QoS-Aware Task Allocation and Scheduling in Three-Tier
Cloud-Fog-IoT Architecture Using Double Auction
Nikita Joshi and Sanjay Srivastava
Dhirubhai Ambani Institute of Information and Communication Technology, Gandhinagar, India
Keywords:
Internet of Things, Fog Computing, Task Scheduling, Perishable Resources, Double Auction, Competitive
Bidding, Truthful Bidding.
Abstract:
There is a lot of time-sensitive data provided by IoT applications that needs to be analysed. A Fog-Integrated
Cloud Architecture is available to process these data. Due to the QoS requirements of IoT applications, the
perishable nature of fog cloud resources, and the competition among users and service providers, task alloca-
tion in such an architecture is challenging. In this paper, we propose a competitive bidding (CompBid) strategy
and a QoS-based task allocation and scheduling (QoTAS) algorithm using double auctions that aim to max-
imize user and service provider profit while also satisfying QoS requirements. A remote patient monitoring
system is used to compare QoTAS performance to that of two previous studies, DPDA and MADA. In both
the truthful and CompBid strategies, QoTAS achieves a higher task allocation ratio and resource utilization
than DPDA and MADA. It has 89% more system utility than DPDA and 54% more user utility than MADA.
Furthermore, the CompBid strategy increases QoTAS system utility by 25%.
1 INTRODUCTION
The benefits of various use cases of the Internet of
Things (IoT) have resulted in a significant increment
in IoT adoption. IoT applications generate large
amount of time-critical data, that should be processed
in a timely manner. Internet Database Connector
(IDC) forecast estimated total of 41.6 billion con-
nected IoT devices or “things” will generate 79.4
zettabytes (ZB) of data by 2025 (IoT, 2022). How-
ever, IoT devices cannot process all these data be-
cause of resource and energy constraints. Processing
of this enormous volume of data typically takes place
at remote cloud data centres (Suh et al., 2011).
Cloud servers have sufficient resources to process
IoT data. However, the delay between cloud and IoT
devices may not be acceptable by critical IoT appli-
cations such as healthcare and Vehicular Adhoc Net-
work(VANET). Also, IoT applications generate large
amount of data which requires hundreds of Gbps net-
work bandwidth, which is more than traffic capacity
of WAN in use. IoT applications need context-aware
computing and communication in order to have intel-
ligent interactions. Contextual data can include things
like the user’s location, actions, neighbours’ gadgets,
the time of day, etc. To deal with such IoT appli-
cation requirements, fog computing concept is intro-
duced in which network devices between IoT devices
and cloud server are used for preprocessing, filtering
and compressing the data.
In three tier Cloud-Fog-IoT architecture, IoT Ser-
vice Providers (IoSPs) outsource the task to a Cloud
Service Providers(CSPs) as shown in Figure 1. Fog
Service Providers (FSPs) are registered with CSP.
CSP allocates appropriate FSP to each IoSP. When
allocating resources in such a distributed architecture,
monetary cost must be considered. The cost of com-
puting resources in edge computing, like cloud com-
puting, is determined by their availability and usage
(Jin et al., 2015).
According to J. Weinman, “fogonomics” is a
branch of study that focuses on the economic aspects
impacting the architecture of the fog computing sys-
tem (Weinman, 2017). Fogonomics discusses a va-
riety of pricing techniques, such as static price and
dynamic price. In such a setting, dynamic pricing en-
courages healthy competition among FSPs, which is
advantageous for both IoSPs and FSPs. However, dy-
namic pricing makes it challenging for FSPs to deter-
mine resource prices and for IoSPs to establish bud-
gets dependent on the status of the market (Baranwal
et al., 2018). The challenges of dynamic pricing are
often overcome with the help of many modifications
of the auction dynamic pricing method (Joshi and Sri-
Joshi, N. and Srivastava, S.
QoS-Aware Task Allocation and Scheduling in Three-Tier Cloud-Fog-IoT Architecture Using Double Auction.
DOI: 10.5220/0011967400003488
In Proceedings of the 13th Inter national Conference on Cloud Computing and Services Science (CLOSER 2023), pages 253-260
ISBN: 978-989-758-650-7; ISSN: 2184-5042
Copyright
c
2023 by SCITEPRESS Science and Technology Publications, Lda. Under CC license (CC BY-NC-ND 4.0)
253
Figure 1: Fog IoT architecture.
vastava, 2019).
