cloud to address oscillatory peaks of workload. -â In this context ... to private clouds and a provisioning strategy .... more VMs, it uses the public cloud to host the.
Provisioning and Resource Alloca2on for Green Clouds
Guilherme Arthur Geronimo, Jorge Werner, Carlos Becker Westphall, Carla Merkle Westphall, Leonardo Defen2 Networks and Management Laboratory Federal University of Santa Catarina JANUARY 29TH, SEVILLE, SPAIN
IARIA ICN 2013 -‐ GLOBENET 2013
1
Summary 1 -‐ Introduc2on 2 – State of the Art 3 – Model 4 – Proposal (Results) 5 -‐ Conclusions 6 – Future Works 7 – Some References JANUARY 29TH, SEVILLE, SPAIN
IARIA ICN 2013 -‐ GLOBENET 2013
2
(J. Werner, G. A. Geronimo, C. B. Westphall et al. CLEI EJ 2012)
JANUARY 29TH, SEVILLE, SPAIN
IARIA ICN 2013 -‐ GLOBENET 2013
3
1 Introduc2on -‐ The aim of Green Cloud Compu2ng is to achieve a balance between the resource consump2on and quality of service. -‐ Dynamic provisioning and alloca2on strategies are needed to regulate the internal se_ngs of the cloud to address oscillatory peaks of workload. -‐ In this context, we propose strategies to op2mize the use of the cloud resources without decreasing the availability. JANUARY 29TH, SEVILLE, SPAIN
IARIA ICN 2013 -‐ GLOBENET 2013
4
1 Introduc2on -‐ This work introduces two hybrid strategies based on a distributed system management model, describes the base strategies, opera2on principles, tests, and presents the results. -‐ We extended CloudSim to simulate the organiza2on model upon which we were based and to implement the strategies, using this improved version to validate our solu2on. JANUARY 29TH, SEVILLE, SPAIN
IARIA ICN 2013 -‐ GLOBENET 2013
5
1 Introduc2on -‐ We aim to propose an alloca2on strategy to private clouds and a provisioning strategy for Green Clouds, which suits the oscillatory workload and unexpected peaks. -‐ We will focus on finding a solu2on that consumes low power and generates acceptable request losses. JANUARY 29TH, SEVILLE, SPAIN
IARIA ICN 2013 -‐ GLOBENET 2013
6
1 Introduc2on Organiza2on of this presenta2on: -‐ 2. comments the state of the art based in some references; -‐ 3. explains under which model the strategies were based; -‐ 4. presents the proposal, tests, and the results; -‐ 5. concludes this presenta2on; and -‐ 6. addresses some future works. JANUARY 29TH, SEVILLE, SPAIN
IARIA ICN 2013 -‐ GLOBENET 2013
7
2 State of the Art -‐ The reference [8] uses a Dynamic Voltage Frequency Scaling (DVFS) strategy to decrease the energy consump2on in PMs used as virtualiza2on hosts. -‐ It adapts the clock frequency of the CPUs with the real usage of the PMs. It decreases the frequency in idle nodes and increases when is needed. -‐ JANUARY 29TH, SEVILLE, SPAIN
IARIA ICN 2013 -‐ GLOBENET 2013
8
2 State of the Art -‐ The workload balance strategy for clusters in [9], tries to achieve a lower energy consump2on unbalancing the cluster workload, genera2ng idle nodes and turning off them. -‐ The paper [10] tries to decrease the hos2ng costs in public and/or federated clouds using the costs and fines in contracts as metrics to bejer allocate the resources. JANUARY 29TH, SEVILLE, SPAIN
IARIA ICN 2013 -‐ GLOBENET 2013
9
