Soft Reservations
Uncertainty-aware Resource Reservations in IaaS Environments
Seyed Vahid Mohammadi
1
, Samuel Kounev
2
, Adrian Juan-Verdejo
1
and Bholanathsingh Surajbali
1
1
CAS Software A.G, CAS-Weg 1-5, Karlsruhe, Germany
2
Institute for Program Structures and Data Organization, Karlsruhe Institute of Technology, Karlsruhe, Germany
vahid.mohammadi@cas.de, kounev@kit.edu, {adrian.juan, b.surajbali}@cas.de
Keywords:
Cloud Computing, IaaS, Resource Reservations.
Abstract:
Modern Infrastructure-as-a-Service (IaaS) provides flexible access to data center resources on demand in an
elastic fashion to meet the highly variable workload requirements of cloud applications. Cloud providers aim
to provision resources as efficiently and as quickly as possible to their consumers. However, the lack of infor-
mation about the hosted applications and their workloads makes it hard for cloud providers to anticipate the
future resource demands of their customers so that they can plan the capacity of their infrastructure. Cloud
providers can receive arbitrary requests for allocating resources on-the-fly in a completely unpredictable man-
ner. Given this unpredictability, it may happen that providers might not be able to provision the requested
resources quickly enough, or in the worst case, they might ran out of capacity and may not be able to satisfy
their customers resource demands. To address these concerns, in this paper we propose a new resource reser-
vation mechanism, based on the concept of soft reservations, addressing the issue of uncertainty and lack of
information concerning the expected future customer workloads and corresponding resource demands. The
proposed resource reservation mechanism makes it possible for cloud providers to better plan the capacity
of their infrastructure and continuously optimize the placement of virtual machines on physical nodes thus
improving the infrastructure cost and energy efficiency. It also takes into account the uncertainty of resource
demand estimations and enables proactive online capacity planning resulting in cost benefits for both cloud
providers and cloud customers.
1 INTRODUCTION
Cloud Computing is emerging as a new comput-
ing paradigm providing cloud consumers (henceforth
called consumers) with on-demand access to data cen-
ter resources by integrating computing, storage and
networking platforms in a transparent manner. One of
the major factors for the success of cloud computing
is its elasticity property and the pay-per-use pricing
strategy.
Elasticity is one of the major essential properties
of the cloud paradigm providing the ability to deal
with load variations by automatically provisioning /
deprovisioning resources on-the-fly to match the cur-
rent demand, i.e., adding more resources during high
load periods and consolidating the resources to fewer
nodes when the load decreases.
Ideally, this implies that the amount of resources
such as CPU, main memory and network bandwidth
are assigned and utilized in an optimal manner. For
a system deployed in a pay-as-you-go cloud environ-
ment, such as Infrastructure-as-a-Service (IaaS), elas-
ticity is critical to minimize operating cost while en-
suring acceptable performance during high load peri-
ods. It allows consolidation of the system to consume
less resource and thus minimize the operating costs
during periods of low load while allowing it to dy-
namically scale up as the load increases.
This elastic scaling is typically implemented us-
ing virtualization technology where consumers de-
ploy their applications packaged in virtual machines
(VMs) on a virtualized infrastructure. Each VM hosts
a complete software stack (operating system, middle-
ware, application components) and instances of the
VM can be dynamically added or removed based on
the load variation. This fine-grained allocation is
referred in the literature as on-the-fly elasticity (Vi-
jayakumar et al., 2010).
However, complex workload patterns highly af-
fect elasticity. Indeed, time-varying workload intensi-
ties are already challenging to handle in todays Inter-
net systems. Workloads can vary for multi-tier appli-
223
Mohammadi V., Kounev S., Juan-Verdejo A. and Surajbali B.
Soft ReservationsUncertainty-aware Resource Reservations in IaaS Environments.