Money should not be the only consideration when
allocating resources in this three-tiered design. Other
factors to take into account include the QoS needs of
IoT applications and the perishable nature of cloud
and fog resources, which means they cannot be reused
as double resources if not allocated at a time. It is
possible that the FSP bids high in the in an effort to
increase its profit, and no IoSP will be able to match
that bid. It ultimately results in those resources not
being auctioned in that round. These resources are
not eligible for double usage in the following round.
Additionally, if FSP places a low bid, its profit gets af-
fected. As a result, there is a trade-off between profit
and a loss from unsold resources. This problem has
been addressed in our work.
We propose a competitive bidding strategy
(CompBid) in this paper to maximize the benefit of
competition among IoSPs and FSPs. QoS-based Task
Allocation and Scheduling (QoTAS) is proposed to
maximize resource utilization and profit of all mar-
ket agents (IoSPs/FSPs/CSP) while taking perishabil-
ity of resources into account at FSPs. A remote pa-
tient monitoring (Verma and Sood, 2018) application
is used to compare QoTAS performance to that of pre-
vious studies DPDA (Joshi and Srivastava, 2020) and
MADA (Peng et al., 2020).
The structure of this paper is as follows:Section 2
discusses related work. The system model and prob-
lem statement are explained in Section 3. The task
allocation algorithm is explained in Section 4. Sec-
tion 5 discusses the simulation results, and Section 6
discusses the conclusions.
2 RELATED WORK
In a distributed Cloud-Fog-IoT architecture, IoSPs,
FSPs, and CSPs share resources by purchasing them
from one another. Each has its own utility function
based on the cost and payment received. When con-
sidering task allocation using both QoS requirements
and monetary costs, the literature divides this problem
into two subproblems. These are the allocation and
price determination problems. The allocation prob-
lem meets QoS requirements, while the price determi-
nation problem seeks to maximize profit for all mar-
ket agents such as FSP, IoSP and CSP. Literature is
divided into three categories depending on the agents’
bidding strategies: forward auction-based allocation,
reverse auction-based allocation, and double auction-
based allocation.
Forward Auction-Based Allocation: In a forward auc-
tion, buyers submit bids, and the seller allocates re-
sources to the highest paying buyer. (Zu et al., 2019)
proposed an ascending bid auction for task offload-
ing problem in fog computing using KKT conditions.
(Zhang et al., 2019) presented a Parking Vehicle As-
sistance (PVA) in VANET using forward auction.
Forward auction is appropriate when there are many
competing buyers but only one seller or when there is
no competition among sellers. In our scenario, how-
ever, multiple sellers compete with one another to sell
their resources for the highest possible profit. Com-
petitive bidding strategy is proposed in this work.
Reverse Auction-Based Allocation: In a reverse auc-
tion, buyers choose the seller with the lowest asking
price after competing sellers compete to sell their re-
sources. Reverse auction was utilized by (Guo et al.,
2020) to select the target fog nodes and convey the
data from the source to the destination fog nodes.
Minimizing overall energy use was the goal. For
the IoT fog, (Aggarwal et al., 2021) suggest reverse
auction-based resource allocation, which also satisfies
the QoS criteria in IoT applications. This technique
only allows service providers to submit bids. As a re-
sult, the seller is unable to profit from the competition
among buyers.
Double Auction-Based Allocation: Buyers and sell-
ers submit bids simultaneously in a double auction
process, and the auctioneer matches the right buyers
and sellers by employing various methods of price de-
termination. (Joshi and Srivastava, 2020) proposed a
VCG-inspired price determination algorithm. How-
ever, in some cases, VCG may result in negative util-
ity for the broker. CSP serves as a broker in our case.
As a result, using CSP may have a negative impact.
(Peng et al., 2020) has proposed a resource alloca-
tion method that creates a bipartite graph of users and
CLOSER 2023 - 13th International Conference on Cloud Computing and Services Science
254
fog nodes. Edges in the graph are weighted based on
QoS requirements. Allocation is then performed us-
ing maximum matching of a bipartite graph. The ex-
tended McAfee (Yang et al., 2011) method finds the
final price for buyers and sellers. Matching is used
to find allocation, so an FSP can serve only one user,
even if some of its resources can satisfy the needs of
other users. In an actual situation, one FSP can satisfy
more than one IoSP.
