructure providers. In this section, we
explain in detail how service providers and infrastructure
providers can use our mechanism.
A. Resource Estimation at Service Providers
Previous works assume service providers will ask for a
specific amount of resources when they send infrastructure
providers a virtual network request. However, previous works
do not explain how to calculate the amount of resources
needed and do not consider how to satisfy QoS requirements.
Thus, our approach can complement previous efforts.
Specifically, when service providers want to create a new
virtual network, they will define the minimum QoS which
should be achieved. They can then use Algo.1 to calculate
the required bandwidth allocation. When service providers
determine the amount of resources needed, they can then send
infrastructure providers VN requests and ask for that resource
amount. Infrastructure providers can then use previous works
(e.g., [9, 11, 14]) to assign resources.
However, there is a limitation when service providers use
ourmechanism.Inourmechanism,b
TCP
equalsTCPthrough-
put of an isolated connection when the number of concurrent
connections is low (Eq.5). There are several factors that can
affectTCPthroughput,suchasRTTandqueuesizes.However,
since service providers do not know which physical paths will
be assigned, the only information they have is the capacity
of the virtual link. Other network information (e.g., RTT and
queue sizes) is typically unavailable. Therefore, the equation
used by service providers to calculate b
TCP
is
C
k
.
In order to observe whether such a limitation will affect
accuracy of estimation results, we use different physical links
to evaluate our mechanism in Sec.V. We also discuss effects
of network parameters, such as RTT and queue size, in Sec.V.
B. Resource Estimation at Infrastructure Providers
Ourmechanismcanalsobeusedbyinfrastructureproviders.
Here, we argue that service providers can ask for a certain
QoS (e.g. connection completion rate) instead of the amount
of resources (e.g. bandwidth) in virtual network requests. In-
frastructureprovidersneedtoestimatetheamountofresources
which should be assigned to satisfy the QoS specifications.
However, there is a requirement for our estimation mech-
anism: infrastructure providers should have measured p
abort
,
r
session
, p
retry
, p
next
, T
abort
, b
TCP
, k, and S
completed
for
their specific application. If infrastructure providers do not
have such data, they cannot use our estimation mechanism.
To address this issue, we propose a two-step approach.
Suppose the service provider specifies a QoS for connection
completion rate r
rate
, and the infrastructure provider must
determine the capacity allocation C for the virtual link. In the
first step, infrastructure provider first collects data on r
session
,
b
TCP
, k, and S
completed
. Since they do not have data on user
behavior yet, they initially set all probabilities to 0, and use a
lower bound C = r
rate
∗S
completed
, like in Sec.III-B. After
collectingsufficientdataonp
abort
,T
abort
,etc.,thesecondstep
uses the user and network curves to adjust C, like in Algo.1.
The infrastructure providers should also be able to modify
resource assignment dynamically. For example, user arrival
rate in the daytime is typically higher than at night. Thus,
to maintain the same QoS, infrastructure providers should
allocate more bandwidth in the daytime. Another example
is that of service providers making requests for higher QoS.
Our mechanism allows infrastructure providers to efficiently
estimate the amount of bandwidth necessary to satisfy such
requests, based on existing measurement data.
V. EVALUATION
A. Evaluation Environment
We use the GT-ITM tool [26] to generate a substrate
network topology. The GT-ITM tool has been widely used in
research that requires practical network topology generation.
7
The substrate network is configured to have 100 nodes and
around 500 links, a scale that corresponds to a medium-size
ISP. The link bandwidths follow a uniform distribution from
40MBps to 200MBps.
In this paper, we consider two services, web and short
video. For web services, each download transfers thirty 536-
byte packets, and T
abort
is 20 seconds. The settings are the
same as used in [18]. For short video services, the average
file size of a video is 10 MBytes. T
rate
and T
quality
are 6
minutes and 30 seconds, respectively. These settings are based
on measurement results from [19].
We use the ns-2 simulator [25] to evaluate the network
curve. The user curve can be plotted as follows: In [18],
collected traces are analyzed andp
retry
is found to be close to
0.97; use p
abort
measured from the network curve simulation;
do a linear regression fit for p
next
. We give two examples for
each traffic type in Fig.5(a) and Fig.5(b).
InFig.5(a),ther
session
isfixedat0.4,andweusetwokinds
of bandwidth for web traffic, 1MBps and 2MBps. Fig.5(a)
