Broker-based optimization of SLA negotiations in cloud computing

Abstract

Cloud computing provisions and allocates resources, in advance or real-time, to dynamic applications planned for execution. This is a challenging task as the Cloud-Service-Providers (CSPs) may not have sufficient resources at all times to satisfy the resource requests of the Cloud-Service-Users (CSUs). Further, the CSPs and CSUs have conflicting interests and may have different utilities. Service-Level-Agreement (SLA) negotiations among CSPs and CSUs can address these limitations. User Agents (UAs) negotiate for resources on behalf of the CSUs and help reduce the overall costs for the CSUs and enhance the resource utilization for the CSPs. This research proposes a broker-based mediation framework to optimize the SLA negotiation strategies between UAs and CSPs in Cloud environment. The impact of the proposed framework on utility, negotiation time, and request satisfaction are evaluated. The empirical results show that these strategies favor cooperative negotiation and achieve significantly higher utilities, higher satisfaction, and faster negotiation speed for all the entities involved in the negotiation.

Keywords

Cloud computing agents negotiation strategies SLA

1. Introduction

Cloud service provides virtual resources and secure and adaptable infrastructures through Infrastructure-as-a-Service (IaaS) delivery model. However, the Cloud-Service-Providers (CSPs) do not have infinite amount of resources. Also, the Cloud-Service-Users (CSUs) consistently have budget and time constraints in availing the resources of the CSPs. This situation can be optimized if the CSUs and CSPs negotiate the Service-Level-Agreements (SLAs) regarding their resources (i.e., utilities). This research presents a Broker ( $B$ ) based SLA negotiation framework in which the CSUs (through User Agents (UAs)) and CSPs are represented as buyers and sellers, respectively. A list of some acronyms used in this research, along with their meanings, is provided in Table 1.

Table 1
A list of some acronyms and their meaning

Notations	Meaning
ACN	Automated concurrent negotiations
SRCNS	Static resource cost negotiation scheme
ABAP	Agreeable by all proposals
$\textit{AC}_{\text{oppo}}(\cdot)$	Offer acceptance criterion
$\textit{AC}_{\textit{UA}_{i}}$	Offer acceptable by $\textit{UA}_{i}$ from $\textit{CSP}_{y}$
$\textit{AC}^{\prime}_{\textit{UA}_{i}}$	Partial offer acceptable by $\textit{UA}_{i}$ from $\textit{CSP}_{y}$
$B$	Broker
$B_{x}$	‘ $x_{\text{th}}$ ’ broker
CSU	Cloud service user
CSUs	Cloud service users
$\textit{CSU}_{id}$	Cloud service user identified with $i d$
UA	User Agent
UAs	User Agents
$\textit{UA}_{i}$	‘ $i_{\text{th}}$ ’ user agent
CSP	Cloud service provider
CSPs	Cloud service providers
$\textit{CSP}_{y}$	‘ $y_{\text{th}}$ ’ cloud service provider
IaaS	Infrastructure-as-a-Service
NPCA	Negotiation through participation of CSU agents
$\beta_{op}$	Concession Speed of $\textit{CSP}_{y}$
PRACP	Proposal ranking according to CSU proposals
QoS	Quality-of-service
$\textit{RESV\_UTIL}_{\textit{CSP}_{y}}$	Minimum utility expected by $\textit{CSP}_{y}$ for offer to be accepted from $\textit{UA}_{i}$
$\textit{RESV\_UTIL}_{\textit{UA}_{i}}$	Minimum utility expected by $\textit{UA}_{i}$ for offer to be accepted from $\textit{CSP}_{y}$
RSL	Random Selection of Leader
$t$	Current round of the negotiation
$\textit{UTIL}_{\textit{UA}_{i}}$	$\textit{UA}_{i}$ Agent’s Utility achieved
$s_{\textit{CSP}_{y}}(\cdot)$	$\textit{CSP}_{y}$ ’s concession strategy
SLA	Service level agreement
$T_{\textit{CSP}_{y}}$	Opponent, $\textit{CSP}_{y}$ ’s, deadline
$\textit{UTIL}_{\textit{CSP}_{y}}(\cdot)$	Utility of the $\textit{CSP}_{y}$
VM	Virtual machine
$X$	Complete proposal
$X^{\prime}$	Partial proposal
$X^{t}_{\textit{UA}_{i}\rightarrow\textit{CSP}_{y}}$	Offer received by $\textit{CSP}_{y}\textit{from}\textit{UA}_{i}$

SLAs have developed as a new technique to control resource utilization in the Cloud, and a number of Cloud services, such as Amazon EC2 and Google Cloud Platform, have already implemented them. As a result, it is only reasonable to explore ways to include SLAs into job scheduling [1]. To accomplish it, effective and flexible solutions that facilitate dynamic SLA negotiation are required.

A dynamic SLA is the agreement that is negotiated each time the service is to be delivered. The services to be negotiated are resource provisioning services, which must deliver higher-level services such as solving a complex problem, running a given application, and so forth [12]. These services necessitate the utilization of a variety of resources, such as computing nodes, network connections, storage locations, or any combination thereof [22].

Agent framework, which is a collection of autonomous software agents, is significant in resource negotiation strategies in Cloud computing [14]. Agents negotiate over SLAs for the utilities associated with service provisioning and availability of resources. Agent-based negotiation strategies comprise a novel subject of research pertaining to negotiation strategies [14]. Schemes to permit agent-based negotiation strategies are required for handling SLA negotiations among CSUs and CSPs.

This research considers a concurrent one-to-many negotiation mechanism for resource allocation in which a UA negotiates, for multiple resources, with multiple CSPs. It involves agents viz. UA and $B$ for the SLA negotiation. $B$ mediates SLA negotiation between CSUs (through UAs) and CSPs by using multiple negotiation strategies. A strategy between $\textit{UA}_{i}$ and $\textit{CSP}_{y}$ defines what, how, and when the decisions have to be taken by both these entities while negotiating for resources. The following four negotiation strategies are proposed in this research: (i) Random Selection of Leader (RSL), (ii) Negotiation through Participation of CSU Agents (NPCA), (iii) Proposal Ranking According to CSU Proposals (PRACP), and (iv) Agreeable By All Proposals (ABAP). The proposed framework allows the UAs, $B s$ , and CSPs to develop an adaptive strategy towards concurrent negotiations in an environment with incomplete information in which a CSU knows its own negotiation parameters but has incomplete information about its opponent’s (i.e., CSP’s) parameters.

