Robust Cloud Monitor Placement for Availability Veriﬁcation

Melanie Siebenhaar, Ulrich Lampe, Dieter Schuller and Ralf Steinmetz

Multimedia Communications Lab (KOM), Technische Universit

at Darmstadt, Darmstadt, Germany

Keywords:

Cloud Computing, Service Level Agreements, Veriﬁcation, Monitoring, Placement, Performance, Availability.

Abstract:

While cloud computing provides a high level of ﬂexibility, it also implies a shift of responsibility to the

cloud provider and thus, a loss of control for cloud consumers. Although existing means such as service

level agreements or monitoring solutions offered by cloud providers aim to address this issue, there is still

a low degree of trust on consumer side that cloud providers properly measure compliance against SLAs. A

solution lies in designing reliable means for monitoring cloud-based services from a consumer’s perspective.

We already proposed such a monitoring approach in our former work. However, our experiments revealed that

our approach is sensitive to network impairments. Hence, in the work at hand, we introduce the Robust Cloud

Monitor Placement Problem and present a formal optimization model. Based on the model, we propose an

initial optimization approach, that allows to obtain an exact solution using off-the-shelf algorithms.

1 INTRODUCTION

Cloud computing provides highly conﬁgurable com-

puting resources on-demand and with minimal man-

agement effort via the Internet (Mell and Grance,

2011). Hence, it promises a high level of ﬂexibil-

ity similar to utilities like electricity or water (Buyya

et al., 2009). However, using services from the

cloud implies a shift of responsibility to the cloud

provider and thus, a loss of control for cloud con-

sumers. A negotiation of so-called service level

agreements (SLAs) between a cloud consumer and a

cloud provider addresses this issue by providing qual-

ity guarantees to the cloud consumer and specifying

corresponding penalties for the cloud provider in case

of violating a part of the contract. The quality guar-

antees typically refer to certain quality levels such

as thresholds for performance parameters like avail-

ability. Although a negotiation of SLAs provides a

certain level of control to cloud consumers, this so-

lution does not seem to be sufﬁcient. Still, there

is only a low degree of trust on consumer side that

providers conduct sufﬁcient measurements of perfor-

mance against SLAs (CSA and ISACA, 2012). How-

ever, cloud providers often additionally provide some

means for monitoring to their customers. Neverthe-

less, solely relying on these solutions cannot be per-

ceived as a reliable evidence base for detecting and

documenting SLA violations from a consumer’s per-

spective, particularly in the context that providers of-

ten put the burden of SLA violation reporting on their

customers (Patel et al., 2009).

A solution lies in the design and provision of re-

liable means to enable consumers to verify compli-

ance with SLAs from their perspective. Such an ap-

proach has already been proposed in our former work

in (Siebenhaar et al., 2013). Our approach involves

the placement of monitoring units within both, the

cloud provider’s and consumer’s infrastructure in or-

der to verify the availability of cloud applications.

However, our experiments revealed that our solution

is very sensitive to network impairments. Hence, we

examine the robust placement of monitoring units de-

pending on current network reliability in the work at

hand. This paper introduces the Robust Cloud Mon-

itor Placement Problem (RCMPP) as a new research

problem and presents a formal optimization model.

Furthermore, we propose transformations that can be

applied to this nonlinear, multi-objective optimization

problem in order to obtain an exact solution using off-

the-shelf optimization algorithms.

The remainder of this paper is structured as fol-

lows: Section 2 gives an overview of related ap-

proaches to our work. Section 3 brieﬂy summarizes

our former approach for availability veriﬁcation. In

Section 4, we describe the RCMPP in detail and in-

troduce a formal optimization model. Based on the

formal model, we propose an optimization approach

in Section 5. The paper closes with a summary and

directions for future work in Section 6.

193

Siebenhaar M., Lampe U., Schuller D. and Steinmetz R..

Robust Cloud Monitor Placement for Availability Veriﬁcation.