All of the studies discussed thus far allocate all of
the resources required by the task in a single round,
resulting in a scarcity of resources in that round,
which can only satisfy a limited number of tasks.
Furthermore, none of them have taken into account
the perishable nature of resources when allocating re-
sources. (Miyashita, 2014) and (Safianowska et al.,
2017) have proposed a bidding strategy for perishable
resource sellers. The allocation algorithm in both of
these works lacks a unique allocation technique for
allocating maximum resources to avoid loss due to
perishability. To avoid loss due to perishability, our
algorithm employs a novel resource allocation tech-
nique.
Furthermore, bidding strategy is an important
component of the auction mechanism. In a double
auction mechanism, the strategy must converge to a
point where both parties make the most profit while
meeting each other’s needs.(Chowdhury et al., 2018)
and (Thavikulwat and Pillutla, 2008) have proposed
a bidding strategy for periodic double auctions where
bids are cleared at regular intervals. It is appropriate
for our architecture. However, in this bidding strat-
egy, they are capable of wasting a few initial rounds
in which the buyer may not receive any resources and
the seller may not sell anything in order to find the
equilibrium point. However, because our resources
are perishable, we must ensure that resources are al-
located from the first round itself. To the best of our
knowledge, no bidding strategy is proposed for the
fog market.
3 PROBLEM FORMULATION
We consider a three-tier architecture,in which there
are N IoSPs in the IoT layer, , M FSPs in the Fog
layer, and one CSP in the Cloud layer. FSPs com-
municate with one another via a mesh network. It is
assumed that IoSPs receive virtualized services from
FSPs (Zhang et al., 2017). As a result, IoSPs can only
connect to FSPs via CSP. Also, time is divided into
rounds. Allocation occurs at the start of each round.
IoSPs and FSPs register with CSP, as shown in Figure
2. CSP sends summary report of the previous round
to all registered IoSPs. When the IoSP requires re-
sources, it sends a request to the CSP. A single IoSP
can send multiple task requests. IoSPs send their bid
for the task, CSP communicates with FSPs and as-
signs the best FSP to IoSP. IoSP sends input data to
FSP, who performs the required task and returns the
results to IoSP. Following task completion, IoSP sub-
mits a QoS report; CSP calculates the final payment
for FSPs and IoSPs.
Figure 2: System model.
Resources are represented as Virtual Machines
(VMs). Each VM has a set of fixed resources such as
memory, network bandwidth, and computing power.
Each FSP provides a varying level of QoS service.
When resources are limited, higher QoS level VMs
are prioritised. It is assumed that the valuation of
VMs becomes zero at the end of each round.
Each FSP registers with the CSP by sending fog
information I
j
= {A
j
, b
f
jy
} where A
j
is the number
of available VMs at j
th
FSP and b
f
jx
is bid for x
th
QoS level. Because CSP stores this information, it
is only necessary to send it once during FSP registra-
tion. If IoSP wants to execute a task, it sends task
requirements to CSP at the beginning of the round.
task requirement of i
th
IoSP for k
th
task contains
Q
i,k
= {λ
i,k
, d
i,k
, D
i,k
, s
i,k
} where λ
i,k
is required num-
ber of VMs, d
i,k
is deadline,D
i,k
is data size, s
i,k
is re-
quired QoS level at fog devices. At the beginning of
each round t CSP sends a summary report (S
t
). Using
this report IoSP calculate its bid b
u
i,k
using competi-
tive bidding strategy(CompBid) proposed in the next
section and send it to CSP.
1. Maximization of IoSPs utility: IoSP wishes to
maximize its utility by adjusting the price paid
to CSP. Constraint (1.1) indicates that the task
must be completed before the deadline. Con-
QoS-Aware Task Allocation and Scheduling in Three-Tier Cloud-Fog-IoT Architecture Using Double Auction
255
straint (1.2) states that each task can only be as-
signed to one FSP, implying that the task cannot
be divided into multiple sub-tasks. (1.3) denotes
that each IoSP pays less than its valuation, result-
ing in a positive utility.
Subproblem-1: Maximization of IoSP utility
argmax
r
u
i,k
N
i=1
U
h
i
(1.0)
Subject to T
i,k
d
i,k
, ik (1.1)
τ
i,k, j
{0, 1} (1.2)
r
u
i,k
< v
i,k
, i (1.3)
2. Maximization of FSP utility: Each FSP seeks to
maximize utility by varying the price and number
of allocated VMs (a
j
). Constraint (2.1) states that
FSP may not allocate more VMs than are avail-
able. Constraint (2.2) states that each FSP must
be priced higher than its maintenance cost in or-
der to be utility positive.