Table 2

A summary of some existing resource negotiation schemes

Schemes	Description	Limitations
Automated concurrent negotiations (ACN) [15]	Interactions are realized in the form of negotiations over a number of item characteristics	In case of a centralized approach, many messages are required for the communication between the coordinator and threads
Static resource cost negotiation scheme (SRCNS) [7]	Concurrent and cooperative cost negotiation model using game theory	Considers only the cost for negotiation
Fuzzy negotiation [16]	Provides optimization of convergence speed in the negotiation process	QoS preferences such as reliability and security are not considered
Reputation-based competitive pricing negotiation [21]	Decentralized price negotiations in a network of microgrids	Utility revenue decreased by 21–27% when allowing microgrids to trade
Gradual bargaining protocol [8]	Construction of two alternating-offer games	CSP agent finds joint gains without concern for the users
Ontology-based negotiation framework [23]	Ability to integrate a semantic mapping between client’s request and the provider’s service	Needs to test the effect of changing other Cloud service contextual parameters in the SLA document
Market-driven agents [13]	Negotiation agents are flexible in their bargaining, depending on resource demands to acquire the resource successfully	Real-time resource updates are not considered
Incomplete information, non-cooperative and alternating offers [11]	Formulates a fair pricing strategy to determine an amount that the user has to pay to providers	Time complexity is more and deals with complicated calculations
Uncertainty-based negotiation [18]	Bilateral bargaining by considering the deadline and reserved price	Rules for multi-party negotiations are undefined

The rest of the paper is organized as follows. The next section presents some background on the issues of negotiation and SLA in Cloud computing. Section 3 presents the proposed framework for SLA negotiation strategies. Section 4 presents the simulation studies and results. Finally, Section 5 concludes this research and highlights some future work.

2. Literature review

Negotiation is a widely studied topic and there are numerous publications addressing various aspects of negotiation. The current research differs from the existing negotiation strategies such as automated concurrent negotiations, fuzzy negotiation, etc. shown in Table 2.

Researchers [2] surveyed the need for devising a technique for resolving the Cloud service negotiation challenge that is well aligned with the needs of both users and the providers. The work in [3] presents a method for aggregating the outcomes of simple task requirement in order to ensure end-to-end composite service requirements. The research in [24] provides a novel agent-based technique to negotiate permission for users and services to exchange private data.

Existing literature also presents the broker-based models to study the issues in SLA negotiation. A broker acts as a middleware between the CSUs and CSPs. Most of the research in this domain is designed for supporting negotiation in a single market. However, such research does not consider dynamic or parallel negotiations in multiple inter-related markets such as e-commerce, web services, etc. (e.g., [4]). For the directed networks with some Byzantine agents, a hybrid censoring technique is devised to reach robust consensus for cooperative agents [25]. A fuzzy membership function is used to represent imprecise Quality of Service (QoS) preferences for SLA negotiation [16]. However, it fails to investigate SLA re-negotiations for dynamic resource capacities of the CSPs.

Some significant research is available on the negotiations in Cloud environment [18, 4]. However, that research does not include dynamicity with respect to the SLA negotiation strategies between the CSUs and the CSPs. The current research is intended to overcome that drawback, resulting in more number of successful negotiations.

Figure 1.

A negotiation scenario.

3. Broker-based mediation in cloud computing

A one-to-one negotiation between a CSU and a CSP may not be an optimal approach for negotiation, as the CSU may have to pay more price for the resource(s) and the CSP may not be able to utilize all its available resources, i.e., Virtual Machines (VMs). In this research, a $B s$ mediated SLA negotiation framework for the UAs and CSPs to enhance the utilities of the CSUs (through UAs) and CSPs is addressed. Utility is defined as the profit or satisfaction, in terms of price, deadline, and resource types, received from the negotiation process. UA is responsible for consolidating its CSUs’ requirements for the resources and negotiating with one or more CSPs on behalf of those CSUs. The utilities of the UAs are in proportion to the resource requirements made by its CSUs (e.g., a CSU requesting more resource shall gain higher utility than another CSU that requests for lesser resource). The Bs mediated negotiation scenarios for UAs and CSPs in cooperative [7] and non-cooperative [15] environments, with the utilities of the UAs and CSPs, are considered.

A new agent-based dynamic SLA negotiations framework for many-to-many concurrent multilateral negotiations in an open, dynamic, and parallel Cloud service environment, as shown in Fig. 1, is proposed. This framework allows the UAs, $B s$ , and CSPs to develop an adaptive strategy towards concurrent negotiations in an environment with incomplete information in which a CSU knows its own negotiation parameters but has incomplete information about its opponent’s (i.e., CSP’s) parameters. The steps encountered from resource request made by a single CSU all the way to resource allocation made by a CSP are shown in Fig. 2. The sequence diagram of an UA strategy for negotiation is shown in Fig. 3.

Figure 2.

The steps encountered in the process of resource request made by a single CSU all the way to resource allocation by a CSP.

Figure 3.

Sequence diagram for the negotiations of SLAs. S1: RSL, S2: NPCA, S3: PRACP, and S4: ABAP are the negotiation strategies discussed in Section 3.2.

KwInInputKwOutOutputKwDesDescriptionKwbeginBegin KwendEnd

[t!] $B$ service module

Resource request from $\textit{UA}_{i}$ : $\textit{CSU}_{id}$ with needs ( $\textit{VM}<\textit{CPU},\textit{RAM},\textit{Storage}>$ )

allocated VMs to $\textit{CSU}_{id}$

call $B_{x}$ _service_intialization() //Given in Algorithm 3

for ( $id=1$ ; $id\leqslant$ 250; $id++$ )

{

if (SLA between [ $\textit{CSU}_{id}$ and $\textit{CSP}_{y}$ ] is acceptable) assign (by $\textit{CSP}_{y}$ ) $\textit{VM}<\textit{CPU},\textit{RAM},\textit{Storage}>$ to $\textit{CSU}_{id}$ ;

else GOTO 8 // for all $\textit{CSU}_{id}$ not accepting the SLA

}

Call (Mediation of SLA negotiation) //Given in Algorithm 3

return $\textit{VM}<\textit{CPU},\textit{RAM},\textit{Storage}>$ to the remaining $\textit{CSU}_{id}$

[t!] $B_{x}$ _service_intialization()

(i) resource request from $\textit{UA}_{i}$ , (ii) list of registered CSPs with $B_{x}$ $\textit{CSP}_{y}$ providing the resources requested by $\textit{CSU}_{id}$

for each of the CSPs registered with $B_{x}$

analyze the resource request from each $\textit{CSU}_{id}$ with needs ( $\textit{VM}<\textit{CPU},\textit{RAM},\textit{Storage}>$ )

search for the CSP that matches $\textit{CSU}_{id}s^{\prime}$ resource requirement

return $\textit{CSP}_{y}$ and exit // this is the appropriate CSP to serve the required resources

if $\textit{CSP}_{y}$ $==$ NULL // in case an appropriate CSP is not in the registered list

search the list of CSPs not registered with $B_{x}$

(i.e., outside $B_{x}$ ’s database)

call for CSPs to register with $B_{x}$ //given in Algorithm 3

GOTO 1

An UA is an approaching point for multiple CSUs with similar resource requirement specifications for running the resource-intensive applications of those CSUs. There could be many UAs in the Cloud market catering to varied resource requirements of the CSUs. For example, one UA could assist the CSUs for medium-size VMs, while another UA could assist for large-size VMs.