DOI: 10.5220/0004958101930198

In Proceedings of the 4th International Conference on Cloud Computing and Services Science (CLOSER-2014), pages 193-198

ISBN: 978-989-758-019-2

 2014 SCITEPRESS (Science and Technology Publications, Lda.)

2 RELATED WORK

Only a few approaches have been proposed so far for

monitoring the performance of cloud resources from a

consumer’s perspective. For example, (Sharma et al.,

2013) propose a network monitoring framework that

enables tenants to monitor the connectivity between

their allocated compute nodes. Other related publica-

tions focus on robust placement problems in networks

in general. For example, (Natu and Sethi, 2008) try

to minimize the number of monitor locations while

accounting for a maximum of k node/edge failures,

and (Bin et al., 2011) consider relocatable VMs in the

presence of up to k host failures. In the ﬁeld of Op-

erations Research, the fault-tolerant facility location

problem, ﬁrst studied by (Kamal and Vazirani, 2000),

where each city must at least be connected to a certain

number of distinct facilities, is related to the work at

hand. But none of the related approaches addresses

robust monitor placement by jointly considering cur-

rent network reliability, monitor redundancy, and re-

source as well as location constraints.

3 AVAILABILITY VERIFICATION

OF CLOUD APPLICATIONS

In this section, we give an overview of our former

work for verifying the availability of cloud applica-

tions introduced in (Siebenhaar et al., 2013), which

serves as a foundation for the work at hand. Our for-

mer work proposed a hybrid monitoring approach for

verifying the availability of cloud applications from a

consumer’s perspective, since availability is one of the

very few performance parameters that are part of the

SLAs of today’s cloud providers (e.g., Elastic Com-

pute Cloud (EC2) by (Amazon, 2008)). However, fur-

ther performance parameters can be easily incorpo-

rated in our approach as well. Basically, our former

approach combines consumer- and cloud-side moni-

toring. In doing so, consumers not only obtain means

to assess the status of a cloud application indepen-

dently from a cloud provider, but also obtain visibil-

ity of the end-to-end performance of a cloud-based

service. Hence, our approach enables consumers to

attribute downtimes to causes either on consumer or

provider side. Three different component types are

considered in our design (cf. Figure 1): Monitoring

units placed on VMs on consumer- and cloud-side ob-

serve the availability of a cloud application and the

VM, where the application is running on. For our

approach we make the assumption that a consumer

has access to the VM where a cloud application is

hosted. However, a monitoring unit on cloud-side ob-

cloud applications

VM 1

VM 2

Consumer-side

Monitor A

Consumer-side

Monitor B

App

VM Monitor

Cloud-side

Monitor

Broker

push to pull from

Figure 1: Overview of the monitoring framework.

serving a speciﬁc application is never placed on the

VM where the application is hosted. In this case,

a monitoring unit would not be able to report any

downtimes if the underlying VM crashes. Neverthe-

less, a lightweight software component, a VM moni-

tor, is placed on each VM where a cloud application

is hosted in order to gain access to predeﬁned pro-

cesses that are essential for running the cloud applica-

tion. A periodical pull model is applied by the mon-

itoring units in order to invoke predeﬁned services of

the cloud application or request the status of essential

processes from VM monitors. The set of monitored

services and essential processes of a cloud application

can be deﬁned in advance, e.g., as part of the SLAs.

Finally, a broker component is introduced which col-

lects and aggregates the monitoring data of all mon-

itoring units. The monitoring units apply an event-

based push model to send data to the broker. The bro-

ker maintains a list of the downtimes reported by each

monitoring unit and periodically computes the overall

availability. For more information on the computation

of the overall availability, the interested reader is re-

ferred to (Siebenhaar et al., 2013). Since failures in

the network between a consumer and provider could

prevent monitoring units on consumer-side from de-

tecting a downtime of a cloud application, monitoring

units should be placed on different network domains.

We assume that an enterprise running several data

centers at different branches could assign independent

consumer-side monitors to different locations.

4 ROBUST CLOUD MONITOR

PLACEMENT PROBLEM

4.1 Problem Statement

Our work focuses on a scenario, where an enterprise

cloud user has several regional branches worldwide,

each running a private data center, and makes use of

applications running in the data centers of a cloud

CLOSER2014-4thInternationalConferenceonCloudComputingandServicesScience

194

Internet

Customer

Network

Chicago

Provider

Network

Los Angeles

Provider

Network

New York

Customer

Network

New York

Customer

Network

Miami

Figure 2: Robust Cloud Monitor Placement scenario.

provider (cf. Figure 2). We take the perspective of the

enterprise cloud user, who wants to verify the avail-

ability of the running cloud applications. For this

purpose, the enterprise cloud user applies our hybrid

monitoring approach as described in the previous sec-

tion. In doing so, s/he places monitoring units for

each cloud application on consumer- and provider-

side. However, since the experiments in (Siebenhaar

et al., 2013) revealed that our approach is sensitive to

network impairments, the placement of the monitor-

ing units must be conducted in such a way that the ro-

bustness of our monitoring framework is maximized.