demonstrates that different bandwidths will result in different
connection completion rates when the network traffic reaches
equilibrium. The connection completion rate increases when
the bandwidth increases.
In Fig.5(b), the bandwidth for video traffic is 30MBps.
There are two different r
session
values, 0.02 and 0.03. In this
example, when r
session
increases, the connection completion
rate decreases. Therefore, to satisfy the same QoS require-
ment, infrastructure providers should increase the bandwidth.
Fig.5 demonstrates how network conditions (e.g., band-
width) and user behavior (e.g., r
session
) affect the network
equilibrium.
B. Accuracy of Performance Estimation Mechanism
We presented our performance estimation mechanism in
Sec.III. Here, we evaluate its accuracy. As mentioned above,
measured data is necessary to carry out the estimation mecha-
nism. We collect data from ns2. The bandwidth for web traffic
and for video traffic is 1MBps and 20MBps, respectively. We
use this data as measured data in the estimation mechanism.
First, we fix the value of r
session
and use different connec-
tioncompletionraterequirementstosimulateserviceproviders
changing QoS requirements. For web service, the maximum
r
out
value is about 8, so we randomly pick a QoS value
uniformly between 4 and 80 for the connection completion
rate; for video service, the maximum r
out
value is about 0.25,
and we similarly pick a QoS value between 0.2 and 2.
To satisfy this changing QoS, an infrastructure provider
should be able to allocate bandwidth dynamically (e.g.,
Fig.5(a)). When a connection completion rate is given, we
determine the estimated bandwidth, C
model
, using our mecha-
nism; we use ns2 to determine the actual amount of bandwidth
needed, C
real
. Then, we compare C
model
with C
real
and
calculate the error rate =|
C
real
−C
model
C
real
|. The average error
rate is presented in Fig.6.
We then fix the connection completion rate (QoS) re-
quirement and change the values of r
session
. This scenario
simulates the case where the user arrival rate may change
dynamically. For web service, r
session
is selected uniformly
at random between 0.2 and 2; for video service, r
session
is
similarlychosenfrombetween0.01and0.1.Theinfrastructure
provider has to adjust resource allocation to satisfy the fixed
QoSrequirement despitethechangingr
session
(e.g.,Fig.5(b)).
We calculate the error rate and depict the results in Fig.6.
Fig.6 shows that our mechanism can estimate the amount
of resources needed accurately in both scenarios. In this
evaluation, we only use one set of measured data for each
service. If infrastructure or service providers have multiple
sets of measured data (e.g., data measured from different
bandwidth capacities), they may have better accuracy.
C. Robustness: Effect of Link Delay and Queue Sizes
We demonstrated how service providers can use our mecha-
nism in Section IV-A. However, there was a noted limitation -
since service providers do not know which physical paths will
be assigned, the only information they have is the capacity
of the virtual link; other network information (e.g., RTT
and queue sizes) is typically unavailable. Here, we evaluate
the impact of unknown link delay and queue sizes on our
performance estimation mechanism.
We use web traffic as an example. The default settings are:
bandwidth of 1MBps, link delay of 10ms, and queue size of
50. We increase link delay to 40ms in the first experiment, and
we increase queue size to 100 in the second experiment. Then,
we observe the network equilibrium under different settings.
The results are depicted in Fig.7 where r
session
is 0.4.
Fig.7(a) shows that link delay affects the network curve
when the number of concurrent connections is small, because
b
tcp
is dominated by RTT there (Eq.5); when the number of
concurrent connections is large enough, b
tcp
is controlled by
bandwidthcapacity,butnotRTT.Moreover,theusercurvesare
almost the same and are not affected by link delay. Different
link delays therefore do not affect the network equilibrium
significantly.
Fig.7(b) shows that queue sizes affect the network curve
when the number of concurrent connections is large, because
packets experience longer queuing delays. When network con-
gestion starts, long queuing delays make the situation worse.
This explains why the connection completion rate decreases
more rapidly when large queue sizes are used.
From Fig.7, we can observe that link delay and queue size
donotaffectnetworkequilibriumsignificantly.Thatis,service
providers can still use our mechanism even if they only know
bandwidth capacity.
To evaluate the effect of link delay and queue sizes on
our mechanism, we select a path from our topology and
collect traffic data from this path. We use the measurements to
estimate the amount of resources needed. Then, we randomly
select 50 paths which have different link delays and queue
