Professional Documents
Culture Documents
Technical Overview of BonFIRE Facility Architecture entry points, Available infrastructure capacity, Setting up experiments Florian Schreiner, Fraunhofer Institute FOKUS Kostas Kavvousanakis, EPCC W: www.bonfire-project.eu E: bonfire@bonfire-project.eu
Agenda BonFIREs Experimental Facility Architecture BonFIREs Infrastructure, available Resources BonFIREs Experiment Support Examples of current BonFIRE Experiments Demo of BonFIRE Experiment Setup and Monitoring Procedure
Fraunhofer Institute FOKUS, ATOS Origin and other members of the BonFIRE consortium 2010
BonFIRE 2
Fraunhofer Institute FOKUS, ATOS Origin and other members of the BonFIRE consortium 2010
BonFIRE 3
Fraunhofer Institute FOKUS, ATOS Origin and other members of the BonFIRE consortium 2010
BonFIRE 4
Monitoring dashboard
Architecture
LDAP
Identity Server
(Used by Portal, Experiment Manager, Broker and Testbeds)
Monitoring
Read/ Write
Message Queue
SSH
Monitoring GUI
Monitoring API
VM (Monitoring Aggregator)
Fraunhofer Institute FOKUS, ATOS Origin and other members of the BonFIRE consortium 2010
BonFIRE 5
Fraunhofer Institute FOKUS, ATOS Origin and other members of the BonFIRE consortium 2010
BonFIRE 6
Fraunhofer Institute FOKUS, ATOS Origin and other members of the BonFIRE consortium 2010
BonFIRE 7
Architecture Principles
It must always be possible to include testbeds over which BonFIRE has no control in a BonFIRE federation if the appropriate adapters are written and deployed into the BonFIRE system. It should not be necessary to force a testbed to change its functionality in order to be included in BonFIRE. Always provide APIs to BonFIRE functionality, in addition to any BonFIRE graphical user interfaces (GUIs). Allow experimenters full access to the specific functionality of particular testbeds; i.e. functionality that belongs only to one of the clouds and is not available through the BonFIRE API. Allow federated functionality to exclude specific functionality of a testbed if this makes common tasks easier to achieve. Support incremental adoption of the BonFIRE system by experimenters. Support declarative specification of experiments as far as possible.
Fraunhofer Institute FOKUS, ATOS Origin and other members of the BonFIRE consortium 2010
BonFIRE 8
Agenda BonFIREs Experimental Facility Architecture BonFIREs Infrastructure, available Resources BonFIREs Experiment Support Examples of current BonFIRE Experiments Demo of BonFIRE Experiment Setup and Monitoring Procedure
Fraunhofer Institute FOKUS, ATOS Origin and other members of the BonFIRE consortium 2010
BonFIRE 9
Fraunhofer Institute FOKUS, ATOS Origin and other members of the BonFIRE consortium 2010
To ensure the operation of the experimental facility Appropriate planning and control of the evolution of the permanently provided infrastructure Provision of on-request resources for large scale experiments in a well defined environment Support monitoring of experiments under well defined conditions Allowing repetition of experiments under similar conditions
11
Fraunhofer Institute FOKUS, ATOS Origin and other members of the BonFIRE consortium 2010
BonFIRE 11
Component 1 The Core Infrastructure Provided permanently by the BonFIRE consortium members Used as general purpose platform for medium scaled experiments Architecture and software provided by BonFIRE
12
Fraunhofer Institute FOKUS, ATOS Origin and other members of the BonFIRE consortium 2010
BonFIRE 12
Component 2 Infrastructure/Resources on request On-request provisioning Not exclusively for BonFIRE Availability for limited times for large scale testing
13
Fraunhofer Institute FOKUS, ATOS Origin and other members of the BonFIRE consortium 2010
BonFIRE 13
Component 3 Experiment-owned infrastructure Experiments may have own small scale testbeds BonFIRE as extension to these testbeds or vice versa Effort needed to allow the connection of experiments infrastructure and the BonFIRE infrastructure
