22
1/22 Efficient Configuration of Monitoring Slices for Cloud Platform Administrators Efficient Configuration of Monitoring Slices for Cloud Platform Administrators arcio Barbosa de Carvalho, Rafael Pereira Esteves, Guilherme da Cunha Rodrigues, Clarissa Cassales Marquezan (2), Lisandro Zambenedetti Granville, Liane Margarida Rockenbach Tarouco Institute of Informatics – Federal University of Rio Grande do Sul – Brazil (2) Paluno, University of Duisburg-Essen, Germany 19 th IEEE Symposium on Computer and Communications June 24 th , 2014

Efficient Configuration of Monitoring Slices for Cloud Platform Administrators

Embed Size (px)

DESCRIPTION

Monitoring is an important issue in cloud environments because it assures that acquired cloud slices attend the user's expectations. However, these environments are multi-tenant and dynamic, requiring automation techniques to offload cloud administrators. In a previous work, we proposed FlexACMS: a framework to automate monitoring configuration related to cloud slices using multiple monitoring solutions. In this work, we enhanced FlexACMS to allow dynamic and automatic attribution of monitoring configuration tasks to servers without administrator intervention, which was not available in previous version. FlexACMS also considers the monitoring server load when attributing configuration tasks, which allows load balancing between monitoring servers. The evaluation showed that enhancements reduced FlexACMS response time up to 60% in comparison to previous version. The scalability evaluation of enhanced version demonstrated the feasibility of our approach in large scale cloud environments.

Citation preview

Page 1: Efficient Configuration of Monitoring Slices for Cloud Platform Administrators

1/22

Efficient Configuration of Monitoring Slices for Cloud Platform Administrators

Efficient Configuration of Monitoring Slices for

Cloud Platform Administrators

Marcio Barbosa de Carvalho, Rafael Pereira Esteves, Guilherme da CunhaRodrigues, Clarissa Cassales Marquezan (2), Lisandro Zambenedetti Granville,

Liane Margarida Rockenbach TaroucoInstitute of Informatics – Federal University of Rio Grande do Sul – Brazil

(2) Paluno, University of Duisburg-Essen, Germany

19th IEEE Symposium on Computer and CommunicationsJune 24th, 2014

Page 2: Efficient Configuration of Monitoring Slices for Cloud Platform Administrators

2/22

Efficient Configuration of Monitoring Slices for Cloud Platform Administrators

Outline

1 Introduction

2 FlexACMSNew architecture

3 EvaluationComparativeScalability

4 Conclusions and future work

Page 3: Efficient Configuration of Monitoring Slices for Cloud Platform Administrators

3/22

Efficient Configuration of Monitoring Slices for Cloud Platform Administrators

Introduction

Motivation

Cloud providers grant computational resources (e.g., compute,storage, network) to cloud users in form of cloud slices

Cloud slices must be closely monitored by the cloud providerto avoid wasting expensive physical resources while stillsatisfying the cloud users expectations

A single monitoring solution does not address all cloudmonitoring requirements, which imposes to cloudadministrators the utilization of multiple monitoring solutions

Once a cloud slice is created, a set of monitoring solutionsmust be configured in order to start monitoring the computingresources that form the new cloud slice

Page 4: Efficient Configuration of Monitoring Slices for Cloud Platform Administrators

4/22

Efficient Configuration of Monitoring Slices for Cloud Platform Administrators

Introduction

Monitoring Slices

Monitoring Slices

Monitoring Slices reflect all the monitoring information abouta cloud slice, which is composed by the collected values of themonitored metrics and the configuration of the monitoring solutionsthat are needed to collect these metrics.