DOI: 10.5220/0004775802230229
In Proceedings of the Third International Symposium on Business Modeling and Software Design (BMSD 2013), pages 223-229
ISBN: 978-989-8565-56-3
Copyright
c
2013 by SCITEPRESS Science and Technology Publications, Lda. All rights reserved
cations by orders of magnitude within the same busi-
ness day which makes it hard for the cloud providers
(referred as providers) to optimally allocate VMs.
Through awareness of workload changes, providers
can more effectively overcome this challenge with
less difficulty in the future.
Consumers are in the best position to predict
how their workloads would change over time. How-
ever, the separation of cloud providers (referred as
providers) and consumers hinders the former in hav-
ing access to such information. Providers do not have
direct access to the applications running inside the
hosted VMs. Therefore, they cannot predict the appli-
cations future workload needs and consequently their
future resource demands. Similarly, consumers do not
have access to the hardware infrastructure where VMs
are deployed. Therefore, they cannot anticipate the
effects of sharing resources with third-party applica-
tions deployed by the provider on the same virtualized
infrastructure. Because of this lack of information,
consumers can only specify the type and amount of
resources (e.g., number of CPUs) that should be allo-
cated to their VMs by means of resource reservations
communicated to the provider.
Due to above described information gap, cloud
providers are not in the position to predict the future
resource demands of consumers so that they can plan
the capacity of their infrastructure accordingly. Cloud
providers can receive arbitrary requests for allocat-
ing resources on-the-fly in a completely unpredictable
manner. Given this unpredictability, it may happen
that providers might not be able to provision the re-
quested resources quickly enough, or in the worst
case, they might ran out of capacity and may not be
able to satisfy the consumers resource demands. The
latter may lead to violations of Service Level Agree-
ments (SLA) leading to loss of customers and rep-
utation for both the cloud providers and cloud con-
sumers. As a result, in order to guarantee SLAs, cloud
consumers are forced to reserve more resources than
they actually need resulting in over-provisioning and
associated over-subscribed costs.
We propose a new reservation mechanism in or-
der to protect consumers and providers from the
cost overhead incurred due to over-provisioning. In
this reservation mechanism, consumers can issue pre-
reservations, referred to as soft reservations and then
claim the pre-reserved resources by issuing normal
reservations, referred to as hard reservations, if they
actually end up needing them. Soft reservations cap-
ture the estimated amount of resources that will be
required by a consumer at a given future point of time
as well as the probability of actually needing these
resources. This approach also aims at closing the in-
formation gap between consumers and providers by
supplying a communication mechanism to exchange
the relevant information for both parties.
The proposed approach comprises mechanisms to
exploit the exchanged information in a beneficial way
for both consumers and providers. Consumers would
be able to use low level information about utiliza-
tion of physical resources to better estimate their ac-
tual resource demands for running their services at
the desired Quality of Service (QoS) level. Mean-
while, providers would be able to exploit the infor-
mation about the expected future resource demands
of their consumers to better plan the capacity of their
infrastructure and continuously optimize the mapping
of logical to physical resources resulting in lower data
center operating costs and energy consumption.
The rest of the paper is organized as follows: Sec-
tion 2 describes our resource reservation approach.
We describe types of information we need to ex-
change between consumers and providers in Section
3. In Section 4, we survey the previous works in this
area. Our end-to-end envisioned approach is summa-
rized in Section 5. Finally, we summarize the paper
in Section 6.
2 RESOURCE RESERVATIONS
Cloud providers provide on-demand access to scal-
able computing, storage and networking resources
over a wide-area network. Consumers are able to de-
ploy the VMs required to satisfy their SLAs with their
customers (SaaS end-consumers). Consumers may
dynamically ask for resources by placing resource
reservations with the provider to match their varying
workloads and respective resource demands. Once a
consumer submits a request to the provider the request
will be accepted, provided that enough resources are
available. Otherwise, it would be rejected or some
other options could be offered, i.e., a counter-offer
(Lu et al., 2011) may be made. If the request is ac-
cepted, the provider will need to find a mechanism
to satisfy the request. Consumers take into account
several aspects such as amount, level of granularity,
validity period, certainty, and provisioning intervals
when making reservations. For example, a reserva-
tion could look like:
”I need 10 nodes, each with 1 GB of memory,
right now” or ”I need 4 nodes, each with 2
CPUs and 2GB of memory, from 2pm to 4pm
tomorrow”.