sizes and assign different QoS requirements to each of them.
We compare the estimated results with measured results for
the 50 paths. The average error rate is less than 10%.
8
2
4
6
8
10
12
14
16
18
20
22
24
0 20 40 60 80 100 120 140 160 180 200
connection completion rate
the number of concurrent connections
network curve(2MBps)
network curve(1MBps)
user curve(2MBps)
user curve(1MBps)
(a) The network curve and the user
curve in web services.
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
20 40 60 80 100 120 140 160 180
connection completion rate
the number of concurrent connections
network curve
user curve(r
session
=0.02)
user curve(r
session
=0.03)
(b) The network curve and the user
curve in video services.
Fig. 5. The connection completion rate at equilibrium is where the user demand (Eq.1) and the
network supply (Eq.2) curves intersect. Notice how the curves differ in shape between (a) and (b).
0
1
2
3
4
5
6
7
changing QoS changing r
session
average error rate (%)
web
video
Fig. 6. The average error rate for changing QoS
and changing r
session
In other words, the results demonstrate the robustness of
our mechanism with respect to uncertainty over link delays
and queue sizes.
VI. DISCUSSION
Long videos: In this paper, we consider two types of
services, web and short videos. We assumed that users will
watch the entire video, if they do not abort a video download.
Since the length of videos in YouTube is short, we believe this
assumption is reasonable. However, this assumption may not
be valid when considering Video-on-Demand (VoD) services
that provide long videos, such as movies.
Yu et al. have studied user behavior in VoD systems [27],
where a typical file is roughly 100 minutes in length. They
found that more than 75% of sessions were terminated within
25 minutes. Most of the users just scanned through videos
rather than watched the entire movie. In such a case, it is
difficult to define what is a completed connection.
In addition, compared to short video sharing systems, user
behavior in such systems is more complicated, e.g., users may
seek forward or backward while viewing a movie [28]. Ex-
tending the traffic equilibrium analysis model to such services
is part of future efforts.
Network or application configurations: In this paper, we
do not focus on any special configurations for specific services
or TCP protocol. However, today, some service providers tune
TCP to improve their performance. For instance, Google mod-
ifies TCP on their routers for better performance. Some appli-
cations may open multiple connections to reduce downloading
time. Moreover, Dobrian et al. find that buffer algorithms in
video services have an impact on user engagement [29]. These
configurations may change network equilibrium by affecting
the network curve, or the user curve. Since our approach
focuses on general cases, such issues are outside the scope
of this paper and are part of future efforts.
Combining with existing efforts: A number of heuristics
have been proposed to address the VN embedding problem.
SuchworksassumethatVNrequestsindicatetheexactamount
of resources required. Our approach can be combined with
theseeffortstoestimatetheamountofrequiredresources.That
is, service providers or infrastructure providers can use our
approach to estimate the amount of required resources first;
then, they can use existing heuristics to assign resources.
In [14], Yu et al. propose two mechanisms to simplify
virtual network embedding: i) split a virtual link over multiple
substrate paths and ii) path migration. These mechanisms have
also been considered in [11] and [9]. In this paper, we only
consider the mapping of virtual links onto physical paths.
However, there is no conflict between our approach and
path splitting and migration mechanisms, and our approach
canalsobeusedinthesemechanisms.InSec.IV,wedescribed
how to use our mechanism at the service provider side and at
the infrastructure provider side. If path splitting and migration
are allowed, our mechanism can be used at the infrastructure
provider side. Service providers should ask for specific QoS
requirements, but not for specific bandwidth. The reason is
that the sum of QoS performance from split paths may not
be equal to the QoS performance of the single path. Because
serviceprovidersdonotknowwhetherinfrastructureproviders
will split paths or not, it is difficult for service providers to ask
for correct amounts of resources. Therefore, service providers
should send infrastructure providers virtual network requests
with QoS requirements instead. The goal of infrastructure
providers would then be to make sure that the sum of QoS
performance of all paths can satisfy these QoS requirements.
VII. CONCLUSIONS
We introduced an alternative direction for VN requests,
namelythatofusingQoSasconstraints.Incontrasttoprevious
efforts that assume a VN request will indicate the amount of