Subproblem-2: Maximization of FSP utility
argmax
r
f
j
,a
j
M
j=1
U
f
j
(2.0)
Subject to a
j
A
j
, j (2.1)
r
f
j
> E
j
, j (2.2)
3. Maximization of CSP utility: CSP wants to max-
imize its utility by matching IoSPs and FSPs and
determining their prices. The requirement in con-
straint (3.1) states that IoSPs’ payments must ex-
ceed the price that must be paid to FSPs. Thus, its
utility is positive.
Subproblem-3: Maximization of CSP utility
argmax
τ
i,k, j
,r
u
i,k
,r
f
j
U
c
(3.0)
Subject to
N
i=1
r
u
i,k
M
j=1
r
f
j
(3.1)
These subproblems are interdependent. IoSPs want
to maximize their utility by lowering the price to be
paid. However, because of this, it may not receive any
resources at all. Furthermore, the utility of CSP and
FSP may decrease as they receive less payment from
IoSP. FSP wishes to maximize its utility by charging
higher fees, which may reduce its total allocated re-
sources as well as IoSP and CSP utility. This is an ex-
ample of a joint optimization problem. It is NP-hard
to maximize the utility of all three nodes at the same
time(Zhang et al., 2017). Furthermore, perishable na-
ture of cloud-fog resources makes this problem more
difficult.
4 PROPOSED SOLUTION
We propose a heuristic-based solution that has two
components namely Competitive bidding (CompBid)
strategy and QoS-based Task Allocation and Schedul-
ing (QoTAS).
4.1 Competitive Bidding (CompBid)
To maximize service providers’ benefits as a result
of competition among IoSPs, an IoSP’s bid must in-
crease in conjunction with an increase in the demand
supply ratio, also known as normalised load. To put
this logic into action, we design a bidding strategy in
which CSP shares a summary report at the start of
each round (t). Which contains S
t
= {n
t1
, c
t1
y
, n
t
}.
n
t
represents normalized load of t
th
round and c
t1
y
is
average VM price of the previous round for y
th
QoS
level. Each IoSP has a summary report published by
CSP as an input, as well as its own aggressiveness
factor (β
i
). Each IoSP determines its own bid for QoS
level y using
b
u
i,k
= min(B
0
+ β
i
n
t1
n
t
c
t1
y
, v
i,k
) (1)
Here, B
0
is the base price, which should be equal to
or greater than the VM’s maintenance cost. β
i
can be
any value between 0 and 1. The greater the value of
β
i
, the more aggressive the IoSP and the higher the
bid.
4.2 QoS-Based Task Allocation and
Scheduling (QoTAS)
We propose a double auction-based algorithm for
QoS-based Task Allocation and Scheduling (Qo-
TAS) in commercial three-tier Architecture. Algo-
rithm 1 depicts the steps of the QoTAS algorithm. In
the first step, we identify FSP that meets QoS require-
ments for each task. The final winner and price are de-
termined in the second step based on the bids of ISPs
and FSPs. Because VMs are perishable resources, in
the third step, if there are tasks with eligible FSPs
based on QoS requirements that did not win due to a
lower bid, we assign VMs of those FSPs to the IoSPs
with a penalty.
Algorithm 1: QoS-based Task Allocation and
Scheduling (QoTAS).
1: Find eligible FSP for each task(Algorithm 2)
2: Price and final winner determination (Algorithm 3)
3: Allocation of remaining VMs (Algorithm 4)
CLOSER 2023 - 13th International Conference on Cloud Computing and Services Science
256
4.2.1 Find Eligible FSP for Each Task
In this algorithm, we try to find an eligible FSP for
each task that has the minimum service delay for
the task. We also schedule VMs based on resource
scarcity on each FSP for load balancing and efficient
VM utilization. In the first step, we identify candi-
date FSPs for each task. Candidate FSP is an FSP
who is capable of completing tasks before the dead-
line. We may receive a more than one candidate FSPs
for each task. Then, using Algorithm 2, we filter them
to find the most eligible FSP for each task. To dis-
tribute VMs with load balancing so that each FSP re-
ceives an equivalent number of tasks. It first computes
each FSP’s resource scarcity using Eq. (2), which is
the ratio of VM demand based on the candidate FSP
list to total available VMs on a specific FSP.
r
s
j
=
N
i=1
X
k=1
w
i, j,k
q
i,k
A
j
(2)
Where w
i, j,k
is one if j
th
FSP is candidate FSP for k
th
task of i
th
IoSP. The minimum penalty p
min
i,k
for a ser-
vice delay of d
i,k+1
is calculated for all tasks. It is a
penalty if this task is completed one unit of time af-
ter its deadline. Following that, all FSPs are sorted
in ascending order of r
s
j
, with the FSPs with the low-
est resource scarcity being allocated first. Tasks, on
the other hand, are prioritised using a penalty system.