Every cloud slice is coupled with a monitoring slice, whose goal isto detect cloud slice malfunctioning

Page 5: Efficient Configuration of Monitoring Slices for Cloud Platform Administrators

5/22

Efficient Configuration of Monitoring Slices for Cloud Platform Administrators

Introduction

Monitoring slices

CPU

Memory

Network

CPU Utilization

Memory Utilization

Network Utilization

Monitoring SlicesCloud Slices

OpenStack

Ceilometer

Page 6: Efficient Configuration of Monitoring Slices for Cloud Platform Administrators

6/22

Efficient Configuration of Monitoring Slices for Cloud Platform Administrators

Introduction

Problem

Problem

However, the lack of integration between some monitoring solutionsand cloud platforms imposes for cloud administrators to manuallyset up monitoring solutions or develop scripts to automate the mon-itoring slice creation

Page 7: Efficient Configuration of Monitoring Slices for Cloud Platform Administrators

7/22

Efficient Configuration of Monitoring Slices for Cloud Platform Administrators

FlexACMS

Approach

In a previous work1, we propose a framework to automaticallycreate monitoring slices when cloud slices are created

Flexible Automated Cloud Monitoring Slices (FlexACMS):

enables self-configurable cloud monitoring strategiesindependent of the monitoring solutions employedcreates monitoring slices automatically using monitoringsolutions that satisfy the cloud administrator’s needsfacilitates the reuse of scripts developed by cloudadministrators to automate the creation of monitoring slicessupports diverse cloud platforms and monitoring solutionscollects information from cloud platforms and triggersappropriated components to build monitoring slices

1CNSM 2013, Poster session

Page 8: Efficient Configuration of Monitoring Slices for Cloud Platform Administrators

8/22

Efficient Configuration of Monitoring Slices for Cloud Platform Administrators

FlexACMS

Approach

Now, we enhanced FlexACMS:

Dynamic and automatic attribution of configuration tasks in aMessage-Queueing fashion: enables the automatic mappingbetween type of metrics to be monitored in a monitoring sliceand the type of monitoring server to consume the request -Enhancement!Load balancing during the configuration of the monitoringslices: enables the monitoring servers to volunteer to receive atask (i.e., setting up the monitoring slice) when they havecapacity for doing so - Enhancement!

Page 9: Efficient Configuration of Monitoring Slices for Cloud Platform Administrators

9/22

Efficient Configuration of Monitoring Slices for Cloud Platform Administrators

FlexACMS

New architecture

FlexACMS Core

Gatherers

Configurators

REST

WebService

...

Cloud Monitoring

... Monitoring

Slices

Cloud Platform

... Cloud

Slices

Requests

Queue

Request

Workers

Changes

Queue

Change

WorkerChange

WorkerConfigurator

Workers

Change

WorkerChange

WorkerChange

Workers

Configurator

Queues

12

3

4

5

7

8

9

6

...

Gatherers:

responsible for collecting information from cloud platforms(e.g., @slice.ip, @slice.identifier, @slice.owner)handle peculiarities of APIs (e.g., Amazon EC2 API)send collected information to FlexACMS Core through RESTWeb service

Page 10: Efficient Configuration of Monitoring Slices for Cloud Platform Administrators

10/22

Efficient Configuration of Monitoring Slices for Cloud Platform Administrators

FlexACMS

New architecture

FlexACMS Core

Gatherers

Configurators

REST

WebService

...

Cloud Monitoring

... Monitoring

Slices

Cloud Platform

... Cloud

Slices

Requests

Queue

Request

Workers

Changes

Queue

Change

WorkerChange

WorkerConfigurator

Workers

Change

WorkerChange

WorkerChange

Workers

Configurator

Queues

12

3

4

5

7

8

9

6

...

FlexACMS Core:

responsible for processing information collected by gatherersbreaks the request into small tasks to be processed in a parallelmanner - Enhancement!detects operations performed in the platform (e.g., cloud slicecreation) (Request Workers)

Page 11: Efficient Configuration of Monitoring Slices for Cloud Platform Administrators

11/22

Efficient Configuration of Monitoring Slices for Cloud Platform Administrators

FlexACMS

New architecture

FlexACMS Core

Gatherers

Configurators

REST

WebService

...

Cloud Monitoring

... Monitoring

Slices

Cloud Platform

... Cloud

Slices

Requests

Queue

Request

Workers

Changes

Queue

Change

WorkerChange

WorkerConfigurator

Workers

Change

WorkerChange

WorkerChange

Workers

Configurator

Queues

12

3

4

5

7

8

9

6

...