In the IaaS scenario that is considered in this research, initially, one or more CSUs submit(s) queries for resource requirements to multiple UAs. Each UA in turn contacts multiple $B s$ . Each $B$ analyzes the queries it received from an UA to assess the requirement of those queries. The $B$ searches its own database to suggest an appropriate CSP for the incoming request for VMs from the UA. The SLA information pertaining to the appropriate CSP is passed to the CSU (that submitted the query) through its UA. Now, if the CSU agrees to the SLA specified by the CSP, the resource requests and allocation procedure starts between that CSU and that CSP, respectively, i.e., on one-to-one basis.

[t!] Mediation of SLA negotiation

$\textit{CSU}_{id}$ (from $\textit{UA}_{i}$ ) that requires SLA negotiation a negotiation strategy between a $\textit{CSU}_{id}$ and a $\textit{CSP}_{y}$

mediate negotiation strategies of $\textit{CSU}_{id}$ with $\textit{CSP}_{y}$ (through feasible negotiation strategies discussed in Section 3.2)

[t!] CSPs’ registration with $B_{x}$

registration request from $B_{x}$ to register the CSPslist of new CSPs successfully registered with $B_{x}$

send the required VM specification details of each $\textit{CSU}_{id}$ to all the CSPs

receive the specifications of VMs available at all the CSPs

create entry for all the CSPs in $B_{x}$ ’s database (indicating CSPs’ successful registration with $B_{x}$ ’s database)

return to Step 6 of Algorithm 3

If one or more CSUs have budget constraints and prefer to negotiate with a CSP (exactly one CSP which can match the resource requirements of the CSUs), those $\textit{CSU}(s)$ communicate(s) the same to their appropriate UAs. An appropriate UA for a CSU would be the one that represents that CSU in negotiating with a CSP optimally, considering the resource types, deadline and price parameters specified by the CSU. Depending on the negotiation profile (cooperative or non-cooperative) of similar requests in the past, $B$ (exactly one $B$ ) initiates multiple strategies to enable a cooperative or non-cooperative negotiation between that UA and a CSP (one-to-one) for the incoming requests (for VMs). If the incoming request from an UA has not been encountered by a $B$ during the previous requests, the $B$ queries the details (e.g., the VM type, its usage, number of users, etc.) from that CSU (through its UA).

After processing the query, the $B$ adds the current request details to its database(s) for future reference. The proposed work considers a concurrent many-to-many negotiation mechanism in which a UA (one UA for multiple CSUs) and a CSP negotiate (i.e., one-to-one negotiation between UA and CSP), and this negotiation is mediated by a single $B$ . The service request(s) along with the mediation procedures and service response(s) by $B$ are given in Algorithms 1–4. Algorithm 3 accepts the requirements from the CSUs then pass on those requirements to the CSP. The SLA negotiation for the requested service is monitored. If the SLA is acceptable in the original form then the resources are exchanged between the CSU and the CSP, else a mediation of SLA negotiation is performed through Algorithm 3. Algorithms 3 and 4 aid in exploring new CSPs registering with $B$ and fulfilling the requirements of the CSUs.

The Iterated Contract Net Protocol (ICNP) [6], which is a commonly used negotiation protocol, is used in this study. Recursive negotiation is supported by ICNP, as is multi-round iterative negotiation, to reach a common agreement. Further, by exchanging the modified proposals and counter-proposals, a more acceptable negotiation result is likely to be reached. Next, the proposed SLA negotiation strategies among the UAs and CSPs through $B s$ are discussed.

3.1 Negotiations strategies

In this section, the assumptions and the negotiation strategies employed for the proposed model are presented.

There are many CSUs and many UAs in the Cloud market. Different UAs assist the CSUs in getting the required resources needed to run the applications of the CSUs. For example, one UA could cater to assisting the CSUs for medium-size VMs, while another UA could assist in getting large-size VMs. There are several $B s$ for mediation during negotiation, and several CSPs for resource provisioning and allocation. CSP agents are not considered in this work because, currently, in the market, the number of CSPs is small when compared to the number of UAs and $B s$ .

In this research, there is exactly one UA for all the CSUs that require medium-size VMs. That UA is represented by $\textit{UA}_{i}$ . In this research, the range of CSUs associated with the $\textit{UA}_{i}$ is in between 1 and 250. The $\textit{UA}_{i}$ aims to negotiate a successful deal for its CSUs (belonging to the $\textit{UA}_{i}$ ) and $\textit{CSP}_{y}$ , where $\textit{CSP}_{y}$ is one of the available CSPs. In this research, the total number of available CSPs is in between 1 and 5. The negotiation between $\textit{UA}_{i}$ and $\textit{CSP}_{y}$ is mediated by $B_{x}$ , where $B_{x}$ is one of the available $B s$ . In this research, the total number of available $B s$ is in between 1 and 5.

In the mathematical modeling of the negotiation strategies of this work, a single $B$ ( $B_{x}$ ) and a single CSP ( $\textit{CSP}_{y}$ ) are considered. However, the same modeling is applicable to a scenario that contains multiple UAs, CSPs, and $B s$ . Each entity, i.e., $\textit{UA}_{i}$ and $\textit{CSP}_{y}$ , that is involved in the negotiation process aims to maximize its own utility. For example, an utility of 0.60 for $\textit{UA}_{i}$ indicates that out of 100 negotiations between $\textit{UA}_{i}$ and $\textit{CSP}_{y}$ , 60 negotiations are as per the satisfaction of $\textit{UA}_{i}$ (and hence for the satisfaction of its corresponding CSUs). The $\textit{CSP}_{y}$ has a reservation utility $\textit{RESV\_UTIL}_{\textit{CSP}_{y}}$ . Any proposal from $B_{x}$ whose utility is lower than $\textit{RESV\_UTIL}_{\textit{CSP}_{y}}$ will be rejected by $\textit{CSP}_{y}$ . A time-based concession tactic [9, 19] to model the tactic of the opponent (i.e., $\textit{CSP}_{y}$ ), as expressed in Eq. (1), is used in this work.

$\displaystyle s_{\textit{CSP}_{y}}(t)=1-(1-\textit{RESV\_UTIL}_{\textit{CSP}_{% y}})\left(\frac{t}{T_{\textit{CSP}_{y}}}\right)^{\frac{1}{\beta_{\textit{CSP}_% {y}}}}$ (1)

In Eq. (1), $t$ is the current negotiation round number, $s_{\textit{CSP}_{y}}(\cdot)$ is the $\textit{CSP}_{y}$ ’s concession strategy, $T_{\textit{CSP}_{y}}$ is the deadline set by $\textit{CSP}_{y}$ to complete the negotiation, and $\beta_{\textit{CSP}_{y}}$ is a parameter that governs the concession speed of the negotiation. The concession speed $\beta_{\textit{CSP}_{y}}$ of $\textit{CSP}_{y}$ during the negotiation process determines the final outcome of the agreement for the CSUs of a $\textit{UA}_{i}$ . For instance, if the $\textit{CSP}_{y}$ concedes very quickly towards its reservation utility, better agreements for the CSUs may come earlier in the negotiation process.

The $\textit{CSP}_{y}$ uses an offer acceptance criterion $\textit{AC}_{\textit{CSP}_{y}}(\cdot)$ during the negotiation process. Inspired by the research in [9, 19], the criterion is formalized as shown in Eq. (2).

