Keywords

1 Introduction

Cloud computing (CC) has been a novel paradigm in the field of Information and Communication Technology which offers a different style of computing where users do not own their computing resources but rent and pay for them. NIST [1] defines it as “a model for enabling ubiquitous, convenient, on-demand network access to a shared pool of configurable computing resources (e.g., networks, servers, storage, applications and services) that can be rapidly provisioned and released with minimal management effort or service provider interaction.” Hence, resources are provisioned/released to/from the user on demand over the network and payment is made accordingly. This new paradigm has become very popular in recent times.

Out of various challenges faced by Cloud such as security, availability, performance etc., the issue of energy consumption has attracted the attention of many researchers. The issue is being addressed in many facets such as VM Consolidation, Load balancing, VM Migration etc. In this research, we aim to address the issue of energy consumption by optimally selecting (minimum number of) VMs to be migrated from one host to another in such a way that a minimum number of the host are to be kept on, resulting in a reduction in electrical power. Authors of [2] have addressed the issue of minimizing migration but it works fine for the cases where only one VM is to be migrated to consolidate the system. But, in real situation, where a number of hosts in a data center is huge and the number of VMs in a host is large, we need to migrate multiple VMs to stabilize the system. In such scenario, the existing method does not give an optimum solution. Hence, we find the scope of improvement in the existing method and we propose a modified version of minimization of migration algorithm to handle the cases of multiple VM selection.

Proper utilization of hosts leads into keeping an optimum number of hosts alive and rest to be idle resulting in saving in power consumption. VM consolidation is a process of appropriately distributing tasks on VMs and mapping or migrating VMs on or across hosts. The process of VM consolidation is divided among four phases [3] viz. (a) Overloaded host detection (b) VM selection from overloaded host (c) VM placement for selected VMs (d) Underloaded host detection. In this research, we propose a modified minimization of migration algorithm for VM selection (as mentioned in b) for optimally selecting the combination of VMs to be migrated. The Efficient method to select the best combination of VMs would result into significant improvement in power consumption.

The overall paper has been organized as follows. Section 2 discusses related work, followed by our proposed scheme in Sect. 3. Experimentations and results are described in Sect. 4. The overall research has been concluded in Sect. 5 followed by a list of references.

2 Related Work

Anton et al. [2] have been one of the few researchers who worked in the domain of achieving energy efficiency by proper resource allocation. Along with defining an architectural framework and survey of research in energy-efficient computing, the authors have proposed energy-efficient resource allocation policies and scheduling algorithms and claimed significant cost saving and improvement of energy efficiency. Minimization of Migrations (MM) is one of the key contributions by the authors in the direction of selecting a combination of VMs from an overloaded host for subsequent migration. The algorithm covers two aspects of migration viz. (a) single VM selection and (b) multiple VMs selection. The algorithm works efficiently for (a) but there seems to be a scope of improvement in (b) which is one of the main sources of motivation for this research. In [3], authors identified dynamic consolidation of VMs using live migration and switching idle nodes to sleep mode, as the mean to optimize resource usage and reduce energy consumption. Further, the authors propose heuristics for dynamic consolidation of VMs to claim a reduction in energy consumption and maintenance of SLA. Different statistical methods such as (a) Median Absolute Deviation (MAD) (b) Interquartile Range (IQR) (c) Local Regression (LR) and (d) Robust Local Regression (LRR), have been used for detection of the overloaded host. Various methods for VM selection such as (a) Minimum Migration Time (MMT) (b) Random Choice (RC) and (c) Maximum Correlation (MC) have been discussed. Power Aware Best Fit Decreasing (PABFD) policy has been proposed for VM placement. A research paper [4], aimed to maximize utilization and minimize cost by proper management of resources and allocation strategies. Authors proposed performance analysis based resource allocation scheme which follows the best fit strategy for efficient VM allocation. The execution time of CPU and performance of memory are considered as two important factors with a flexibility of assigning weight to resource. Subsequently, authors claimed improvement in resource utilization without compromising the allocation time. In the survey paper, [5] identified resource utilization and energy consumption as two important factors to be considered in the process of VM consolidation. To address the factors, authors have suggested having intelligent workload placement and relocation techniques. Authors have exhaustively described terminology involved such as virtualization, VM migration, VM consolidation (static, dynamic) etc. Authors of [6] proposed a QoS-aware VM consolidation approach (based on resource utilization history of VMs) to improve QoS metrics and energy consumption. Initially, the proposed algorithm detects an overloaded host. Then, VMs are selected for migration from these overloaded hosts. Next, detect underloaded hosts and select all VMs from them for migration. At the end, a new placement is searched for VMs to be migrated. Authors claimed a reduction in energy consumption and SLA violation.

3 Proposed Algorithm: Modified Minimization of Migration

The Minimization of Migration (MM) [2] policy selects the minimum number of VMs to migrate from overloaded hosts to reduce the CPU utilization under the upper utilization threshold. The algorithm tries to selects best possible VM combination that satisfies two conditions viz. (a) keep the number of VMs to be migrated as minimum as possible (b) keep the host utilization under and nearer to the upper threshold, after migration. The algorithm works well where the resultant VM is single. But, there are cases where single VM selection is not sufficient for bringing host utilization under the upper threshold. In such scenarios, MM algorithm selects one VM (by default) with the highest utilization and proceeds to make best possible combination by selecting next VM(s) for bringing the host utilization under the upper threshold. The main issue with the MM algorithm is that it does not provide the optimum combination of VMs which brings the host utilization nearer to the upper threshold. To address this issue, for the cases of multiple VM selection, we have modified existing MM algorithm to maximize utilization of host while keeping the number of VMs to be migrated minimal. We start the selection procedure for a number of VM to be migrated equal to 2 and check whether host utilization (after migration) goes below upper threshold or not. If not, the selection of VM proceeds for all nCr combinations incrementally, where n is total number of VMs in a host and r ranges from 2 to n.