FlexACMS Core (Change Workers):

evaluates the rules predefined by cloud administrators totrigger configuration taskseach Configurator has an interest and a set of conditionsputs the configuration task in the Configurator Queuepredefined by the cloud administrator - Enhancement!

Page 12: Efficient Configuration of Monitoring Slices for Cloud Platform Administrators

12/22

Efficient Configuration of Monitoring Slices for Cloud Platform Administrators

FlexACMS

New architecture

Examples of Configurator rules

Configurator Attribute ValueName: nagios host basic Interest New SliceQueue: nagios basic Condition @slice.MaaS =˜ /basic/Name: nagios cpu basic Interest New ResourceQueue: nagios basic Condition @resource.identifier =˜ /CPU/

Condition @slice.MaaS =˜ /basic/Name: nagios host plat Interest New SliceQueue: nagios platinum Condition @slice.MaaS =˜ /platinum/Name: nagios cpu plat Interest New ResourceQueue: nagios platinum Condition @resource.identifier =˜ /CPU/

Condition @slice.MaaS =˜ /platinum/

Page 13: Efficient Configuration of Monitoring Slices for Cloud Platform Administrators

13/22

Efficient Configuration of Monitoring Slices for Cloud Platform Administrators

FlexACMS

New architecture

FlexACMS Core

Gatherers

Configurators

REST

WebService

...

Cloud Monitoring

... Monitoring

Slices

Cloud Platform

... Cloud

Slices

Requests

Queue

Request

Workers

Changes

Queue

Change

WorkerChange

WorkerConfigurator

Workers

Change

WorkerChange

WorkerChange

Workers

Configurator

Queues

12

3

4

5

7

8

9

6

...

Configurator Queues:

represent pools of monitoring servers that share somecapability (e.g., monitoring solutions, monitoring purpose,server capacity) - Enhancement!

Page 14: Efficient Configuration of Monitoring Slices for Cloud Platform Administrators

14/22

Efficient Configuration of Monitoring Slices for Cloud Platform Administrators

FlexACMS

New architecture

FlexACMS Core

Gatherers

Configurators

REST

WebService

...

Cloud Monitoring

... Monitoring

Slices

Cloud Platform

... Cloud

Slices

Requests

Queue

Request

Workers

Changes

Queue

Change

WorkerChange

WorkerConfigurator

Workers

Change

WorkerChange

WorkerChange

Workers

Configurator

Queues

12

3

4

5

7

8

9

6

...

Configurator Workers:

consume configuration tasks from the Configurator Queue ifthe monitoring server has capacity to perform the task (e.g.,appropriated server load) - Enhancement!execute the Configurator and send its status and output toREST Web service for past analysis and debug

Page 15: Efficient Configuration of Monitoring Slices for Cloud Platform Administrators

15/22

Efficient Configuration of Monitoring Slices for Cloud Platform Administrators

Evaluation

Comparative

Comparative

Goal: evaluate what is the impact of the enhancements in theFlexACMS performance

Cloud platform: OpenStack

Monitoring solutions: Nagios and MRTG

Created cloud slices: from 1 to 10 cloud slices

Monitoring slices: from 1 to 10 monitoring slices, with 2 and52 metrics

Workers: 10 configurator workers

Page 16: Efficient Configuration of Monitoring Slices for Cloud Platform Administrators

16/22

Efficient Configuration of Monitoring Slices for Cloud Platform Administrators

Evaluation

Comparative

Previous vs Enhanced version

1 2 3 4 5 6 7 8 9 10

Created Cloud Slices

Re

sp

on

se

tim

e(s

)

020

40

60

80

100

120

140

27.1 %

30.9 %36.9 %

36.7 %

35.5 %

34.5 %

35.6 %

36.7 %

35.9 %37.4 %

OpenStack

FlexACMS

(a) Previous (2 metrics)