Table 1 gives an overview of the symbols introduced

in the following and used in the formal model. The

upper part of the table describes the basic entities,

whereas the lower part contains the parameters.

The scenario considered in this work consists of

a set of data center sites, formally denoted as S =

{1, ..., n}. This set S is a union of the set S

{1, ..., d}) of data center sites of the enterprise cloud

user and the set S

= {d +1,...,n}) of data center sites

of the cloud provider. A set V

= {1, ..., i} of VMs is

running on each data center site s ∈ S. All VMs V

are potential candidates for monitor placement. On

the data center sites s

∈ S

of the cloud provider,

each VM v

∈ V

is provided to the enterprise cloud

user, e.g., as part of a virtual private network, run-

ning a set of cloud applications C

= {1, ..., j} to

be monitored. The VMs v ∈ V

representing potential

monitor candidates are interconnected with the VMs

∈ V

of the cloud applications C

via a set of

links L = {l(sv  s

)}. Since the enterprise cloud

user is not aware of the underlying network topolo-

gies of the Internet service provider and the cloud

provider when connecting to a cloud application, we

assume that the enterprise cloud user is only able to

measure the end-to-end performance between a given

pair of VMs represented by a single link l ∈ L be-

tween this pair of VMs in the model. Each moni-

Based on http://docs.aws.amazon.com/AmazonVPC/

latest/NetworkAdminGuide/images/Branch Ofﬁces.png

Table 1: Used symbols in the formal model.

Symbol Description

S = {1, ..., n} set of n data center sites

= {1, ..., d}) consumer sites, S

⊂ S

= {d + 1, ..., n}) provider sites, S

⊂ S

= {1, ..., i} VM candidates for monitor

placement on site s ∈ S

= {1, ..., j} cloud applications to moni-

tor on VM v

∈ V

, s

∈ S

L = {l(sv  s

)} links interconnecting VM

monitor candidates V

and

VMs of applications C

R = {1, ..., k} set of k considered VM re-

source types

∈ R

resource demand for moni-

toring application c ∈ C

for resource r ∈ R

svr

∈ R

resource supply of VM

v ∈ V

for resource r ∈ R

r f

∈ N

redundancy factor for moni-

toring application c ∈ C

l(svs

)

∈ R

observed reliability for each

link l ∈ L

∈ R

observed reliability for each

VM v ∈ V

toring unit for a cloud application c ∈ C

exhibits

a certain resource demand of rd

∈ R

for a spe-

ciﬁc resource type r ∈ R = {1, ..., k} such as, e.g.,

CPU power or memory. Furthermore, each VM v ∈ V

on a site s ∈ S is able to provide a speciﬁc resource

supply of rs

svr

∈ R

. For reasons of fault-tolerance,

the enterprise cloud user speciﬁes a redundancy fac-

tor r f

∈ N

for each cloud application c ∈ C

indicating that application c has to be monitored by

r f

different monitoring units. We further assume

that the enterprise cloud user is able to utilize tradi-

tional network measurement tools in order to estimate

the reliability of a given VM v ∈ V

on a site s ∈ S or

a link l ∈ L between a given pair of VMs. With re-

spect to these measurements, p

∈ R

denotes the

observed reliability for a VM v ∈ V

on a site s ∈ S

and p

l(svs

)

∈ R

denotes the observed reliability

for a given link l ∈ L.

The challenge for the enterprise cloud user now

consists in assigning the monitoring units of all cloud

applications to VMs on consumer and provider sites.

In doing so, the objective of the enterprise cloud user

is to maximize the reliability of the monitoring frame-

work expressed by the probability that at least one

of the redundant monitoring units for each cloud ap-

plication is working. The resource supplies of all

VMs must not be exceeded by the assigned moni-

toring units. Furthermore, all monitoring units for a

RobustCloudMonitorPlacementforAvailabilityVerification

195

Internet

S’’

S’

Figure 3: Robust Cloud Monitor Placement example.

speciﬁc cloud application must be placed on different

sites and, following our hybrid monitoring approach,

at least one monitor must be placed on consumer and

provider side, respectively. In addition, a monitoring

unit for a speciﬁc cloud application is not allowed to

be placed on the VM where the cloud application is

running. As already mentioned, this is due to the fact

that a monitoring unit is not able to report a downtime

when the underlying VM is currently down.