The latter describes the amount of required resources
for a specified time window (from 2 to 4 pm), whereas
the former requests them immediately. Consumers
Third International Symposium on Business Modeling and Software Design
224
Hard Reservation
Soft Reservation
Client 1
Client 2
Client 3
Client 4
Time in Advance of
Sending Soft Reservation
Time
Number of Servers
60%
80%
50%
60%
Figure 1: Snapshot of reservations received in infrastructure.
would normally not be able to predict in advance the
exact amount of resources they will require due to the
uncertainty about their future workloads and resource
demands. However, consumers should normally be
able to indicate how confident they are about the pro-
vided estimation. The latter can be done in differ-
ent ways, e.g., by specifying the probability of actu-
ally requiring the reserved resources or by providing
a cumulative distribution function of the amount of
required resources.
2.1 Hard and Soft Reservations
As mentioned earlier, the actual resource utilization
will only be known at runtime, so it is not possible for
consumers to anticipate the real resource consump-
tion in advance. In traditional resource reservation
mechanisms, consumers have to pay for the requested
resources even if they do not end up using them which
is not aligned with the pay-per-use model in cloud
computing.
Even though consumers may not be able to an-
ticipate their exact resource demands, they would
normally be able to approximately estimate the ex-
pected resource consumption based on workload fore-
casts and performance predictions with some level
of certainty (Herbst et al., 2013). In the envisioned
approach, consumers pre-reserve the forecasted re-
quired resources in long term through soft reserva-
tions. In these reservations, they basically specify
how much resource they will need, in what time span,
and how certain they are about their estimation.
The issuing of a soft reservation does not grant a
consumer the requested resources. Nevertheless, once
a consumer becomes more certain about his resource
demands (near to the point of actual usage) tradi-
tional on-the-fly reservations (henceforth called hard
reservations) can be issued to claim the resources that
were previously pre-reserved through the correspond-
ing soft reservations. If a consumer does not claim the
required resources by means of hard reservations, the
resources will not be allocated.
Figure 1 illustrates a snapshot of hard and soft
reservations received by an IaaS provider and the time
in advance of sending them by consumers over a pe-
riod of time.
Consumers can choose only between using on-
the-fly hard reservations or also taking advantage of
the new soft reservations that allow them to book
in advance resources from provider for a given time
horizon (e.g., minutes, hours) in order to account for
the perceived risk that a workload surge may occur.
Informally, soft reservations will act as a form of in-
surance for consumers about obtaining resources at
lower costs when needed provided that a correct bid
for their future resource demands is communicated in
advance to providers, whereas more expensive hard
reservations (that are not preceded by a previous soft
reservation) will be used to obtain unanticipated ca-
pacity that is required to process the current workload.
Soft reservations cater a win-win solution for both
consumers and providers. For consumers, soft reser-
vations will be much cheaper than hard reservations,
since they only offer the right to obtain a set of fu-
ture resources within a certain amount of time if they
turn out to be required. If these resources are truly
allocated at some point in the future, the consumer
will have to pay additional compensation, but they
will still save money compared to the on-the-fly hard
reservations that would otherwise have to be used as
workloads vary. If not allocated, the soft reservations
would have instead merely served as an insurance pol-
icy for the consumer against high resource provision-
ing costs. However, the pricing model should provide
a policy to avoid oversubscribed unclaimed soft reser-
vations.
Similarly, providers will utilize the information
provided through soft reservations as a basis for on-
Soft Reservations - Uncertainty-aware Resource Reservations in IaaS Environments
225
line capacity planning driving infrastructure manage-
ment decisions. Upon observing changes in hard
reservations, providers would dynamically allocate
new capacity using standard mechanisms such as
provisioning of new servers previously in stand-by
mode. Providers can then use heuristics to optimize
the placement for the new servers taking into account
the currently active hard and soft reservations. Such
heuristic algorithms should prioritize placements that
improve the Total-Cost-of-Ownership (TCO) of the
infrastructure.