The greater the minimum penalty, the more important
that task. As a result, for All tasks, the descending
order of p
min
i,k
is used.
Each task in the sorted list is compared against
each FSP in the sorted list. If w
i, j,k
= 1, task schedul-
ing is done. We provide a task list, fog information,
and candidate FSPs for each task as input to Algo-
rithm 2. For each task, the algorithm returns eligible
FSP.
4.2.2 Price Determination
Following the identification of eligible FSPs for each
task, we must determine the final winners and the final
price to be paid by ISPs, as well as the final payment
to be received by FSPs based on their bid. For double
auctions, there are numerous winner determination
and pricing methods available. We map the require-
ment of our problem statement with terms of auction
theory to find the best method for our scenario; from
that exercise it is concluded that, we need Individ-
ual rationality (IR), Weekly budget baance (WBB),
Truthull (TF) and Allocative efficient (AE) pricing
mechanism. We compared k-DA, VCG, Trade reduc-
tion, McAfee and extended McAfee techniques and
concluded that extended McAfee (Peng et al., 2020)
is most suitable pricing mechanism for our model.
Algorithm 2: Find eligible FSP.
Input: Task list, Fog info(I), Candidate list
Output: Eligible Task-FSP pair
1: for Each FSP in Fog info do
2: Calculate r
s
j
using Eq. (2)
3: Sort FSPs in ascending order of r
s
j
4: for Each Task do
5: Calculate p
min
i,k
6: Sort Task in descending order of p
min
i,k
7: for Each Task in sorted Task list do
8: for Each FSP in sorted Fog info do
9: if FSP is candidate FSP for Task then
10: if Deadline of task is one round then
11: Allocate all required VMs in one round
12: else if Deadline(d
i,k
) is less than r
s
j
then
13: Allocate VMs equally till d
i,k
rounds
14: else
15: Allocate VMs equally till r
s
j
rounds
However, we have a bid of IoSPs for different QoS
levels so we can’t not allocate them altogether. Algo-
rithm 3 shows solution to handle it.
Algorithm 3: Winner and price determination.
1: Create a group of task requests for each QoS level
2: We have the price of all FSPs for each QoS level
3: Apply extended McAfee for each group and find
winners and price
4.2.3 Allocation of Remaining VMs
We may not keep VMs idle if IoSPs’ bid is less than
the required cost since they are perishable. However,
if we allocate the same VMs at a lower price, users
may always bid lower to maximize their utility. In re-
sponse to this situation, we propose algorithm 4 where
we are allocating lesser VMs than demanded and QoS
satisaction is not guaranteed.
Algorithm 4: Allocation of remaining VMs.
Input: Task list, Fog info, Eligible Task FSP pair,
Winner pairs
Output: Updated winner pairs
1: for Each Task in Task list do
2: for Each FSP in Fog info do
3: if FSP is eligible for Task but not winner then
4: Declare that Task FSP pair as winner
5: Price to be paid is equal to bid of IoSP
6: Allocate γ percentage of required VMs
with QoS level one
QoS-Aware Task Allocation and Scheduling in Three-Tier Cloud-Fog-IoT Architecture Using Double Auction
257
5 SIMULATION AND RESULTS
A remote patient monitoring case study is used to
compare the performance of QoTAS in truthful bid-
ding and CompBid strategy with Deadline and Prior-
ity enabled Double Auction (DPDA) (Joshi and Sri-
vastava, 2020) and Multi Attribute Double Auction
(MADA) (Peng et al., 2020).
5.1 Simulation Setup
There are four types of tasks in remote patient mon-
itoring. Namely, client module task, pre-processing
task, abnormality task, and prediction task. Out of
that client module task, pre-processing task, abnor-
mality task are processed on fog nodes (Mahmud
et al., 2022). This task request can be generated by a
variety of patients. The monitoring period and dead-
line for processing collected data differs depending
on the criticality of the patient (Kumar et al., 2008).