3 Model -‐ Management Systems based on the Organiza2on Theory, providing the means to describe why / how elements of the cloud environment should behave to achieve global system objec2ves, which are (among others): op2mum performance, reduce opera2ng costs, appointment of dependence, service level agreements, and energy efficiency. JANUARY 29TH, SEVILLE, SPAIN
IARIA ICN 2013 -‐ GLOBENET 2013
10
3 Model -‐ Managing Cloud through the principles of the Organiza2on Theory provides the possibility for an automa2c con figura2on management system, since adding a new element (e.g., V i r t u a l M a c h i n e s , P h y s i c a l M a c h i n e s , Uninterrupted Power Supply, Air Condi2oning) is just a majer of adding a new service on the Management Group. JANUARY 29TH, SEVILLE, SPAIN
IARIA ICN 2013 -‐ GLOBENET 2013
11
3 Model -‐ The proposed strategies are based on a proac2ve management of Clouds, which is based on the distribu2on of responsibili2es in holes, as seen in next figure. The responsibility of management of the cloud elements is distributed among several agents, separated in holes, and each agent controls individually, a Cloud element that suits him. JANUARY 29TH, SEVILLE, SPAIN
IARIA ICN 2013 -‐ GLOBENET 2013
12
3 Model
JANUARY 29TH, SEVILLE, SPAIN
IARIA ICN 2013 -‐ GLOBENET 2013
13
4 Proposal -‐ For the conscious resource provisioning of the data center, we propose a hybrid strategy that uses public cloud as an external resource used to mi2gate probable Service level Agreements (SLAs) breaches due to unexpected workload peaks. JANUARY 29TH, SEVILLE, SPAIN
IARIA ICN 2013 -‐ GLOBENET 2013
14
4 Proposal -‐ In parallel, to the op2mal use of local resources, we propose a strategy of dynamic reconfigura2on of the VMs ajributes, allocated in the data center. -‐ Given the distributed model presented in the previous sec2on, we use the Cloud simula2on tool CloudSim to simulate the university data center environment. JANUARY 29TH, SEVILLE, SPAIN
IARIA ICN 2013 -‐ GLOBENET 2013
15
4 Proposal (Alloca2on) -‐ The resource alloca2on strategy is a proposal that introduces a composi2on of two other approaches: (1) the migra2on of VMs, which aims to focus on the processing of cloud, and (2) the Dynamic Reconfigura2on of VMs, which aims to relocate dynamically the resources used by the VMs. JANUARY 29TH, SEVILLE, SPAIN
IARIA ICN 2013 -‐ GLOBENET 2013
16
4 Proposal (Alloca2on)
(J. Werner, G. A. Geronimo, C. B. Westphall et al. CLEI EJ 2012) JANUARY 29TH, SEVILLE, SPAIN
IARIA ICN 2013 -‐ GLOBENET 2013
17
4 Proposal (Alloca2on)
Parameter
Value
VM – Image size
1GB
VM -‐ RAM
256MB
PM -‐ Engine
Xen
PM -‐ RAM
8GB
PM -‐ Frequency
3.0GHZ
PM -‐ Cores
2
PROPOSED SCENARIO CHARACTERISTCS (J. Werner, G. A. Geronimo, C. B. Westphall et al. CLEI EJ 2012)
JANUARY 29TH, SEVLLE, SPAIN
IARIA ICN 2013 -‐ GLOBENET 2013
18
4 Proposal (Alloca2on)
(J. Werner, G. A. Geronimo, C. B. Westphall et al. CLEI EJ 2012) JANUARY 29TH, SEVLLE, SPAIN
IARIA ICN 2013 -‐ GLOBENET 2013
19
4 Proposal (Alloca2on) 1) VMs Migra+on Strategy: This strategy aims to reduce power consump2on by disabling the idle PMs of the Cloud. To induce idleness in the PMs, the VMs are migrated and concentrated in a few PMs. 2) VMs Dynamic Reconfigura+on Strategy: It adjusts the parameters of the VM, without migra2ng it or turning it off. For example, we can increase or decrease the parameters of CPU and memory allocated.