Our proposed approach envisions different levels
of soft reservations. Currently, four different dimen-
sions are considered in order to define the softness
level:
Provisioning interval of a soft reservation refers to
the amount of time in which the softly-reserved
resources are guaranteed to be provisioned if they
end up being requested by issuing a respective
hard reservation. The smaller the provisioning in-
terval, the faster resources are guaranteed to be
provisioned if they are claimed.
Validity period represents the validity time frame of
a soft reservation. A reservation for one month
would normally have higher importance and more
implications for capacity planning than a reserva-
tion only for one day.
Time in advance represents the time in advance of
sending a soft reservation before the desired pe-
riod of its validity begins. A soft reservation for
next week would normally have higher priority
and more implications for capacity planning than
a soft reservation with validity period beginning
after one month.
Level of uncertainty refers to the estimated proba-
bility that the reserved resources will actually end
up not being needed.
All four dimensions influence the degree of soft-
ness of soft reservations which in turn would normally
influence the price for placing them. The softer reser-
vations are, the cheaper their price would normally be
expected to be.
Figure 2 illustrates the problems arises without
soft reservations and benefits of the proposed hard and
soft reservation compare to traditional mechanism. In
this example, the provisioning interval and the level of
uncertainty are fixed. The arrows depict the points in
time at which soft reservation are submitted. One is-
sue with the traditional reservation mechanism is the
potential delay between the arrival of hard reserva-
tions and the actual provisioning of the requested re-
sources (dashed red box on the left side of the first di-
agram). Another problem which might occur without
soft reservations is that the provider might not be able
to provision all of the requested resources (dashed red
box on the right side of each diagram). The soft reser-
vations help to address these two issues by enabling
the provider to plan the capacity of the infrastruc-
ture such that all requested resources can be provided
in time. The extent to which the softly reserved re-
sources are guaranteed to be provisioned when plac-
ing a hard reservation depends on the four dimensions
of the softness level explained above. Both exemplary
Figure 2: Example of hard and soft reservations.
problems described above are rooted in the inability
of providers to anticipate what resources will be re-
quired by consumers in the future and thereby plan
their capacity accordingly. The envisioned hard and
soft reservation mechanism will tackle this problem
by making it possible for consumers to communicate
their estimated future resource demands.
3 INFORMATION EXCHANGE
BETWEEN CLOUD
PROVIDERS AND CONSUMERS
As mentioned earlier, the separation of cloud
providers and consumers hinders providers to opti-
mize their placement algorithms for provisioning IaaS
resources. Moreover, the effects of sharing resources
with third-party applications deployed in the same
virtualized infrastructure are hidden from consumers.
Hence, the exchange of information between the two
stakeholders is beneficial for both. Providers can es-
timate the consumers future resource requirements
and plan the allocation of VMs to Physical Machines
(PM) improving the resource provisioning time and
making significant cost savings at both ends. More-
over, information about the actual measured hardware
utilization will allow consumers to only reserve and
pay for the resources that are actually needed and to
run their services at the service level required to fulfil
the end-customers’ (SLAs).
Third International Symposium on Business Modeling and Software Design
226
We identify two types of information which needs
to be exchanged by consumers and providers, namely:
(1) infrastructure monitoring-data supplied by the
provider and (2) SLAs containing resource reserva-
tions placed by consumers.
3.1 Infrastructure Monitoring Data
Consumers suffer from the lack of direct access to the
physical infrastructure level which is necessary to ac-
curately monitor their resource consumption. There-
fore, in order to enable consumers to accurately mon-
itor and predict their future resource usage, providers
must supply information and monitoring data about
the physical infrastructure.
To provide such information, the provider needs to
know the accuracy level and granularity of the mea-
surement data required by each consumer. It is also
important to know for how long historical data about
resource utilization should be stored on the provider
side in order to prevent infinitely growing large log
files.