$\displaystyle AC_{\textit{CSP}_{y}}(X_{\textit{UA}_{i}\rightarrow\textit{CSP}_% {y}}^{t})=\begin{cases}\text{accept}&\text{if∼{}}s_{\textit{CSP}_{y}}(t+1)% \leqslant\textit{UTIL}_{\textit{CSP}_{y}}(X_{\textit{UA}_{i}\rightarrow\textit% {CSP}_{y}}^{t})\\ \text{reject}&\text{otherwise}\end{cases}$ (2)

where $X^{t}_{\textit{UA}_{i}\rightarrow\textit{CSP}_{y}}$ is the proposal received by $\textit{CSP}_{y}$ from the $\textit{UA}_{i}$ , $\textit{UTIL}_{\textit{CSP}_{y}}(\cdot)$ is the utility function of the $\textit{CSP}_{y}$ , $s_{\textit{CSP}_{y}}(\cdot)$ is the $\textit{CSP}_{y}$ ’s concession strategy, and $t$ is the current round number of the negotiation process.

The $\textit{CSP}_{y}$ proposes $X_{\textit{CSP}_{y}\rightarrow\textit{UA}_{i}}^{t}$ , whose utility is given by $\textit{UTIL}_{\textit{CSP}_{y}}(X_{\textit{CSP}_{y}\rightarrow\textit{UA}_{i}% }^{t})=s_{\textit{CSP}_{y}}(t)$ . Inspired by [20], it is assumed that the $\textit{CSP}_{y}$ attempts to propose the offer that is more similar to the last offer received from $\textit{UA}_{i}$ , and whose utility is mathematically represented as $\textit{UTIL}_{\textit{CSP}_{y}}(X_{\textit{CSP}_{y}\rightarrow\textit{UA}_{i}% }^{t})=s_{\textit{CSP}_{y}}(t)$ .

3.2 UA strategies for negotiation

A strategy between $\textit{UA}_{i}$ and $\textit{CSP}_{y}$ defines what, how, and when the decisions have to be taken by both these entities while negotiating for resources. The following four negotiation strategies are proposed in this research: (i) RSL, (ii) NPCA, (iii) PRACP, and (iv) ABAP. In these four strategies that are discussed in the following subsections, $B_{x}$ represents the ‘ $x_{\text{th}}$ ’ broker. The framework w.r.t. a single $\textit{UA}_{i}$ is discussed. However, the same is applicable to a scenario with multiple $\textit{UA}_{i}s$ .

3.2.1 Random selection of leader (RSL)

RSL is a simple strategy employed by $B_{x}$ , in which $\textit{UA}_{i}$ initiates the negotiation on behalf of its CSUs that need resources for running the resource intensive applications. One of the CSU members from $\textit{UA}_{i}$ , $csu_{iL}$ , is randomly selected as the leader by $B_{x}$ . The leader negotiates with $\textit{CSP}_{y}$ , according to the leader’s ( $csu_{iL}^{\prime}s$ ) own utility function ${csu_{iL}}(\cdot)$ , on behalf of all the CSUs looking for similar resource(s) over the Cloud. The criteria for selecting the leader (among several CSUs of the $\textit{UA}_{i}$ ) is decided by $B_{x}$ through one of the following two methods:

1.
Request-based hierarchy: The CSU for which the amount of each of the needed resources is the highest will be the leader. If a CSU needs more resources of type $a$ and another needs more resources of type $b$ , the cumulative sum of different resources that are required is prioritized for the selection of the leader. In case of a tie, the history of the CSUs’ transactions is considered, and the CSU that requested more resources in the past will be the leader.
2.
Capacity-based hierarchy: The CSU for which the resource usage duration is the highest will be the leader. In case of a tie, the history of the CSUs’ transactions is considered, and the CSU that utilized more resources efficiently in the past will be the leader.

Being a time-bounded negotiation, the leader $csu_{iL}$ employs a time-based concession tactic, $s_{csu_{iL}}(t)$ , to negotiate with the $\textit{CSP}_{y}$ , as provided in Eq. (3).

$\displaystyle s_{csu_{iL}}(t)=1-(1-\textit{RESV\_UTIL}_{csu_{iL}})\left(\frac{% t}{csu_{iL}}\right)^{\frac{1}{\beta_{csu_{iL}}}}$ (3)

Any proposal $X_{\textit{UA}_{i}}\rightarrow\textit{CSP}_{y}^{t}$ proposed by $\textit{csu}_{iL}$ to $\textit{CSP}_{y}$ at round $t$ will obey the condition given in Eq. (4).

$\displaystyle\textit{UTIL}_{\textit{csu}_{iL}}(X_{\textit{UA}_{i}\rightarrow% \textit{CSP}_{y}}^{t})=s_{\textit{UA}_{\textit{icsu}_{iL}}}(t)$ (4)

With the $\textit{CSP}_{y}$ ’s strategy, the leader $csu_{iL}$ selects, in current negotiation round number $t$ , the offer that is more similar to the previous offer (during previous negotiation in round number $t-1$ ), given in Eq. (5). Here, $X|\textit{UTIL}_{csu_{iL}}(X)$ denotes the utility associated with the selection of proposal $X$ .

$\displaystyle X_{\textit{UA}_{i}\rightarrow\textit{CSP}_{y}}^{t}=\max_{X|% \textit{UTIL}_{\textit{csu}_{iL}}(X)=s_{csu_{iL}}(t)}\textit{Sim}(X,X_{\textit% {CSP}_{y}\rightarrow\textit{UA}_{i}}^{t-1})$ (5)

The leader accepts the $\textit{CSP}_{y}$ ’s offer, $X_{\textit{CSP}_{y}\rightarrow csu_{iL}}^{t}$ , at round $t$ , provided the utility $\textit{UTIL}_{\textit{csu}_{iL}}$ $(X_{\textit{CSP}_{y}\rightarrow\textit{csu}_{iL}}^{t})$ is greater than or equal to $s_{\textit{csu}_{iL}}(t+1)$ .

Type of negotiation: Since $B_{x}$ mediates the negotiations according to $csu_{iL}$ utility function, $B_{x}$ cannot guarantee any kind of unanimity regarding the leader’s decision (i.e., acceptance of leader’s decision by other CSU members of the $\textit{UA}_{i}$ ). Hence, it is considered as a non-cooperative negotiation between $\textit{UA}_{i}$ and $\textit{CSP}_{y}$ . It appears as if this strategy is not worth being used because no unanimity can be assured. However, when all CSU members of the $\textit{UA}_{i}$ tend to be very similar in their resource requirements, this strategy yields fruitful results.
3.2.2 Negotiation through participation of CSU agents (NPCA)

Whenever a new proposal from one or more $\textit{CSU}(s)$ have to be proposed to the $\textit{CSP}_{y}$ at round $t$ , $B_{x}$ opens a call for resource proposals (containing the description of the required VMs with the specifications of the CPU, RAM, and storage requirements) among the $\textit{UA}_{i}$ members that are looking for resources. Each such $\textit{CSU}_{id}$ of the $\textit{UA}_{i}$ anonymously sends a proposal $X_{\textit{UA}_{i}\rightarrow\textit{CSU}_{id}}^{t}$ to $\textit{CSP}_{y}$ . Once every proposal has been gathered, $B_{x}$ opens a voting process where the proposals are made public among the members of $\textit{UA}_{i}$ that participate in the voting. Then, each $\textit{CSU}_{id}$ of the $\textit{UA}_{i}$ communicates a multi-vote, $\textit{Vote}_{\textit{CSU}_{id}}$ , to the $B_{x}$ .