We denote this problem as the Robust Cloud Mon-

itor Placement Problem (RCMPP). A simpliﬁed ex-

ample for a RCMPP instance neglecting resource de-

mands and supplies for ease of clarity is depicted in

Figure 3. The instance exhibits two provider sites

, S

and three consumer sites S

, S

each running

two VMs. A set of applications is running on each

VM. For example, C

denotes the set of applications

running on VM no. 2 on site S

. The corresponding

link between this VM and VM no. 1 on consumer site

exhibits a reliability of p

l(1,14,2)

= 0.99 and VM

no. 1 exhibits a reliability of p

= 0.97. In our work,

we are only interested in monitoring the cloud appli-

cations on provider side. However, the applications

running on consumer side are also considered in the

model in the form of remaining resource supplies of

their VMs. If we assume that only one application is

part of each set C

on provider side in the example

and that the enterprise cloud user speciﬁes a redun-

dancy factor r f

= 2 for each application, 8 moni-

toring units will be assigned to VMs in total, thereby

placing one monitoring unit on consumer side and one

on provider side for each application.

4.2 Formal Model

In order to develop strategies to solve the RCMPP,

we transform the problem into an optimization model.

The resulting optimization model is depicted in

Model 1 and will be described in the following.

Model 1: Robust Cloud Monitor Placement Problem.

Maximize {p

mon

(x)|s

∈ S

, v

∈ V

, c ∈ C

} (1)

mon

(x) = 1 −

∏

s∈S,v∈V

path

svs

)

svs

(2)

path

svs

= [(1 − p

) + (1 − p

l(svs

)

) (3)

−(1 − p

) (1 − p

l(svs

)

)]

subject to

∑

s∈S,v∈V

svs

= r f

(4)

∀s

∈ S

, v

∈ V

, c ∈ C

, r f

≥ 2

∑

∈S

∈V

,c∈C

svs

≤ rs

svr

(5)

∀s ∈ S, v ∈ V

, r ∈ R

∑

v∈V

svs

≤ 1 (6)

∀s ∈ S, s

∈ S

, v

∈ V

, c ∈ C

∑

s∈S,v∈V

svs

≥ 1 (7)

∀s

∈ S

, v

∈ V

, c ∈ C

, s = {d + 1, ..., n}

∑

s∈S,v∈V

svs

≥ 1 (8)

∀s

∈ S

, v

∈ V

, c ∈ C

, s = {1, ..., d}

svs

= 0 (9)

∀c ∈ C

, s = s

and v = v

svs

∈ {0, 1} (10)

∀s ∈ S, v ∈ V

, s

∈ S

, v

∈ V

, c ∈ C

As already stated, the objective of the RCMPP is

to assign the monitoring units of all cloud applications

to VMs on consumer and provider side so that the re-

sulting placement maximizes the probability that at

least one of the monitoring units for each cloud ap-

plication is working. The objective can be described

by a set of multiple potentially conﬂicting objective

functions that we want to maximize simultaneously

(cf. Equation 1). Each objective function expresses

the probability p

mon

(x) to reliably monitor a speciﬁc

application, i.e., at least one of the redundant mon-

itoring units is working properly. p

mon

(x) equals 1

minus the probability that all monitors for a speciﬁc

cloud application c ∈ C

fail (cf. Equation 2). The

decision variable x

svs

is deﬁned in Equation 10 as

binary and indicates whether VM v ∈ V

running on

CLOSER2014-4thInternationalConferenceonCloudComputingandServicesScience

196

Model 2: Optimization of Complementary Objectives.

Minimize {q

mon

(x)|s

∈ S

, v

∈ V

, c ∈ C

} (11)

mon

(x) =

∏

s∈S,v∈V

path

svs

)

svs

(12)

path

svs

= [(1 − p

) + (1 − p

l(svs

)

) (13)

−(1 − p

) (1 − p

l(svs

)

)]

site s ∈ S is selected to monitor cloud application c ∈

running on VM v

∈ V

on site s

∈ S

. x denotes

the vector of all decision variables x

svs

(cf. Equa-

tion 1). Equation 3 determines the probability q

path

svs

for a monitoring unit of cloud application c ∈ C

placed on VM v ∈ V

on site s ∈ S to fail, thereby

considering the reliability of the VM v ∈ V

and the

reliability of the path interconnecting VM v ∈ V

and

the VM of the cloud application. Equation 4 ensures

that a cloud application c ∈C

is monitored by r f

redundant monitoring units. For each application, at

least two monitoring units must be assigned to VMs,

one on consumer side and one on provider side (cf.