We assume that each patient’s monitoring period, also
known as task generation period, is a multiple of the
auction period. As a result, at the start of each auction
round, requests from all patients are received.
We implement a scenario in which there are 30
patients, 6 FSPs and one CSP. We consider two tasks
here, preprocessing and abnormality detection, be-
cause they are expected to be performed on fog.
The resource requirements for both of these tasks are
listed in (Mahmud et al., 2022). Total simulation time
is 500s and allocation is performed at every 30s. Sim-
ulation parameters are shown in Table 1. Here,λ
1
ik
rep-
resents VM requirement of preprocessing task and λ
2
ik
represents VM requirement of abnormality detection
task (Mahmud et al., 2022)
Table 1: Simulation parameters.
Parameter Value
A
j
30
E
j
N(250,50)
v
i,k
N(500,100)
B
0
N(250,50)
β
i
N(0,1)
λ
1
ik
4
λ
2
ik
5
The patients are classified in four groups here
based on their criticality(sensitivity) as shown in Ta-
ble 2.
Table 2: Patient information.
Class % of patient Periodicity Deadline Sensitivity
0 10% 30s 10s 3
1 20% 60s 30s 3
2 30% 120s 60s 2
3 40% 240s 90s 1
The delay between the FSP and the ISP is mod-
elled in Netsim by simulating real-time scenarios like
congestion, link down, and packet loss, and then used
it in Python code for allocation. After this setup, we
have compare the following parameters in the alloca-
tion mechanism:
1. Task Allocation Ratio(TA): It is ratio of number
of tasks successfully executed and number of task
requests accepted by CSP.
2. Resource Utilization (RU): It is the ratio of total
VMs allocated and total VMs available.
3. Average User Utility(UU): It is the ratio of aver-
age IoSP utility per task and task valuation for all
assigned tasks.
4. System Utility (SU): It is the sum of the average
FSP utility(FU) and the average CSP utility(CU).
Where FU is the ratio of FSP utility per task to
task maintenance cost. CU is the ratio of the CSP
utility per task and the IoSP valuation for that task.
5.2 Results
We measure QoTAS performance using two bidding
strategies: truthful bidding and competitive bidding.
Normalized load (NL) is changed from 0 to 1 with
step size 0.1.
5.2.1 Truthful Biding Strategy
We assume that FSP and IoSP both bid their true task
valuations. In this bidding strategy, agents do not con-
sider market scenarios such as resource demand and
other agents’ bids. As shown in Figure 3(a), MADAs
TA is the lowest because it uses a matching algorithm
for VM allocation, allowing it to allocate only one
task per FSP. Also, because of that its TA decreases
as NL increase. DPDA has more TA than MADA be-
cause each FSP can be assigned multiple tasks. Qo-
TAS has the highest TA because it assigns resources
to tasks in multiple rounds if necessary. It also em-
ploys extended McAffee, which allocates the maxi-
mum number of FSP and IoSP pairs. Furthermore,
QoTAS allocates remaining resources at a low cost
as part of perishable allocation, so QoTAS has the
highest TA. Overall, QoTAS has 54% more TA than
DPDA and 69% more than MADA.
Because more resources are used when more tasks
are assigned, the resource utilization (RU) of QoTAS
is maximum and the RU of MADA is minimum as
shown in Figure 3(b). Also, for MADA, RU remains
constant since it always allocated constant number of
tasks which is equal to number of FSPs. Overall, Qo-
TAS has 63% more RU than DPDA and 75% more
than MADA.
CLOSER 2023 - 13th International Conference on Cloud Computing and Services Science
258
(a) Truthful bidding: NL vs. TA. (b) Truthful bidding: NL vs. RU. (c) Truthful bidding: NL vs. UU.
(d) Truthful bidding: NL vs. SU. (e) CompBid: NL vs. UU. (f) CompBid: NL vs. SU.
Figure 3: Comparision of QoTAS with DPDA and MADA.
As shown in Figure 3(c), the UU of VCG is the
highest because in VCG, buyers pay the seller’s ask at
the breakeven point. In contrast, in extended McAfee,
buyers pay the buyer’s bid at the break even point.