When all the votes are assembled, $B_{x}$ summarizes the quantity of positive votes, and the most upheld proposition $X_{\textit{UA}_{i}\rightarrow\textit{CSP}_{y}}^{t}$ is chosen according to Eq. (6). It is then made public among the $\textit{CSU}_{id}s$ , and the winning offer is communicated to $\textit{CSP}_{y}$ .

$\displaystyle X_{\textit{UA}_{i}\rightarrow\textit{CSP}_{y}}^{t}=(\mathop{% \operatorname{arg\,max}}\limits_{X_{j}\in X_{\textit{UA}_{i}^{t}}})\sum_{% \textit{CSU}_{id}\in\textit{UA}_{i}}\textit{Vote}_{\textit{CSU}_{id}}(j)$ (6)

Since the negotiation is time-bound, $\textit{UA}_{i}$ members follow a time-based concession strategy with the $\textit{CSP}_{y}$ , where the concession speed $\beta_{\textit{UA}_{i}}$ is common and agreed by the $\textit{UA}_{i}$ members at the beginning of the negotiation process given in Eq. (7).

$\displaystyle s_{\textit{UA}_{i}(t)}=1-(1-\textit{RESV\_UTIL}_{\textit{UA}_{i}% })\left(\frac{t}{\textit{UA}_{i}}\right)^{\frac{1}{\beta_{\textit{UA}_{i}}}}$ (7)

In case of multiple offers with such utility, the $\textit{UA}_{i}$ needs to pick at random one of those offers received from $\textit{CSU}_{id}$ s. Let $X$ denotes an offer. If the $\textit{CSU}_{id}$ needs its offer $X_{\textit{CSU}_{id}\rightarrow\textit{UA}_{i}}^{t}$ to be accepted by $\textit{UA}_{i}$ , it ought to boost its likelihood being the most accepted by other CSU s as given in Eq. (8).

$\displaystyle X_{\textit{CSU}_{id}\rightarrow\textit{UA}_{i}}^{t}=\mathop{% \operatorname{arg\,max}}\limits_{X|\textit{UTIL}_{\textit{UA}_{i}}(X)=s_{% \textit{UA}_{i}}(t)}p_{\textit{CSP}_{y}}(X)*p_{\textit{UA}_{i}}(X)$ (8)

In Eq. (8), $p_{\textit{CSP}_{y}}(X)$ is the probability for $X$ to be accepted by the $\textit{CSP}_{y}$ , and $p_{\textit{UA}_{i}}(X)$ is the probability for $X$ to be selected by the $\textit{CSU}_{id}$ . The offer selection and rejection mechanism can be specified as in Eq. (3.2.2).

$\displaystyle\textit{AC}_{\textit{UA}_{i}}(X_{\textit{CSP}_{y}\rightarrow% \textit{UA}_{i}}^{t})=\begin{cases}\text{accept}&\displaystyle\text{if∼{}}\sum% \limits_{\textit{CSU}_{id}\in\textit{UA}_{i}}\textit{AC}_{\textit{CSU}_{id}}(X% _{\textit{CSP}_{y}\rightarrow\textit{UA}_{i}}^{t})>\frac{|\textit{UA}_{i}|}{2}% \\ \text{reject}&\displaystyle\text{if∼{}}\sum\limits_{\textit{CSU}_{id}\in% \textit{UA}_{i}}\textit{AC}_{\textit{CSU}_{id}}(X_{\textit{CSP}_{y}\rightarrow% \textit{UA}_{i}}^{t})<\frac{|\textit{UA}_{i}|}{2}\\ \text{random (either accept or reject)}&\text{otherwise}\end{cases}$

The way in which CSUs decide the acceptability of the $\textit{CSP}_{y}$ ’s offer $\textit{AC}_{\textit{CSU}_{id}}(X_{\textit{CSP}_{y}\rightarrow\textit{UA}_{i}}% ^{t})$ follows the rational mechanism that is employed so far in this work. The offer is considered acceptable if it yields a utility which is greater than or equal to the utility demanded by the concession strategy in the next negotiation round $s_{\textit{CSU}_{id}}(t+1)$ . Otherwise, the offer is not considered acceptable. Equation (10) formalizes the acceptance criterion used by $\textit{CSU}_{id}$ in determining the offer made by $\textit{CSP}_{y}$ .

$\displaystyle AC_{\textit{CSU}_{id}}(X_{\textit{CSP}_{y}\rightarrow\textit{UA}% _{i}}^{t})=\begin{cases}\text{accept}&\text{if∼{}}\textit{UTIL}_{\textit{CSU}_% {id}}(X_{\textit{CSP}_{y}\rightarrow\textit{UA}_{i}}^{t})\geqslant s_{\textit{% CSU}_{id}}(t+1)\\ \text{reject}&\text{otherwise}\end{cases}$ (10)

Type of negotiation: This strategy is supported by a plurality and majority of the CSUs, regarding $\textit{UA}_{i}s^{\prime}$ decisions to select the best proposal. Plurality is guaranteed when the offer is proposed (by $\textit{UA}_{i}$ ) to the $\textit{CSP}_{y}$ by selecting that offer from multiple CSUs by using a voting mechanism as explained, and majority is assured when choosing the offers from multiple CSPs, resulting in cooperative negotiation scenario.

3.2.3 Proposal ranking according to CSU proposals (PRACP)

The essential feature of PRACP is similar to NPCA, however the voting rules employed in PRACP are unique. In particular, when each CSU votes on the proposition made by the $\textit{UA}_{i}$ , borda count [10] is utilized to decide the winner, and an unanimity rule is used to decide the acceptance of the proposal from $\textit{CSP}_{y}$ .

As in NPCA, the $B_{x}$ opens a call for proposals. Once all the proposals are gathered (from various CSUs), $B_{x}$ makes public those proposals to the $\textit{UA}_{i}$ and a voting process starts. Each CSU member (i.e., each $\textit{CSU}_{id}$ ) from $\textit{UA}_{i}$ ranks the proposal $X_{\textit{UA}_{i}^{t}}$ (i.e., proposal $X$ of $\textit{UA}_{i}$ received from its CSUs) in ascending order as per its own utility function $\textit{UA}_{i}(\cdot)$ . The $\textit{CSP}_{y}$ ’s offer is accepted if it is acceptable to all the CSUs. The offer acceptance mechanism is formalized in Eq. (11). It gives the number of proposals sent by $\textit{CSP}_{y}$ , and accepted by $\textit{UA}_{i}$ , and this number is same as the number of proposals supported (offered positive votes) by its CSUs.