JANUARY 29TH, SEVILLE, SPAIN
IARIA ICN 2013 -‐ GLOBENET 2013
20
4 Proposal (Alloca2on) Four scenarios were simulated in order to seek the compara2ve analysis between ordinary cloud (Scenario 1), the exis2ng methods (Scenarios: 2 and 3), and the proposed approach (Scenario 4). Those were: No strategies; Migra2ng VMs Strategy; Reconfiguring the VMs Strategy; Reconfiguring and migra2ng VMs Strategy. JANUARY 29TH, SEVILLE, SPAIN
IARIA ICN 2013 -‐ GLOBENET 2013
21
4 Proposal (Alloca2on) Scenario
Reconf. Strategy Migrat. Strategy ConsumpMon
Timeout
1
No
No
-‐
-‐
2
No
Yes
84.3 %
8.0 %
3
Yes
No
0.4 %
-‐
4
Yes
Yes
87.2 %
7.3 %
Table I (RESULTS OF ALLOCATION’S SCENARIOS) shows the results of the simula2ons. It tells what strategies were used in each scenario and what percentage (approximate) reduc2on was obtained, compared to the scenario without strategies. JANUARY 29TH, SEVILLE, SPAIN
IARIA ICN 2013 -‐ GLOBENET 2013
22
4 Proposal (Provisioning) -‐ The hybrid strategy is based on the merge of two other strategies, the On-‐Demand strategy (OD) and the Spare Resources strategy (SR). -‐ It aims to present a power consump2on lower than the SR strategy and a wider availability than the OD strategy. JANUARY 29TH, SEVILLE, SPAIN
IARIA ICN 2013 -‐ GLOBENET 2013
23
4 Proposal (Provisioning) 1) On-‐Demand Strategy: The principle of OD strategy is to ac2vate the resources when they are needed. In our case, when a service reaches a satura2on threshold, new VMs would be instan2ated. When there is no more space to instan2ate new VMs, new PMs would be ac2vated to host the new VMs. The opposite also applies; when a threshold of idleness is reached, the idle VMs and PMs are disabled.
JANUARY 29TH, SEVILLE, SPAIN
IARIA ICN 2013 -‐ GLOBENET 2013
24
4 Proposal (Provisioning) On-‐Demand Strategy proved to be very efficient energe2cally, since it maintains a minimum amount of ac2ve resources. But, it has been shown ineffec2ve in scenarios that had sudden spikes in demand, because the process to ac2vate the resource took too much 2me, and the requests ended up genera2ng losses.
JANUARY 29TH, SEVILLE, SPAIN
IARIA ICN 2013 -‐ GLOBENET 2013
25
4 Proposal (Provisioning) Spare Resource Strategy: To mi2gate the problem of requests 2meouts, originated by a long ac2va2on 2me of resources, we adopt the strategy SR, whose principle is reserve idle resources ready to be used. In our case, there was always one idle VM ready to process the incoming requests and one idle PM ready to instan2ate new VMs. If these resources were used, they were no longer considered idle, and new idle resources were ac2vated. JANUARY 29TH, SEVILLE, SPAIN
IARIA ICN 2013 -‐ GLOBENET 2013
26
4 Proposal (Provisioning) The Spare Resource strategy has been shown effec2ve in remedying unexpected peak demands, but it showed the same behavior of OD strategy in cases where demand rose very rapidly; in other words, the idle feature was not enough to process the demand. Another nega2ve point was the energy consump2on; since they always had an ac2ve and idle resource, the consump2on has been greater than the OD strategy. JANUARY 29TH, SEVILLE, SPAIN
IARIA ICN 2013 -‐ GLOBENET 2013
27
4 Proposal (Provisioning) 3) Hybrid Strategy: Seeking the merger of the strengths of the previous strategies and mi2ga2ng its shortcomings, we propose a hybrid strategy. This strategy aims to reduce the energy consump2on on private cloud and reduce the breakage of SLA’s service in general. JANUARY 29TH, SEVILLE, SPAIN