Equations 7 and 8). Equation 5 deﬁnes that each VM

v ∈ V

offers sufﬁcient resource supplies to serve the

assigned monitoring units and Equation 6 makes sure

that all monitoring units for a speciﬁc cloud applica-

tion c ∈ C

are placed on different sites. Finally,

Equation 9 ensures that a monitoring unit for a spe-

ciﬁc cloud application c ∈ C

is not assigned to the

VM v

∈ V

on site s

∈ S

where the cloud application

is also running on.

As can be seen from Model 1, the RCMPP consti-

tutes a multi-objective optimization problem. Hence,

no unique solution of the problem exists. Further-

more, the RCMPP is nonlinear due to Equation 2.

However, the problem can be linearized and trans-

formed into a single-objective optimization problem

as described in the next section.

5 OPTIMIZATION APPROACH

We propose an initial solution approach based on lin-

earization and transformation of the RCMPP into a

single-objective optimization problem. In doing so,

off-the-shelf optimization algorithms, such as branch-

and-bound (Hillier and Liebermann, 2005), can be ap-

plied in order to obtain an exact solution.

For linearization of the RCMPP, we follow a two-

step approach: In the ﬁrst step, we build a new set

of objective functions representing the complemen-

tary objectives of the former Model 1 and turn the

Model 3: Linearization of RCMPP.

log(q

mon

(x)) = log(

∏

s∈S,v∈V

path

svs

)

svs

) (14)

leads to

Minimize {q

log

(x)|s

∈ S

, v

∈ V

, c ∈ C

} (15)

log

(x) =

∑

s∈S,v∈V

svs

log(q

path

svs

) (16)

path

svs

= [(1 − p

) + (1 − p

l(svs

)

) (17)

−(1 − p

) (1 − p

l(svs

)

)]

path

svs

> 0 ∀s ∈ S, v ∈ V

, s

∈ S

, v

∈ V

(18)

maximization problem into a minimization problem

(cf. Model 2). q

mon

(x) in Model 2 determines the

probability that all monitors for a speciﬁc cloud ap-

plication c ∈ C

fail and thus, represents the comple-

mentary probability to p

mon

(x) in the former Model 1.

In the second step, we can now linearize the problem

by taking the logarithm of both sides (cf. Equation

14), an approach that is also followed by (Andreas

and Smith, 2008). Model 3 shows the result of the

linearization. Since no system is without failure, we

assume q

path

svs

> 0. Please note, that the Equations 4

to 10 of Model 1 also belong to Model 2 and Model 3,

but have been neglected due to lack of space. Finally,

we have to transform Model 3 into a single-objective

optimization problem. For this purpose, we apply a

so-called minimax strategy (Jensen and Bard, 2003),

i.e., our transformation is based on a worst-case anal-

ysis, where we aim to minimize the worst possible

outcome. Hence, the former objective of minimiz-

ing the probability q

log

(x) that all monitors fail for

all cloud applications simultaneously can be trans-

formed into minimizing the maximum probability of

all q

log

(x) (cf. Equation 19 in Model 4). In doing so,

the maximum (i.e., worst) probability of all q

log

(x)

can be represented by a new decision variable z ∈ R.

Furthermore, we have to add |C

| new constraints

∀s

∈ S

, v

∈ V

(cf. Equation 20 in Model 4) to the set

of former constraints already introduced in Model 1.

As can be seen from Model 4, the resulting prob-

lem constitutes a mixed-integer linear programming

problem that can be solved using branch-and-bound

(Hillier and Liebermann, 2005).

RobustCloudMonitorPlacementforAvailabilityVerification

197

Model 4: Transformation to Single-Objective RCMPP.

Minimize z (19)

subject to

log

(x) ≤ z (20)

∀s

∈ S

, v

∈ V

, c ∈ C

, z ∈ R

6 SUMMARY AND OUTLOOK

Although cloud-based service delivery is a very ﬂexi-

ble and convenient way for consumers to obtain com-

puting resources, it is attended with a shift of respon-

sibility to the cloud provider and thus, with a loss

of control for consumers. Negotiating SLAs with

providers and using their provisioned monitoring so-

lutions to verify SLA compliance later on is not con-

sidered as sufﬁcient by consumers. Therefore, we

have designed a hybrid monitoring approach for avail-

ability veriﬁcation of cloud applications from a con-

sumer’s perspective in our former work in (Sieben-

haar et al., 2013). However, since our experiments

revealed that our approach is sensitive to network

impairments, we examined the robust placement of

monitoring units in the work at hand. In this paper,

we introduced the Robust Cloud Monitor Placement

Problem (RCMPP) and a corresponding, formal opti-

mization model. Furthermore, we proposed an initial

solution approach based on transformations turning

the nonlinear, multi-objective RCMPP into a mixed-

integer linear programming problem. An exact solu-

tion can then be obtained using the branch-and-bound

optimization algorithm.