Type of negotiation: The $\textit{CSP}_{y}$ ’s offer is accepted by the $\textit{UA}_{i}$ if it is acceptable by every member (every CSU) due to the unanimity rule employed. Hence, PRACP yields a cooperative negotiation between the CSU members of $\textit{UA}_{i}$ and $\textit{CSP}_{y}$ .

3.2.4 Agreeable by all proposals (ABAP)

The last strategy, ABAP, seeks to reach unanimity regarding the decisions made by $\textit{UA}_{i}$ and $\textit{CSP}_{y}$ . In order to determine which offer is sent to $\textit{CSP}_{y}$ , the $B_{x}$ governs an iterated building process. The aim of this iterated process is to build an offer, attribute per attribute, so that the offer sent from $\textit{UA}_{i}$ to the $\textit{CSP}_{y}$ is acceptable to every $\textit{CSU}_{id}$ . Briefly, the iterated building process is as follows.

The iterated process is initiated by $B_{x}$ . Every CSU of that $B_{x}$ is considered a member ( $\textit{CSU}_{id}\in\textit{UA}_{i}$ ) in the building process. In the beginning, the partial proposal from $\textit{UA}_{i}$ to $\textit{CSP}_{y}$ , i.e., $X_{\textit{UA}_{i}\rightarrow\textit{CSP}_{y}}^{\prime t}$ , has attributes (resource types, price, deadline, etc.) whose values are not set. The $B_{x}$ analyzes the attributes (resource types, price, deadline, etc.) that are Not Prioritized (NP) for every $\textit{CSU}_{id}$ , i.e., $\bigcap_{i=1}^{M}\textit{NP}_{\textit{CSU}_{id}}$ . These NP attributes are maximized according to the $\textit{CSP}_{y}$ ’s preferences, so as to acquire its minimum value for the interested attributes. The new attributes’ values are updated for the partial proposal.

The next attribute $j$ , which is not already selected, is selected ( $j\notin\textit{NI}_{\textit{UA}_{i}}$ ). The $\textit{CSU}_{id}$ s interested in $j$ are instructed by the $\textit{CSP}_{y}$ to submit the value $x_{\textit{CSU}_{id},j}$ which represents the offer associated with attribute $j$ to get the offer as close as possible to the aspiration levels of those $\textit{CSU}_{id}$ s. The values $x_{\textit{CSU}_{id},j}$ for each attribute $j$ , are aggregated. In order to aggregate and obtain the final value for the attribute ${j}$ , $\max$ operator is used for monotonically increasing function and $\min$ operator is used for monotonically decreasing function.

When each $\textit{CSU}_{id}$ has been satisfied by the partial offer $X_{\textit{UA}_{i}\rightarrow\textit{CSP}_{y}}^{\prime t}$ , and if there are still some attributes that have not been set, those attributes are maximized according to the $\textit{CSP}_{y}$ ’s preferences. Then, a final proposal $X_{\textit{UA}_{i}\rightarrow\textit{CSP}_{y}}^{t}$ is obtained, made public among the CSUs, and sent to the $\textit{CSP}_{y}$ . The concession strategy $s_{\textit{CSU}_{id}}(t)$ , which determines the level of demand at each negotiation round, is formalized in Eq. (12), which is inspired by [9].

$\displaystyle s_{\textit{CSU}_{id}}(t)=(1-\epsilon_{\textit{CSU}_{id}})-(1-% \epsilon_{\textit{CSU}_{id}}-\textit{RESV\_UTIL}_{\textit{CSU}_{id}})\left(% \frac{t}{\textit{UA}_{i}}\right)^{\frac{1}{\beta_{\textit{UA}_{i}}}}$ (12)

When the CSUs are instructed about a value for $j$ , each CSU communicates anonymously that value, i.e., $x_{\textit{CSU}_{id},j}$ . Taking the linear additive utility function formula, this is calculated as shown in Eq. (13).

$\displaystyle x_{\textit{CSU}_{id},j}=\mathop{\operatorname{arg\,min}}\limits_% {x\in[0,1]}(s_{\textit{CSU}_{id}}(t)-\textit{UTIL}_{\textit{CSU}_{id}}(X_{% \textit{UA}_{i}\rightarrow\textit{CSP}_{y}}^{\prime t})-w_{\textit{CSU}_{id,j}% }V_{\textit{CSU}_{id,j}}(x))$ (13)

In Eq. (13), $s_{\textit{CSU}_{id}}(t)$ is the utility demanded by the $\textit{CSU}_{id}$ at negotiation round number $t$ , and $\textit{UTIL}_{\textit{CSU}_{id}}(X_{\textit{UA}_{i}\rightarrow\textit{CSP}_{y% }}^{\prime t})$ is the utility reported by the current partial proposal, and $w_{\textit{CSU}_{id,j}}V_{\textit{CSU}_{id,j}}(x)$ is the weighted utility reported by the value demanded by the $\textit{UA}_{i}$ .

Type of negotiation: ABAP strategy is capable of guaranteeing unanimity regarding the decisions of $\textit{CSP}_{y}$ and the CSU members of $\textit{UA}_{i}$ . The guarantee of unanimity in the offer acceptance phase is obvious, because a voting mechanism based on the unanimity rule is used, resulting in cooperative negotiations.

3.3 Performance parameters

The performance parameters considered in the simulation discussed in the next section are: (i) Utility: Each player involved in the negotiation is characterised by a pre-defined utility function with respect to the negotiated variables. In other words, utility indicates the profit CSUs and CSPs have obtained after the negotiation game, employing several strategies, (ii) Negotiation time: It is the time taken to converge the negotiation process to an equilibrium, and (iii) Request satisfaction: It is the ratio of total number of jobs executed successfully by CSP to the total number of jobs submitted by the CSUs in any given time interval.

4. Results and discussion

This section describes the simulation procedure and the results.

4.1 Simulation

The proposed work is simulated on a Pentium IV machine using Genius simulator [26]. Based on the workload (i.e., dataset of negotiation scenarios), the utility, negotiation time, and the request satisfaction of the CSUs and CSPs are analyzed. The values of simulation parameters are presented in Table 3. To consider the real-world negotiation data for the initial training of datasets, the scenarios discussed in ACN [15] and SRCNS [7] are considered in this work.

Table 3
Simulation parameters

Symbol	Values
$\textit{CSU}_{id}$	$i d$ ranges from 1–250
$\textit{UA}_{i}$	$i$ ranges from 1–5 nodes
$B_{x}$	$x$ ranges from 1–5 nodes
$\textit{CSP}_{y}$	$y$ ranges from 1–5 nodes
Medium-size VMs per $\textit{CSP}_{y}$	100–500 nodes
Processor’s speed	2–4 Mghz
Processor’s memory	4–8 GB
Jobs	100 applications that require medium-size VMs
Jobs arrival rate	5–100 jobs/ms.
Job_deadline	60–600 sec./job
Buffered resource to serve dynamic requests	100–200 VMs

4.2 Simulation procedure

The simulation procedure contains the following steps:

Step 7: Step 1:
Deploy CSUs, $\textit{UA}_{i}$ , $B_{x}$ , and $\textit{CSP}_{y}$ for the scenario considered in the Cloud environment.
Step 2:
CSUs submit resource requirements to $\textit{UA}_{i}$ .
Step 3:
$\textit{UA}_{i}$ approaches $B_{x}$ .
Step 4:
$B_{x}$ searches for $\textit{CSP}_{y}$ and then recommends it to $\textit{UA}_{i}$ for job execution.
Step 5:
$\textit{UA}_{i}$ and $\textit{CSP}_{y}$ negotiate in cooperative or non-cooperative manner, mediated by $B_{x}$ .
Step 6:
$\textit{CSP}_{y}$ schedules $\textit{UA}_{i}$ jobs, executes and returns back the results.
Step 7:
Performance parameters are computed.