IARIA ICN 2013 -‐ GLOBENET 2013
28
4 Proposal (Provisioning) As shown in next figure, the cloud enables the VMs when the service in ques2on reaches its satura2on threshold, just as the OD strategy. When more PMs space is unable to allocate more VMs, it uses the public cloud to host the new VMs while the PM is passing through the ac2va2on process. This is to fulfill requests that would be lost during the ac2va2on process. JANUARY 29TH, SEVILLE, SPAIN
IARIA ICN 2013 -‐ GLOBENET 2013
29
4 Proposal (Provisioning)
(J. Werner, G. A. Geronimo, C. B. Westphall et al. CLEI EJ 2012) JANUARY 29TH, SEVILLE, SPAIN
IARIA ICN 2013 -‐ GLOBENET 2013
30
4 Proposal (Provisioning) 4) Tests Results: As previously men2oned, we performed some modifica2ons to the CloudSim code, in order to enable the simula2on of scenarios. Before we started the simula2on, we defined some variables for the scenario, such as the satura2on threshold and idleness, for example. Some of these variables are shown in Table II. JANUARY 29TH, SEVILLE, SPAIN
IARIA ICN 2013 -‐ GLOBENET 2013
31
4 Proposal (Provisioning) Variable
Value
Satura2on Threshold (Load 1 minute)
1.0
Idleness Threshold (Load 1 minute)
0.1
Ac2va2on VM 2me (seconds)
10
Ac2va2on PM 2me (seconds)
120
Size of Request (MI)
1000 to 2000
Number of PMs
8
Maximum number of VMs per PMs
5
SLA 2meout threshold (seconds)
10
Table II (SIMULATION’S VARIABLES) JANUARY 29TH, SEVILLE, SPAIN
IARIA ICN 2013 -‐ GLOBENET 2013
32
4 Proposal (Provisioning) To get an overview of how each strategy would behave in different scenarios, we ran a series of tests which varied (1) the amount of requests and (2) the size of the requests. To maintain the defined request distribu2on (explained in the beginning of Sec2on 3), we used mul2pliers to increase the requests. Those mul2pliers started from 2 to 20 in steps of 2 (2, 4, 6, etc.). JANUARY 29TH, SEVILLE, SPAIN
IARIA ICN 2013 -‐ GLOBENET 2013
33
4 Proposal (Provisioning) The size of the requests ranged from 1000 to 2000 MI (Millions Instruc2ons), in steps of 100 (1000, 1100, 1200, etc.). This way, it performed a series of 100 simula2ons. This test evaluated the power consump2on of the private cloud and the total number of 2meouts. Next figures demonstrates 100 simula2ons in two images, the percentage of 2meouts (top) and the energy consump2on of the private cloud (bojom) are plojed. JANUARY 29TH, SEVILLE, SPAIN
IARIA ICN 2013 -‐ GLOBENET 2013
34
4 Proposal (Provisioning)
JANUARY 29TH, SEVILLE, SPAIN
IARIA ICN 2013 -‐ GLOBENET 2013
35
4 Proposal (Provisioning)
JANUARY 29TH, SEVILLE, SPAIN
IARIA ICN 2013 -‐ GLOBENET 2013
36
4 Proposal (Provisioning) Table III shows the results obtained in the ”worst case scenario”, by defini2on, with the mul2plier equal to 20 and the request size equal to 2000 MI. Regarding the results in Table III, it took the Hybrid Strategy as a basis of comparison. In this case, the values listed are for hybrid strategy. For example, the hybrid strategy presented 3% less requisi2on 2meouts than the OD strategy. JANUARY 29TH, SEVILLE, SPAIN
IARIA ICN 2013 -‐ GLOBENET 2013
37
4 Proposal (Provisioning)
Timeouts Consump2on
On demand
Spare
-‐ 3 %
15 %
% -‐ 18
-‐ 52 %
Table III (HYBRID STRATEGY COMPARED TO THE OTHER STRATEGIES) JANUARY 29TH, SEVILLE, SPAIN
IARIA ICN 2013 -‐ GLOBENET 2013
38
5 Conclusions Based on what was presented in the previous sec2ons, and considering the objec2ves set at the beginning of this paper, we consider the intended goal was achieved. Two strategies for alloca2on and provisioning, were proposed; both aimed at op2mizing the energy resource without sacrificing service availability. JANUARY 29TH, SEVILLE, SPAIN