resources needed, we suggested that service providers can use
QoS as constraints when generating VN requests.
We focused on two popular services, web and short videos,
and proposed an allocation mechanism based on analysis of
interaction between user behavior and network performance
in different services. This mechanism can be used by service
providers or infrastructure providers to estimate the amount of
resources needed. It can also dynamically adjust allocations
based on collected measurements.
Our simulation-based experiments demonstrate that the
mechanism can satisfy user performance requirements through
9
1
2
3
4
5
6
7
8
9
10
0 20 40 60 80 100 120 140 160
connection completion rate
the number of concurrent connections
network curve(40ms)
network curve(10ms)
user curve(40ms)
user curve(10ms)
(a) The network and user curves with
different link delay
2
3
4
5
6
7
8
9
10
0 20 40 60 80 100 120 140 160 180 200
connection completion rate
the number of concurrent connections
network curve(100)
network curve(50)
user curve(100)
user curve(50)
(b) The network and user curves with
different queue sizes
Fig. 7. The equilibrium and our estimation mechanism are robust with respect to uncertainty over link delay and queue sizes.
appropriate resource allocation. Moreover, our approach can
adjust resource allocations efficiently, based on collected mea-
surements.
REFERENCES
[1] T. Anderson, L. Peterson, S. Shenker, and J. Turner,
“Overcoming the internet impasse through virtualiza-
tion,” IEEE Computer, April 2005.
[2] J. Turner and D. Taylor, “Diversifying the internet,” in
IEEE GLOBECOM, 2005.
[3] N. Feamster, L. Gao, and J. Rexford, “How to lease the
internet in your spare time,” ACM SIGCOMM Comput.
Commun. Rev., vol. 37, pp. 61–64, Jan. 2007.
[4] A. Bavier, N. Feamster, M. Huang, L. Peterson, and
J. Rexford, “In vini veritas: realistic and controlled
network experimentation,” in ACM SIGCOMM, 2006.
[5] GENI, http://www.geni.net/.
[6] PlanetLab, http://www.planet-lab.org/.
[7] D. G. Andersen, “Theoretical approaches to node assign-
ment,” Dec. 2002, unpublished Manuscript.
[8] S. G. Kolliopoulos and C. Stein, “Improved approxima-
tionalgorithmsforunsplittableflowproblems,”inFOCS,
1997.
[9] N. Chowdhury, M. Rahman, and R. Boutaba, “Virtual
network embedding withcoordinated node and linkmap-
ping,” in IEEE INFOCOM, 2009.
[10] J. Fan and M. H. Ammar, “Dynamic topology configu-
ration in service overlay networks: A study of reconfig-
uration policies,” in IEEE INFOCOM, 2006.
[11] J. Lischka and H. Karl, “A virtual network mapping
algorithm based on subgraph isomorphism detection,” in
VISA, 2009.
[12] J. Lu and J. Turner, “Efficient mapping of virtual net-
works onto a shared substrate,” Washington University.
Technical Report, 2006.
[13] W. Szeto, Y. Iraqi, and R. Boutaba, “A multi-commodity
flow based approach to virtual network resource alloca-
tion,” in IEEE GLOBECOM, 2003.
[14] M. Yu, Y. Yi, J. Rexford, and M. Chiang, “Rethinking
virtual network embedding: substrate support for path
splitting and migration,” SIGCOMM Comput. Commun.
Rev., April 2008.
[15] Y.ZhuandM.H.Ammar,“Algorithmsforassigningsub-
strate network resources to virtual network components,”
in IEEE INFOCOM, 2006.
[16] YouTube, http://www.youtube.com/.
[17] Sandvine Global Internet Phenomena Report - Spring
2011, Sandvine, http://www.sandvine.com/.
[18] Y.C.Tay,D.N.Tran,E.Y.Liu,W.T.Ooi,andR.Morris,
“Equilibrium analysis through separation of user and
network behavior,” Comput. Netw., vol. 52, pp. 3405–
3420, Dec. 2008.
[19] P. Gill, M. Arlitt, Z. Li, and A. Mahanti, “Youtube traffic
characterization: a view from the edge,” in IMC, 2007.
[20] X. Cheng, C. Dale, and J. Liu, “Statistics and social
network of YouTube videos,” in IEEE IWQoS, Jun 2008.
[21] D. N. Tran, W. T. Ooi, and Y. C. Tay, “Sax: A tool for
studying congestion-induced surfer behavior,” in PAM,
2006.
[22] E. Altman, F. Boccara, J. Bolot, P. Nain, P. Brown,
D. Collange, and C. Fenzy, “Analysis of the TCP/IP
flow control in high-speed wide-area networks,” in IEEE
CDC, 1995.
[23] C.BarakatandE.Altman,“AnalysisofTCPwithSeveral
Bottleneck Nodes,” INRIA, Tech. Rep. RR-3620, 1999.
[24] J. Padhye, V. Firoiu, D. F. Towsley, and J. F. Kurose,
“Modeling TCP Reno performance: a simple model and
its empirical validation,” IEEE/ACM Trans. Netw., vol. 8,
pp. 133–145, April 2000.
[25] ns-2, http://isi.edu/nsnam/ns/.
[26] GT-ITM, http://www.cc.gatech.edu/projects/gtitm/.
[27] H. Yu, D. Zheng, B. Y. Zhao, and W. Zheng, “Under-
standing user behavior in large-scale video-on-demand
systems,” in EuroSys, 2006.
[28] B. Cheng, X. Liu, Z. Zhang, and H. Jin, “A measurement
study of a peer-to-peer video-on-demand system,” in
IPTPS, 2007.
[29] F. Dobrian, V. Sekar, A. Awan, I. Stoica, D. Joseph,
A. Ganjam, J. Zhan, and H. Zhang, “Understanding the
impact of video quality on user engagement,” in ACM
SIGCOMM, 2011.