To assess the performance parameters with regard to utility, negotiation time, and request satisfaction, $B_{x}$ recommends negotiation strategies among the members of $\textit{UA}_{i}$ and $\textit{CSP}_{y}$ from a sample of workloads. Each CSU in $\textit{UA}_{i}$ negotiates with $\textit{CSP}_{y}$ in the range of 2 to 6 steps (on average).

The results obtained for the average utility, by using different negotiation strategies between $\textit{UA}_{i}$ and $\textit{CSP}_{y}$ , are presented in Table 4.

0pt 0pt

Table 4
Average utility (#neg.: number of negotiations)

$\textit{CSP}_{y}$ $\textit{UA}_{i}$

$\textit{CSP}_{y}$ $\textit{CSP}_{y}$ $\textit{CSP}_{y}$ $\textit{UA}_{i}$ $\textit{UA}_{i}$ $\textit{UA}_{i}$

(50 #neg.) (100 #neg.) (200 #neg.) (50 #neg.) (100 #neg.) (200 #neg.)

ACN [15] (non-cooperative) 0.13 0.23 0.32 0.39 0.41 0.45

SRCNS [7] (cooperative) 0.11 0.29 0.37 0.39 0.45 0.55

RSL (proposed) (non-cooperative) 0.09 0.25 0.26 0.29 0.35 0.43

NPCA (proposed) (cooperative) 0.17 0.35 0.46 0.23 0.37 0.47

PRACP (proposed) (cooperative) 0.22 0.37 0.58 0.49 0.56 0.69

ABAP (proposed) (cooperative) 0.34 0.44 0.64 0.57 0.65 0.83

Figure 4 shows the negotiation time between $\textit{UA}_{i}$ and $\textit{CSP}_{y}$ , employing different strategies and varied VM instances.

A comparison between the existing and proposed negotiation strategies, in terms of request satisfaction, is provided in Table 5.

Table 5
Comparative analysis among the existing and proposed negotiation strategies

Strategy Nature Unanimity level Pre-negotiation? Mediated? Jobs Successful

(How) (When) submitted negotiations

ACN [15] Non-cooperative Unanimity Information sharing Very active 100 25

SRCNS [7] Cooperative Unanimity Information sharing Very active 100 56

RSL (proposed) Non-cooperative Unilateral Minimum No 100 21

NPCA (proposed) Cooperative Plurality and majority Minimum Only coordination 100 47

PRACP (proposed) Cooperative Semi-unanimity Minimum Only coordination 100 58

ABAP (proposed) Cooperative Unanimity Information sharing Very active 100 78

Figure 4.
Negotiation time vs. negotiation strategies for various instances of VMs.

4.3 Critical analysis

	$\textit{CSP}_{y}$	$\textit{UA}_{i}$
ACN [15] (non-cooperative)	0.13	0.23	0.32	0.39	0.41	0.45
SRCNS [7] (cooperative)	0.11	0.29	0.37	0.39	0.45	0.55
RSL (proposed) (non-cooperative)	0.09	0.25	0.26	0.29	0.35	0.43
NPCA (proposed) (cooperative)	0.17	0.35	0.46	0.23	0.37	0.47
PRACP (proposed) (cooperative)	0.22	0.37	0.58	0.49	0.56	0.69
ABAP (proposed) (cooperative)	0.34	0.44	0.64	0.57	0.65	0.83

Strategy	Nature	Unanimity level	Pre-negotiation?	Mediated?	Jobs	Successful
		(How)	(When)		submitted	negotiations
ACN [15]	Non-cooperative	Unanimity	Information sharing	Very active	100	25
SRCNS [7]	Cooperative	Unanimity	Information sharing	Very active	100	56
RSL (proposed)	Non-cooperative	Unilateral	Minimum	No	100	21
NPCA (proposed)	Cooperative	Plurality and majority	Minimum	Only coordination	100	47
PRACP (proposed)	Cooperative	Semi-unanimity	Minimum	Only coordination	100	58
ABAP (proposed)	Cooperative	Unanimity	Information sharing	Very active	100	78

It is to be noted that the approach discussed in [15] is CSP centric for Cloud environment and it results in non-cooperative negotiation scenario for the CSUs. The authors in [7] have discussed a cooperative negotiation scenario for Cloud environment, with the limitation being the negotiation time and successful offers that affect the scalability of the resources in Cloud. Approaches such as ACN [15] and SRCNS [7] allow more negotiation time between the buyers and the sellers as those negotiations are either non-automated or semi-automated. With increased demand and supply for/of resources, automated negotiation reduces the negotiation time and saves from embarrassment as in the case of manual or physical negotiations.

If the performance results of the proposed negotiation strategies namely RSL, NPCA, PRACP, and ABAP are compared, they show that the CSUs using ABAP gather higher utility on average than the rest of the strategies, as shown in Table 4. RSL’s performance, in terms of utility, is less than that of [7]. This shows that a non-cooperative strategy’s utility is lower than that of a cooperative strategy even if that non-cooperative strategy is automated.

As shown in Fig. 4, RSL takes more negotiation time than ACN and SRCNS for all the VM instances considered. This shows that the negotiation time for non-cooperative automated strategy (i.e., RSL) is worse than that of a non-cooperative non-automated (or semi-automated) strategy (i.e., ACN) also. Therefore, non-cooperative automated strategy is worse than all other strategies in terms of negotiation time needed (i.e., it takes the highest amount of negotiation time). ABAP takes less negotiation time than all other negotiation strategies except SRCNS, specifically when the number of VMs is in higher ranges. As shown in Table 5, cooperative negotiation strategies lead CSUs and CSPy to win-win situations, with ABAP contributing to 78% successful negotiations.

The fact that ABAP approach is the best may be explained due to the fact that it ensures all the negotiating entities are satisfied with those offers received and sent during the negotiation rounds. As expected, all these strategies, especially the cooperative ones, get higher average utility while negotiating with an opponent, involving $\textit{UA}_{i}$ and $\textit{CSP}_{y}$ . This result supports the observation that a successful negotiating agent, considering a single negotiation with the opponent (short-term relationship), should support cooperative strategies for longer term of services (long-term relationship).

The negotiation frameworks used by the models such as [25, 24, 17] are rather inadequate, as they rely on a very simplistic acceptance semantics. A negotiating framework such as [14, 5] circumvents some of these limitations. A protocol for dealing with negotiations has been developed in that research. However, in that framework, it is unclear what the status of the negotiations’ outcome is and how an agent picks an offer to make at a given point during the negotiations. The work that is presented in this research attempts to improve the quality of the negotiated outcome, and demonstrates that cooperative negotiations may always enhance the quality of a negotiation.