Because the seller’s ask at the breakeven point is al-
ways less than the buyer’s bid, buyers must pay less
in VCG. In our case, IoSPs are the buyers. QoTAS
has a higher UU than MADA because it completes
more tasks after accepting, avoiding penalties and in-
creasing UU. Oerall, DPDA has 59% more UU than
QoTAS and QoTAS has 54% more than MADA.
As shown in Figure 3(d), the SU of DPDA is min-
imum because it employs VCG, and in VCG auction-
eer has negative utility because it accepts less pay-
ment from buyers and gives more payment to sell-
ers, necessitating the addition of its own funds. In
our case, CSP is an auctioneer, so while DPDA has
the highest FSP utility , it has the lowest CSP util-
ity, making its total system utility the lowest of all
three. MADA has more SU than QoTAS because,
due to perishable allocation in QoTAS, it must sell re-
sources at a lower price than FSP’s ask, so CSP must
add money from its own pocket to pay FSP. This re-
duces CU and, eventually, SU. Overall, QoTAS has
89% more SU than DPDA and 26% less than MADA.
5.2.2 Competitive Bidding Strategy
In this section, we use CompBid strategy in which
IoSPs increased their bid as resource demand in-
creased. TA and RU has same behaviour as truth-
ful biddding since total number of tasks allocated de-
pends on available resources and QoS requirements.
As illustrated in Figure 3(e), the comparative be-
haviour of DPDA, MADA, and QoTAS is identical
to the Truthull bidding strategy depicted in 3(c). The
only difference is that the UU of MADA and QoTAS
decreases as load increases because IoSPs must pay
more because they are expected to pay the lowest bid
of all IoSPs and bid increases as load increases. UU
for DPDA remains constant because in VCG IoSP
pays seller’s ask, which is constant here. Overall,
DPDA has 78% more UU than QoTAS and QoTAS
has 20% more than MADA.
As illustrated in Figure 3(f), the SU of MADA and
QoTAS increases as load increases due to higher IoSP
bids. For DPDA, SU remains constant because the bid
of IoSP increases as the load increases. This causes
CSP to pay more to FSP, lowering CU and increasing
FU, but the SU is the sum of both, keeping SU con-
stant. Overall, QoTAS has 90% more SU than DPDA
and 12% more than MADA.
As a conclusion, QoTAS increase achieves more
TA and RU than DPDA and MADA in both bidding
strategy. Also, it has more SU than DPDA and more
UU than MADA in both truthful and CompBid strat-
egy. Also, it is observed that CompBid strategy in-
creases SU of QoTAS by 25% with some reduction in
UU.
6 CONCLUSION
Tasks with deadlines are generated by IoT applica-
tions. This task can be handled by cloud-fog service
providers. Task allocation and scheduling are difficult
in commercial cloud and fog architecture. To solve
this problem, a multi-attribute double auction is used.
The perishable nature of cloud-fog VMs motivates
QoS-Aware Task Allocation and Scheduling in Three-Tier Cloud-Fog-IoT Architecture Using Double Auction
259
allocation of the maximum number of VMs in each
round. The algorithm for allocating the remaining
VMs after the auction is proposed. To take advantage
of competition among users and service providers, a
competitive bidding strategy is proposed. To com-
pare QoTAS with existing studies MADA (Peng et al.,
2020) and DPDA (Joshi and Srivastava, 2020) remote
patient monitoring tasks are used. In both the truth-
ful and competitive bidding scenarios, QoTAS outper-
forms DPDA and MADA in both bidding strategies.
It has more 89% SU than DPDA and more 54% UU
than MADA. Furthermore, the CompBid strategy in-
creases SU of QoTAS by 25% while decreasing UU.
REFERENCES
(2022). IoT data. https://www.businesswire.com/news/ho
me/20190618005012/en/The-Growth-in-Connected
-IoT-Devices-is-Expected-to-Generate-79.4ZB-of
-Data-in-2025-According-to-a-New-IDC-Forecast.
[Online; accessed 18-03-2022].
Aggarwal, A., Kumar, N., Vidyarthi, D. P., and Buyya,
R. (2021). Fog-integrated cloud architecture en-
abled multi-attribute combinatorial reverse auctioning
framework. Simulation Modelling Practice and The-
ory, 109:102307.
Baranwal, G., Kumar, D., Raza, Z., and Vidyarthi, D. P.
(2018). Auction based resource provisioning in cloud
computing. Springer.