Linked assets

Computer Science Technical Report Archive

Conceptually similar

PDF

USC Computer Science Technical Reports, no. 924 (2012)

PDF

USC Computer Science Technical Reports, no. 919 (2011)

PDF

USC Computer Science Technical Reports, no. 905 (2009)

PDF

USC Computer Science Technical Reports, no. 923 (2012)

PDF

USC Computer Science Technical Reports, no. 917 (2010)

PDF

USC Computer Science Technical Reports, no. 834 (2004)

PDF

USC Computer Science Technical Reports, no. 918 (2010)

PDF

USC Computer Science Technical Reports, no. 894 (2008)

PDF

USC Computer Science Technical Reports, no. 766 (2002)

PDF

USC Computer Science Technical Reports, no. 914 (2010)

PDF

USC Computer Science Technical Reports, no. 969 (2016)

PDF

USC Computer Science Technical Reports, no. 913 (2009)

PDF

USC Computer Science Technical Reports, no. 904 (2009)

PDF

USC Computer Science Technical Reports, no. 906 (2009)

PDF

USC Computer Science Technical Reports, no. 920 (2011)

PDF

USC Computer Science Technical Reports, no. 888 (2007)

PDF

USC Computer Science Technical Reports, no. 891 (2007)

PDF

USC Computer Science Technical Reports, no. 915 (2010)

PDF

USC Computer Science Technical Reports, no. 770 (2002)

PDF

USC Computer Science Technical Reports, no. 899 (2008)

Description

Bo-Chun Wang, Y.C. Tay, and Leana Golubchik. "Resource allocation for network virtualization through users and network interaction analysis." Computer Science Technical Reports (Los Angeles, California, USA: University of Southern California. Department of Computer Science) no. 928 (2012).

Asset Metadata

Creator Golubchik, Leana (author), Tay, Y.C. (author), Wang, Bo-Chun (author)

Core Title USC Computer Science Technical Reports, no. 928 (2012)

Alternative Title Resource allocation for network virtualization through users and network interaction analysis (title)

Publisher Department of Computer Science,USC Viterbi School of Engineering, University of Southern California, 3650 McClintock Avenue, Los Angeles, California, 90089, USA (publisher)

Tag OAI-PMH Harvest

Format 9 pages (extent), technical reports (aat)

Language English

Unique identifier UC16270712

Identifier 12-928 Resource Allocation for Network Virtualization through Users and Network Interaction Analysis (filename)

Legacy Identifier usc-cstr-12-928

Format 9 pages (extent),technical reports (aat)

Rights Department of Computer Science (University of Southern California) and the author(s).

Internet Media Type application/pdf

Source 20180426-rozan-cstechreports-shoaf (batch), Computer Science Technical Report Archive (collection), University of Southern California. Department of Computer Science. Technical Reports (series)

Access Conditions The author(s) retain rights to their work according to U.S. copyright law. Electronic access is being provided by the USC Libraries, but does not grant the reader permission to use the work if the desired use is covered by copyright. It is the author, as rights holder, who must provide use permission if such use is covered by copyright.

Repository Name USC Viterbi School of Engineering Department of Computer Science

Repository Location Department of Computer Science. USC Viterbi School of Engineering. Los Angeles\, CA\, 90089

Repository Email csdept@usc.edu

Inherited Values

Title Computer Science Technical Report Archive

Description Archive of computer science technical reports published by the USC Department of Computer Science from 1991 - 2017.

Coverage Temporal 1991/2017

Repository Email csdept@usc.edu

Repository Name USC Viterbi School of Engineering Department of Computer Science

Repository Location Department of Computer Science. USC Viterbi School of Engineering. Los Angeles\, CA\, 90089

Publisher Department of Computer Science,USC Viterbi School of Engineering, University of Southern California, 3650 McClintock Avenue, Los Angeles, California, 90089, USA (publisher)