3.2 SLAs Containing Resource
Reservations
Cloud providers normally do not have direct access to
the applications and services running inside the VMs
deployed on their infrastructure.
To bridge this gap, the proposed resource reserva-
tion mechanism offers a means for consumers to sup-
ply information about their expected future resource
requirements based on workload forecasts and per-
formance predictions. Such information should log-
ically be part of the SLAs established between the
consumers and providers.
The SLAs we are considering would not only
cover classical metrics such as service response time
and throughput, but also provide a powerful protocol
for placing resource reservations, cancelling existing,
or changing them. To realize these consumers need to
know what types of resources are available for reser-
vation and at what level of granularity they can be
reserved (differentiating between general provider of-
ferings, static agreements about the maximum alloca-
tions that could be provided to a consumer, and the
availability of resources for possible reservation at a
particular point in time).
4 STATE-OF-THE-ART
A vast amount of research exists in the literature on
resource reservations in grid computing a summary
of which can be found in (Rani et al., 2011). In
cloud computing, advance reservations are an active
area of research. In (Chaisiri et al., 2009), the au-
thors present a stochastic integer program algorithm
that works in an environment with multiple cloud
providers. They propose an optimal virtual machine
placement (OVMP) algorithm to minimize the total
costs of reservations and on-demand resource provi-
sioning. However, this approach does not consider
any insurance policy for consumers allowing them to
obtain their required resources at a cheaper price.
In (Mark et al., 2011), the authors address this
problem in the same environment (Chaisiri et al.,
2009), but they take different approach for handling
future demands. They apply three different prediction
algorithms (i.e., simple Kalman filter, double expo-
nential smoothing, and Markov prediction) to predict
the demands of customers, i.e., they use past usage
history as a basis for forecasting future demands. In
their approach, resource predictions takes place on the
provider side while in ours, consumers are respon-
sible for estimation of their future resource require-
ments.
Similarly, (Lu et al., 2011) provides a solution
for the resource reservation problem in IaaS providers
with limited resource capacity by which they are able
to realize the feasibility of individual requests from
consumers. If they are not able to satisfy the requests
they will be able to provide an alternative offer by
shifting requests in time (backward and forward) to
fulfil them rather than refusing them. In their solu-
tion fragmentation in virtual resources is controlled
and tried to be avoided. Resource requests are SLA-
based and reservations take place during SLA nego-
tiation. They utilize computational geometry for ad-
vanced reservation of resources.
Haizea
1
is a resource manager (”resource sched-
uler”) software component which allows consumers
to request resources from a computational resource.
Haizea uses leases as a basic resource provisioning
abstraction. A lease is ”a negotiated and renegotiable
agreement between a provider and a consumer, where
the former agrees to make a set of resources available
to the latter, based on a set of lease terms presented by
the resource consumer”. In (Sotomayor et al., 2009),
designers of Haizea, present a model for predicting
various runtime overheads involved in using virtual
machines, which efficiently support advance reserva-
tions. They extend Haizea to use this new model in its
scheduling decisions, and use it with the OpenNebula
virtual infrastructure manager so the scheduling deci-
sions will be enacted in a Xen cluster.
In (Wang et al., 2011), the management of QoS
1
http://haizea.cs.uchicago.edu
Soft Reservations - Uncertainty-aware Resource Reservations in IaaS Environments
227
in the presence of resource reservations in cloud en-
vironments is investigated. In order to guarantee QoS
in the near future and maximize the total revenue of
the resource provider, resource reservation requests
should be accepted selectively. The decision is made
based on the analysis of the possible achieved QoS
after resource configuration.