1 2 3 4 5 6 7 8 9 10

Created Cloud Slices

Re

sp

on

se

tim

e(s

)

020

40

60

80

100

120

140

25.8 %

29.1 %28.3 %

27.4 %25.6 %

23.5 %

22.3 %

20.6 %20.5 %

18.2 %

OpenStackFlexACMS

(b) Enhanced (2 metrics)

1 2 3 4 5 6 7 8 9 10

Created Cloud Slices

Re

sp

on

se

tim

e(s

)

020

40

60

80

100

120

140

43.7 %

46.1 %

45.9 %

41.5 %

41.6 %

41.3 %

40.7 %42.9 %

41.3 %

40.7 %

OpenStackFlexACMS

(c) Enhanced (52 metrics)

Figure: Previous vs Enhanced version

Page 17: Efficient Configuration of Monitoring Slices for Cloud Platform Administrators

17/22

Efficient Configuration of Monitoring Slices for Cloud Platform Administrators

Evaluation

Scalability

Scalability

Goal: evaluate the scalability in regards to response timevarying:

number of cloud slices already in place (101,102,103,104)number of new monitoring slices in a burst (10,40,70,100)number of metrics per monitoring slice (5,25,50)

Cloud platform: a gatherer generates requests similar to anactual OpenStack gatherer

Monitoring solutions: Nagios

Workers: 10 configurator workers

Page 18: Efficient Configuration of Monitoring Slices for Cloud Platform Administrators

18/22

Efficient Configuration of Monitoring Slices for Cloud Platform Administrators

Evaluation

Scalability

Scalability0

50

15

02

50

35

0

log(cloud slices already in place)

Re

sp

on

se

tim

e(s

)

New monitoring slices

10 40 70 100

1 2 3 4

(a) 5 metrics permonitoring slice

05

01

50

25

03

50

log(cloud slices already in place)

Re

sp

on

se

tim

e(s

)

New monitoring slices

10 40 70 100

1 2 3 4

(b) 25 metrics permonitoring slice

05

01

50

25

03

50

log(cloud slices already in place)

Re

sp

on

se

tim

e(s

)

New monitoring slices

10 40 70 100

1 2 3 4

(c) 50 metrics permonitoring slice

Figure: Scalability

Page 19: Efficient Configuration of Monitoring Slices for Cloud Platform Administrators

19/22

Efficient Configuration of Monitoring Slices for Cloud Platform Administrators

Conclusions and future work

Conclusions

New FlexACMS features:

Dynamic and automatic attribution of monitoring tasks topools of monitoring serversLoad balancing when attributing monitoring tasks

Performance enhancements:

reduces circa of 10% its influence in the evaluation time(Comparative)FlexACMS time reduced up to 60% (Comparative - 10 newmonitoring slices)

Page 20: Efficient Configuration of Monitoring Slices for Cloud Platform Administrators

20/22

Efficient Configuration of Monitoring Slices for Cloud Platform Administrators

Conclusions and future work

Conclusions

Scalability evaluation:

The number of cloud slices already in place does not affect theresponse timeThe number of metrics per monitoring slice and the number ofmonitoring slices that must be built affect the response time,but they do not affect the response time at same rate ofscenario growth

The automatic configuration of multiple monitoring solutionsfor cloud computing is feasible

FlexACMS can be used to help cloud administrators toachieve their monitoring needs using monitoring solutions thatare not integrated to cloud platforms

Page 21: Efficient Configuration of Monitoring Slices for Cloud Platform Administrators

21/22

Efficient Configuration of Monitoring Slices for Cloud Platform Administrators

Conclusions and future work

Future work

Future work:

Complete the cloud slice life-cycle automation (reconfigurationand destruction of monitoring slices)Improve load balancing strategiesInvestigate FlexACMS feasibility for PaaS and SaaS cloudmodels

Page 22: Efficient Configuration of Monitoring Slices for Cloud Platform Administrators

22/22

Efficient Configuration of Monitoring Slices for Cloud Platform Administrators

Conclusions and future work

Acknowledgments

Thank you for your attention!

Questions?

[email protected]