5. Conclusion

This work simulated the Broker mediated SLA negotiation strategy using agents to have multiple options for selecting a CSP among several CSPs available. This negotiation system is observed to be dynamic. That is, depending on the request from the CSUs, a suitable strategy (among different negotiation strategies proposed) is applied to reduce the negotiation time. This work can be further extended to make a faster, secure, and flexible e-negotiation strategy. As part of future work, more complex market settings involving multiple issues in negotiations will be considered. Also, federated Clouds for designing negotiation strategies that analyze the behavior of the opponents and act accordingly will be analyzed.

Footnotes

Authors’ Bios

Priyanka Bharti received Master’s Degree in Software Engineering from Visvesvaraya Technological University, Belagavi, India. Her research interests include Cloud Computing, Grid Computing, and Data Mining. She has published 10 papers in International conferences and 5 papers in International journals.

Rajeev Ranjan received PhD degree from Indian Institute of Information Technology, Allahabad, India. He is involved in research of Wireless Sensor Networking, Vehicular Ad-hoc networks, E-commerce and Mobile computing. He has published about 30 papers in National and International conferences and 20 papers in National and International journals. He is a professional member of Computer Society of India (CSI), ISTE and a member of ACM.

Bhanu Prasad received Master of Technology and Ph.D. degrees, both in computer science, from Andhra University and Indian Institute of Technology Madras, respectively. He is currently serving as a Professor in the Department of Computer and Information Sciences at Florida A&M University in Tallahassee, Florida, USA. His research interests include Artificial Intelligence, Knowledge-based Systems, and Software Engineering.

References

Badshah

Ghani

Shamshirband

Aceto

and Pescapé

, Performance-based service-level agreement in cloud computing to optimise penalties and revenue, IET Communications 14 (2020), 1102–1112.

Shojaiemehr

Rahmani

A.M.

and Qader

N.N.

, Cloud computing service negotiation: A systematic review, Computer Standards & Interfaces 55 (2018), 196–206.

Shojaiemehr

Rahmani

A.M.

and Qadar

N.N.

, Automated negotiation for ensuring composite service requirements in Cloud computing, Journal of Systems Architecture 99 (2019), 201–211.

Liang

and Tseng

, Evaluation of intelligent agents in consumer-to-business e-commerce, Computer Standards & Interfaces 65 (2019), 122–131.

Pascual

Rzadca

and Trystram

, Cooperation in multiorganization scheduling, Concurrency and Computation: Practice and Experience 21 (2007), 905–921.

Foundation for intelligent physical agents, Available online at http://www.fipa.org/, Accessed on January 10, 2021.

Shyam

G.K.

Manvi

and Prasad

, Concurrent and cooperative negotiation of resources in cloud computing: A game theory based approach, Journal of Multiagent and Grid Systems, 14 (2018), 177–202.

Rocheteaua

T.W.

Lebeaua

and In

, Gradual bargaining in decentralized asset markets, Review of Economic Dynamics, Elsevier, (2020), doi: 10.1016/j.red.2020.10.008.

Lai

Sycara

and Li

, A decentralized model for multi-attribute negotiations with incomplete information and general utility functions, Rational, Robust, and Secure Negotiations in Multi-Agent Systems. Studies in Computational Intelligence 89 (2008), doi: 10.1007/978-3-540-76282-9_3.

10.

Nurmi

, Voting systems for social choice, in: Handbook of Group Decision and Negotiation. Advances in Group Decision and Negotiation, Springer, Dordrecht Kilgour

Eden

, eds, doi: 10.1007/978-90-481-9097-3_11 4 (2010), 167–182.

11.

Weerd

H.D.

Verbrugge

and Verheij

, Negotiating with other minds: The role of recursive theory of mind in negotiation with incomplete information, Autonoumous Agent Multi-Agent Systems 31 (2017), 250–287.

12.

and Xiao

, Dynamic negotiation strategy based parallel-offer in cloud computing, Journal of Computational Information Systems 11 (2015), 3691–3699.

13.

Sim

K.M.

, G-commerce, market-driven G-negotiation agents and grid resource management, IEEE Transactions on Systems, Man and Cybernetics 36 (2006), 1381–1394.

14.

Sim

K.M.

, Agent-based cloud computing, IEEE Transactions on Services Computing 5 (2012), 564–577.

15.

Kolomvatsos

Panagidi

Neokosmidis

Varoutas

and Hadjiefthymiades

, Automated concurrent negotiations: An artificial bee colony approach, Electronic Commerce Research and Applications 19 (2016), 56–69.

16.

Yeo

C.S.

Hsu

and Lai

K.R.

, Agent-based fuzzy constraint-directed negotiation for service level agreements in Cloud computing, Journal of Cluster Computing 21 (2018), 1349–1363.

17.

Lai

K.R.

and Zhu

, Data-driven behavior-based negotiation model for cyber physical social systems, IEEE Access on Visual Analysis for CPS Data 7 (2019), 83319–83331.

18.

Marey

Bentahar

Dssouli

and Mbarki

, Measuring and analyzing agents uncertainty in argumentation-based negotiation dialogue games, Expert Systems with Applications 41 (2014), 306–320.

19.

Faratin

Sierra

and Jennings

N.R.

, Negotiation decision functions for autonomous agents, Robotics and Autonomous Systems 24 (1998), 159–182.

20.

Faratin

Sierra

and Jennings

N.R.

, Using similarity criteria to make issue trade-offs in automated negotiations, Artificial Intelligence 142 (2002), 205–237.

21.

Jankoa

and Johnsonb

N.G.

, Reputation-based competitive pricing negotiation and power trading for grid-connected microgrid networks, Applied Energy 277 (2020), doi: 10.1016/j.apenergy.2020.115598.

22.

Shivani Singh

Singhrova

and Kumar

, A makespan based framework for detection of SLA violations in Cloud computing environment, in: Soft Computing: Theories and Applications. Advances in Intelligent Systems and Computing Pant

Sharma

Verma

Singla

Sikander

, eds, Springer, Singapore 1053 (2020), 503–512. doi: 10.1007/978-981-15-0751-9_47.

23.

Labidi

Mtibaa

Gaaloul

and Gargouri

, Cloud SLA negotiation and re-negotiation: An ontology-based context-aware approach, Journal of Concurrency and Computation: Practice and Experience 32 (2020), 53–61. doi: 10.1002/cpe.5315.

24.

Baarslag

Alan

A.T.

Gomer

Alam

Perera

Gerding

E.H.

and Schraefel

M.C.

, An automated negotiation agent for permission management, in: Proceedings of International Conference on Autonomous Agents and Multiagent Systems, Sao Paulo, Brazil, 2017, pp. 380–390.

25.

Shang

, Consensus of hybrid multi-agent systems with malicious nodes, IEEE Transactions on Circuits and Systems 67 (2020), 685–689.

26.

Genius simulator: http://ii.tudelft.nl/genius, Accessed on January 10, 2021.