IARIA ICN 2013 -‐ GLOBENET 2013
39
5 Conclusions The alloca2on strategy in private clouds, compared to a normal cloud, demonstrated a 87% reduc2on in energy consump2on. It was observed that this strategy is not effec2ve in scenarios where the workload is oscilla2ng. That’s because it ends up genera2ng too much unnecessary reconfigura2ons and migra2ons. Despite this, it s2ll shows a significant gain in energy savings when compared to a cloud without any strategy deployed. JANUARY 29TH, SEVILLE, SPAIN
IARIA ICN 2013 -‐ GLOBENET 2013
40
5 Conclusions The hybrid strategy for provisioning in green clouds, demonstrated a 52% consump2on reduc2on over the SR strategy, and a 2meout rate 3% lower than the OD strategy. Thus, we conclude that the use of this strategy is recommended in situa2ons where the ac2va2on 2me of the resource is expensive for the health of SLA. We also iden2fied that using this is not recommended when the public cloud should be used sparingly due to their course or other factors. JANUARY 29TH, SEVILLE, SPAIN
IARIA ICN 2013 -‐ GLOBENET 2013
41
6 Future Works As future work, we aim at adding the strategy of Dynamic Reconfigura2on of VMs in public clouds. This procedure was not adopted because, during the development of this work, this feature was not a market reality. We also intend to invest in new simula2ons of the cloud extending the variables (such as DVFS and UPS) and, if possible, explore some ar2ficial intelligence techniques such as Bayesian networks. JANUARY 29TH, SEVILLE, SPAIN
IARIA ICN 2013 -‐ GLOBENET 2013
42
6 Future Works Our PCMONS (Private Cloud Monitoring System), open-‐ source solu2ons for cloud monitoring and management, also will help to manage green clouds, by automa2ng the instan2a2on of new resource usage. We foresee, in opposi2on to unexpected peaks scenarios, work with cloud management based on prior knowledge of the behavior of hosted services. It is believed to be necessary to develop a descrip2on language to represent the structure and behavior of a service, enabling the exchange of informa2on between applica2ons for planning, provisioning, and managing the cloud. JANUARY 29TH, SEVILLE, SPAIN
IARIA ICN 2013 -‐ GLOBENET 2013
43
7 Some References -‐ J. Werner, G. A. Geronimo, C. B. Westphall, F. L. Koch, R. R. Freitas, and C. M. Westphall, “Environment, services and network management for green clouds,” CLEI Electronic Journal, vol. 15, no. 2, p. 2, 2012. -‐ R. Buyya, A. Beloglazov, and J. Abawajy, “Energy-‐ Efficient management of data center resources for cloud compu2ng: A vision, architectural elements, and open challenges,” in Proceedings of the 2010 Interna+onal Conference on Parallel and Distributed Processing Techniques and Applica+ons (PDPTA 2010), Las Vegas, USA, July 12, vol. 15, 2010.
JANUARY 29TH, SEVILLE, SPAIN
IARIA ICN 2013 -‐ GLOBENET 2013
44
7 Some References -‐ R. Buyya, “Modeling and simula2on of scalable cloud compu2ng environments and the cloudsim toolkit: Challenges and opportuni2es,” in HPCS 2009. Interna+onal Conference on. IEEE, 2009, pp. 1–11. -‐ G. von Laszewski, L. Wang, A. Younge, and X. He, “Power aware scheduling of virtual machines in dvfs enabled clusters,” in Cluster Compu+ng and Workshops, 2009. CLUSTER ’09. IEEE Interna+onal Conference on, 2009, pp. 1–10.
JANUARY 29TH, SEVILLE, SPAIN
IARIA ICN 2013 -‐ GLOBENET 2013
45