In future work, we will implement and evaluate

our proposed optimization approach. Furthermore,

we plan to extend the proposed model to consider

other objectives such as the total monitoring costs.

ACKNOWLEDGEMENTS

This work was supported in part by the Ger-

man Federal Ministry of Education and Research

(BMBF) under grant no. “01|C12S01V” in the con-

text of the Software-Cluster project SINNODIUM

(www.software-cluster.org), E-Finance Lab Frankfurt

am Main e.V. (http://www.eﬁnancelab.com), and the

German Research Foundation (DFG) in the Collabo-

rative Research Center (SFB) 1053 – MAKI. The au-

thors assume responsibility for the content.

REFERENCES

Amazon (2008). Amazon ec2 service level agreement.

http://aws.amazon.com/ec2-sla/, [last access: 3 De-

cember 2012].

Andreas, A. K. and Smith, J. C. (2008). Mathemati-

cal Programming Algorithms for Two-Path Routing

Problems with Reliability Considerations. INFORMS

Journal on Computing, 20(4):553–564.

Bin, E., Biran, O., Boni, O., Hadad, E., Kolodner, E.,

Moatti, Y., and Lorenz, D. (2011). Guaranteeing High

Availability Goals for Virtual Machine Placement. In

31st International Conference on Distributed Com-

puting Systems (ICDCS), pages 700–709.

Buyya, R., Yeo, C. S., Venugopal, S., Broberg, J., and

Brandic, I. (2009). Cloud Computing and Emerging

IT Platforms: Vision, Hype, and Reality for Deliver-

ing Computing as the 5th Utility. Future Generation

Computer Systems, 25(6):599–616.

CSA and ISACA (2012). Cloud Computing Market

Maturity. Study Results. Cloud Security Alliance

and ISACA. http://www.isaca.org/Knowledge-

Center/Research/Documents/2012-Cloud-

Computing-Market-Maturity-Study-Results.pdf,

[last access: 28 November 2012].

Hillier, F. S. and Liebermann, G. J. (2005). Inroduction to

Operations Research. McGraw-Hill, 8th edition.

Jensen, P. A. and Bard, J. F. (2003). Appendix A: Equiv-

alent Linear Programs. In Supplements to Oper-

ations Research Models and Methods. John Wiley

and Sons. http://www.me.utexas.edu/ jensen/ORMM/

supplements/units/lp models/equivalent.pdf [last ac-

cess: 12 January 2014].

Kamal, J. and Vazirani, V. V. (2000). An Approximation

Algorithm for the Fault Tolerant Metric Facility Loca-

tion Problem. In Jansen, K. and Khuller, S., editors,

Approximation Algorithms for Combinatorial Opti-

mization, volume 1913 of Lecture Notes in Computer

Science, pages 177–182. Springer.

Mell, P. and Grance, T. (2011). The NIST Deﬁnition of

Cloud Computing. http://csrc.nist.gov/publications/

nistpubs/800-145/SP800-145.pdf, [Last access: 28

November 2012].

Natu, M. and Sethi, A. S. (2008). Probe Station Placement

for Robust Monitoring of Networks. Journal of Net-

work and Systems Management, 16(4):351–374.

Patel, P., Ranabahu, A., and Sheth, A. (2009). Service Level

Agreement in Cloud Computing. Technical report,

Knoesis Center, Wright State University, USA.

Sharma, P., Chatterjee, S., and Sharma, D. (2013). Cloud-

View: Enabling Tenants to Monitor and Control their

Cloud Instantiations. In 2013 IFIP/IEEE Interna-

tional Symposium on Integrated Network Manage-

ment (IM 2013), pages 443–449.

Siebenhaar, M., Wenge, O., Hans, R., Tercan, H., and

Steinmetz, R. (2013). Verifying the Availability of

Cloud Applications. In Jarke, M. and Helfert, M., edi-

tors, Proceedings of the 3rd International Conference

on Cloud Computing and Services Science (CLOSER

2013), pages 489–494. SciTe Press.

CLOSER2014-4thInternationalConferenceonCloudComputingandServicesScience

198