In (Diaz et al., 2011), three types of resource re-
quests which are rejected because of finite number of
(PM)s and to the variability of VM resources utiliza-
tion is identified. (1) Immediate Rejection (IR): If
there is not enough available capacity in any PM. (2)
Resources Allocation Rejection (RAR): resources are
allocated, and VMs are already hosted in PMs. How-
ever, due to the variability of VM workload, the sum
of resources utilized by VMs hosted in the same PM
can exceed its capacity. Therefore, one or more VMs
must be suppressed to free the resources in PM. (3)
Total of Rejections: It is the sum of IR and RAR ra-
tios. They propose a new concept of Resource Over-
Reservation (ROR) as a mean to reduce RARs. The
basic idea is to pre-reserve additional resources in or-
der to stick at the load variation. The authors found a
trade-off between the IR and RAR as a value for ROR
to keep percentages of total request rejections low.
To summarize, in all of these works traditional re-
source reservations were considered as an input to the
algorithms for placement of VMs and none of them
consider the gap of information between providers
and consumers to address resource reservations in
cloud IaaS environments.
5 APPROACH
Our contribution is new ”soft” resource reservation
mechanism for consumers. As we described in Sec-
tion 2.1, consumers can issue their soft reservations
based on their expected resource demand in long term
with some level of certainty. Typically this is close to
the actual resource consumption point, when they be-
come certain about their resource requirements, they
can claim their softly reserved resources through hard
reservations.
After receiving soft and hard reservations,
providers can use them to continuously improve the
quality of virtual machine placement decisions and
to plan the capacity of their infrastructure. To be
effective, soft reservations should be coupled with a
pricing model that reduces risk and TCO for both
providers and consumers (Rizou and Polyviou, 2012).
We also propose the importance of data shar-
ing between consumers and providers which was de-
scribed in Section 3. This sharing of information will
help both parties to characterize and dynamically re-
act to unexpected changes of usage patterns of ser-
vices and systems.
Figure 3 summarize our end to end envisioned ap-
proach. Consumers automatically adapt the amount
of resources requested from providers based on the
dynamic performance models which will be auto-
matically calibrated at run-time. These predicted re-
sources will be translated to soft and hard reservation
and will be sent to providers. Providers adapt the
mapping of requested logical resources to physical re-
sources in a dynamic way and strictly accounting for
TCO. One possibility to realize the online forecast-
time
time
Model
Calibration
Short Term
Performance
Prediction
Long Term
Performance
Prediction
Cloud Consumer
Short Term
Workload Forecast
Long Term
Workload Forecast
Hard
reservation
Soft
reservation
Usage Data
Sharing
Dynamic Service
Performance Model
ERP
Customer
Portal
Infrastructure
Optimization
TCO Model
VM
VM
VM
VM
VMM
VMM VMM
VM
VM
VM
VM
VM
Cloud Provider
VM Images/
Configuration
Requirements
Figure 3: Online system performance model.
ing and performance predicting process is using the
Descartes Meta Model (Huber et al., 2012). Within
this approach, consumers and providers will individ-
ually work towards achieving better performance and
TCO, thus promoting a more efficient use of data
center resources at reduced cost. Particularly, con-
sumers will automatically adapt the amount of re-
sources requested from providers based on dynamic
performance and TCO models which will be auto-
matically maintained, calibrated, composed and eval-
uated at runtime (Kounev et al., 2011). This will pro-
duce continuously TCO optimization.
6 CONCLUSIONS
In this paper, we introduced our research roadmap
through the definition of soft and hard reservations.
Third International Symposium on Business Modeling and Software Design
228
This mechanism enables consumers to communicate
their workload forecasts in long term to providers
by means of soft reservations. Consumers can claim
the softly reserved resources when they become cer-
tain about their estimated resource requirements. Soft
reservations act as an insurance policy that guaran-
tees that the consumer will receive the softly reserved
resources with cheaper price once they are requested
through hard reservations. Similarly, providers have
to send hardware utilization data (with respect to pri-
vacy of other customers) to the consumer. By means
of these types of reservations, consumers would re-
serve and pay for the amount of resources they
use and will receive their requested resources faster.
Providers would be able to estimate the amount of
resources they should provide in any point in time.