Chowdhury, M. M. P., Kiekintveld, C., Tran, S., and Yeoh,
W. (2018). Bidding in periodic double auctions using
heuristics and dynamic monte carlo tree search. In
IJCAI, pages 166–172.
Guo, Y., Saito, T., Oma, R., Nakamura, S., Enokido, T.,
and Takizawa, M. (2020). Distributed approach to
fog computing with auction method. In International
Conference on Advanced Information Networking and
Applications, pages 268–275. Springer.
Jin, A.-L., Song, W., Wang, P., Niyato, D., and Ju, P. (2015).
Auction mechanisms toward efficient resource sharing
for cloudlets in mobile cloud computing. IEEE Trans-
actions on Services Computing, 9(6):895–909.
Joshi, N. and Srivastava, S. (2019). Task allocation in three
tier fog iot architecture for patient monitoring sys-
tem using stackelberg game and matching algorithm.
In 2019 IEEE International Conference on Advanced
Networks and Telecommunications Systems (ANTS),
pages 1–6. IEEE.
Joshi, N. and Srivastava, S. (2020). Auction-based deadline
and priority enabled resource allocation in fog-iot ar-
chitecture. In 2020 Futuristic Trends in Network and
Communication Technologies(FTNCT). Springer.
Kumar, S., Kambhatla, K., Hu, F., Lifson, M., and Xiao,
Y. (2008). Ubiquitous computing for remote cardiac
patient monitoring: a survey. International journal of
telemedicine and applications, 2008.
Mahmud, R., Pallewatta, S., Goudarzi, M., and Buyya, R.
(2022). Ifogsim2: An extended ifogsim simulator for
mobility, clustering, and microservice management in
edge and fog computing environments. Journal of
Systems and Software, 190:111351.
Miyashita, K. (2014). Online double auction mechanism
for perishable goods. Electronic Commerce Research
and Applications, 13(5):355–367.
Peng, X., Ota, K., and Dong, M. (2020). Multiattribute-
based double auction toward resource allocation in ve-
hicular fog computing. IEEE Internet of Things Jour-
nal, 7(4):3094–3103.
Safianowska, M. B., Gdowski, R., and Huang, C.
(2017). Preventing market collapse in auctions
for perishable internet of things resources. Inter-
national Journal of Distributed Sensor Networks,
13(11):1550147717743693.
Suh, M.-k., Chen, C.-A., Woodbridge, J., Tu, M. K., Kim,
J. I., Nahapetian, A., Evangelista, L. S., and Sar-
rafzadeh, M. (2011). A remote patient monitoring
system for congestive heart failure. Journal of med-
ical systems, 35(5):1165–1179.
Thavikulwat, P. and Pillutla, S. (2008). Pure and mixed
strategies for programmed trading in a periodic dou-
ble auction. International Journal of Applied Decision
Sciences, 1(3):338–358.
Verma, P. and Sood, S. K. (2018). Fog assisted-iot enabled
patient health monitoring in smart homes. IEEE Inter-
net of Things Journal.
Weinman, J. (2017). Fogonomics the strategic, eco-
nomic, and financial aspects of the cloud. In 2017
IEEE 41st Annual Computer Software and Applica-
tions Conference (COMPSAC), volume 1, pages 705–
705.
Yang, D., Fang, X., and Xue, G. (2011). Truthful auction
for cooperative communications. In Proceedings of
the Twelfth ACM International Symposium on Mobile
Ad Hoc Networking and Computing, pages 1–10.
Zhang, H., Xiao, Y., Bu, S., Niyato, D., Yu, F. R., and Han,
Z. (2017). Computing resource allocation in three-tier
iot fog networks: A joint optimization approach com-
bining stackelberg game and matching. IEEE Internet
of Things Journal, 4(5):1204–1215.
Zhang, Y., Wang, C.-Y., and Wei, H.-Y. (2019). Parking
reservation auction for parked vehicle assistance in ve-
hicular fog computing. IEEE Transactions on Vehicu-
lar Technology, 68(4):3126–3139.
Zu, Y., Shen, F., Yan, F., Yang, Y., Zhang, Y., Bu, Z., and
Shen, L. (2019). An auction-based mechanism for
task offloading in fog networks. In 2019 IEEE 30th
Annual International Symposium on Personal, Indoor
and Mobile Radio Communications (PIMRC), pages
1–6. IEEE.
CLOSER 2023 - 13th International Conference on Cloud Computing and Services Science
260