To understand the scenario and to support our claim, we have taken following example values. It is worth noting here that we have taken the example of the scenario where more than one VMs are to be selected for migration. Current Host Utilization hUtil = 85%, Upper Threshold THRESH_UP = 70% (0.7), Total number of VMs in a host = 4, Utilization of all VM in the host, vmUtil = {13, 11, 9, 5} (sorted in decreasing order based on utilization).

Existing Method (MM) [2]:

MM selects first VM (by default) irrespective of its utilization. That means, in this case, first VM with utilization = 13 is selected. Then, the method checks for all other VMs where the difference between (a) and (b) remains minimum and resulted in utilization under to THRESH_UP, where (a) hUtil - THRESH_UP = 85 − 70 = 15 (b) Addition of utilization of all VMs selected. So for the example values given, a combination of {13, 5} shall be selected as the total utilization is 18 where the difference is 3 [(13 + 5) − (85 − 70)] which is minimum and resultant utilization will nearer and under THRESH_UP [85 − (13 + 5) = 67, 67 < 70].

Our Proposal (Modified Minimization of Migration – M3):

In our proposal, unlike existing method, we do not select the first VM (by default). On the contrary, we look for all the possible combinations of VMs ranging from 2 to n (until we achieve host utilization below THRESH_UP). In above example, possible combinations are as under. For VM = 2, VM combinations are {13, 11}, {13, 9}, {13, 5}, {11, 9}, {11, 5}, {9, 5}, For VM = 3, VM combinations are {13, 11, 9}, {13, 11, 5}, {11, 9, 5}, For VM = n = 4, VM combination is {13, 11, 9, 5}.

In our proposal, the algorithm selects a pair {11, 5} because while doing so difference between (a) and (b) remains minimum and nearer to THRESH_UP. (a) hUtil - THRESH_UP = 85 − 70 = 15 (b) Addition of utilization of all VMs selected (11 + 5 = 16). So, for this value, the difference is 1 (against 3 in existing method) and hUtil will be 69 (against 67 in existing method). For these case example, as we have identified a pair of VM from the first option only, hence the algorithm would not go for checking other combinations with VM = 3 or VM = 4.

figure a

4 Experimentation and Results

In this section, we discuss the analysis of existing Minimization of Migration algorithm along with our proposal, Modified Minimization of Migration. For our experimentation, we have taken dataset as mentioned in Table 1. We have simulated a data center consisting of 4 hosts with 9 to 10 VMs on each. Each host is modeled to have 2 CPU cores with MIPS capacity of either 2660 or 1860 per core. Each VM requires 1 CPU core with utilization requirement as mentioned in Table 1. The CloudSim toolkit [7] has been used as a simulation platform because it supports modeling of on-demand virtualization-enabled resource and application management with a varying workload.

Table 1. Data set
Full size table

For the described simulation setup, we have carried out series of experimentation and generated the results as mentioned in Table 2 and Fig. 1.

Table 2. Results
Full size table
Fig. 1.
figure 1

Host utilization after migration

Full size image

In our experimentation, the value of the upper threshold is taken as 0.7 which can be calculated dynamically using any of the available statistical methods such as Median Absolute Deviation, Interqurtile Range, Local regression etc. As can be seen from Table 2 that for the case of Host 1 and Host 3, there exists a solution with only one VM selection. So, for both these hosts, both the algorithm results in the same outcome. Hence, these results are not considered for comparison. But, for Host 2 and Host 4, the results are compared in Fig. 1. As shown in the figure, M3 gives resultant host utilization (after migration) nearer to the upper threshold value (here, 0.7). Hence we claim more optimized utilization of available hosts, without compromising the number of VMs to be migrated. Figure 2, the difference between the upper threshold and host utilization after migration has been shown. Lower the difference (nearer to the upper threshold) better the host utilization.

Fig. 2.
figure 2

Difference (threshold- host utilization after migration)

Full size image

Further, to understand the impact of migration on power consumption, we analyzed the same empirically. Table 3 and Fig. 3 summarize the same. As can be seen from the table and figure, the resultant power gets reduced after migration in both the methods. But, there is a negligible increase in power consumed by M3 as compared to MM. So, we may conclude that without significant increase in power consumption, one may achieve higher utilization using M3.

Table 3. Power consumption
Full size table
Fig. 3.
figure 3

Power consumption

Full size image

5 Conclusion

The issues of energy consumption and carbon emission by Cloud data center have attracted the attention of many researchers in recent time. In this research, we have modified an existing algorithm which claimed to minimize the VM migration while keeping the hosts optimally utilized. Our contribution through this research is to identify the best combination of VMs to be migrated in such a way that (a) number of VMs to be migrated is minimizes and (b) host utilization after migration remains below and nearer to the upper threshold. Our results show that the proposed algorithm provides an optimum solution as compared to the existing method.