Therefore, they would be able to manage their re-
sources more efficiently by keeping PMs off or turn
off PMs. In our future work, we intend to develop al-
gorithms on the provider side to handle these reserva-
tions. These algorithms will cater for determining the
expected required capacity at a given point of time in
the future. Furthermore, our algorithms will identify
future changes in capacity needs and will optimize the
on-the-fly placement of VMs taking into account the
costs of different reconfiguration options.
ACKNOWLEDGEMENTS
The research presented in this paper has been sup-
ported by the European Union within the FP7 Marie
Curie Initial Training Network ”RELATE”
2
.
REFERENCES
Chaisiri, S., Lee, B.-S., and Niyato, D. (2009). Opti-
mal virtual machine placement across multiple cloud
providers. In Services Computing Conference, 2009.
APSCC 2009. IEEE Asia-Pacific, pages 103 –110.
Diaz, F., Doumith, E., and Gagnaire, M. (2011). Impact
of resource over-reservation (ror) and dropping poli-
cies on cloud resource allocation. In Cloud Comput-
ing Technology and Science (CloudCom), 2011 IEEE
Third International Conference on, pages 470 –476.
Herbst, N. R., Huber, N., Kounev, S., and Amrehn, E.
(2013). Self-Adaptive Workload Classification and
Forecasting for Proactive Resource Provisioning. In
Proceedings of the 4th ACM/SPEC International Con-
ference on Performance Engineering (ICPE 2013),
Prague, Czech Republic, April 21–24.
Huber, N., Brosig, F., and Kounev, S. (2012). Modeling Dy-
namic Virtualized Resource Landscapes. In Proceed-
2
http://www.relate-itn.eu/
ings of the 8th ACM SIGSOFT International Confer-
ence on the Quality of Software Architectures (QoSA
2012), June 25–28, 2012, Bertinoro, Italy. Accep-
tance Rate (Full Paper): 25.6%.
Kounev, S., Brosig, F., and Huber, N. (2011). Self-
Aware QoS Management in Virtualized Infrastruc-
tures (Poster Paper). In 8th International Conference
on Autonomic Computing (ICAC 2011), June 14–18,
2011, Karlsruhe, Germany.
Lu, K., Roblitz, T., Yahyapour, R., Yaqub, E., and
Kotsokalis, C. (2011). Qos-aware sla-based ad-
vanced reservation of infrastructure as a service. In
Cloud Computing Technology and Science (Cloud-
Com), 2011 IEEE Third International Conference on,
pages 288 –295.
Mark, C., Niyato, D., and Chen-Khong, T. (2011). Evo-
lutionary optimal virtual machine placement and de-
mand forecaster for cloud computing. In Advanced In-
formation Networking and Applications (AINA), 2011
IEEE International Conference on, pages 348 –355.
Rani, B., Venkatesan, R., and Ramalakshmi, R. (2011). Re-
source reservation in grid computing environments:
Design issues. In Electronics Computer Technology
(ICECT), 2011 3rd International Conference on, vol-
ume 4, pages 66 –70.
Rizou, S. and Polyviou, A. (2012). Towards value-based
resource provisioning in the Cloud. In 2012 IEEE 4th
International Conference on Cloud Computing Tech-
nology and Science.
Sotomayor, B., Montero, R., Llorente, I., and Foster, I.
(2009). Resource leasing and the art of suspending
virtual machines. In High Performance Computing
and Communications, 2009. HPCC ’09. 11th IEEE In-
ternational Conference on, pages 59 –68.
Vijayakumar, S., Zhu, Q., and Agrawal, G. (2010). Dy-
namic resource provisioning for data streaming appli-
cations in a cloud environment. In Proceedings of
the 2010 IEEE Second International Conference on
Cloud Computing Technology and Science, CLOUD-
COM ’10, pages 441–448. IEEE Computer Society.
Wang, X., Xue, Y., Fan, L., Wang, R., and Du, Z. (2011).
Research on adaptive qos-aware resource reservation
management in cloud service environments. In Ser-
vices Computing Conference (APSCC), 2011 IEEE
Asia-Pacific, pages 147 –152.
Soft Reservations - Uncertainty-aware Resource Reservations in IaaS Environments
229