14
Fraunhofer Institute FOKUS, ATOS Origin and other members of the BonFIRE consortium 2010
BonFIRE 14
BonFIRE 15
Agenda BonFIREs Experimental Facility Architecture BonFIREs Infrastructure, available Resources BonFIREs Experiment Support Examples of current BonFIRE Experiments Demo of BonFIRE Experiment Setup and Monitoring Procedure
16
Fraunhofer Institute FOKUS, ATOS Origin and other members of the BonFIRE consortium 2010
BonFIRE 16
Fraunhofer Institute FOKUS, ATOS Origin and other members of the BonFIRE consortium 2010
Help experimenters to install software/applications and their own tools into the BonFIRE facility Support migration of an experiment from the experimenter's in-house testbed into BonFIREs facility Help experimenters to find appropriate third party tools and install these tools into the BonFIREs facility Clearly define the services a customer may ask for
Enable customers to understand the capabilities of BonFIRE
18
Fraunhofer Institute FOKUS, ATOS Origin and other members of the BonFIRE consortium 2010
BonFIRE 18
Agenda BonFIREs Experimental Facility Architecture BonFIREs Infrastructure, available Resources BonFIREs Experiment Support Examples of current BonFIRE Experiments Demo of BonFIRE Experiment Setup and Monitoring Procedure
19
Fraunhofer Institute FOKUS, ATOS Origin and other members of the BonFIRE consortium 2010
BonFIRE 19
20
Fraunhofer Institute FOKUS, ATOS Origin and other members of the BonFIRE consortium 2010
BonFIRE 20
21
FOKUS, ATOS Origin and other members of the BonFIRE Fraunhofer Institute Centre of Supercomputing of Galicia (CESGA) consortium 2010
BonFIRE 21
22
FOKUS, ATOS Origin and other members of the BonFIRE Fraunhofer Institute Centre of Supercomputing of Galicia (CESGA) consortium 2010
BonFIRE 22
Controlled deployment Controlled network Very fine-grained monitoring Elasticity and cross-site elasticity Steering triggered by events
BonFIRE 23
24
Fraunhofer Institute FOKUS, ATOS Origin and other members of the BonFIRE consortium 2010
BonFIRE 24
Problem Statement
Make better QoS provisioning decisions
Why?
Fraunhofer Institute FOKUS, ATOS Origin and other members of the BonFIRE consortium 2010
Experiment Hypothesis
Hypothesis: If IaaS resources are described in terms of application benchmark scores then PaaS performance models will be faster, cheaper and more accurate Approach: replace low-level resource metrics (e.g. CPU architecture, clockspeed, RAM, EC2 compute units) with benchmark scores based on defined patterns of communication and computation (e.g. dwarfs)
?
IaaS SLA IaaS Provisioning IaaS Virtualisation
26
Fraunhofer Institute FOKUS, ATOS Origin and other members of the BonFIRE consortium 2010
BonFIRE 26
Role in BonFIRE
BonFIRE Capability Heterogeneous hardware Detailed monitoring information Uniform request and access methods Repeatability of hardware allocation Cross-site federation Control of the network Access to long-term storage and experimental data Authentication and authorisation Why the experiment needs it The models must be predictive on a wide range or target platforms Monitoring info on several levels is required to create the models To make access to all the resources at different sites easy So that benchmark scores may be compared with test application data For complex test applications To simulate different network conditions Essential for off-line analysis of experiments Ensuring that experiments are not interfered with
The vast majority of these capabilities are of use to all experiments. The experiment will help BonFIRE by providing concrete requirements for these features and by testing and validating the infrastructure, the documentation and the support systems. BonFIRE 27 27 Fraunhofer Institute FOKUS, ATOS Origin and other members of the BonFIRE consortium 2010
28
Fraunhofer Institute FOKUS, ATOS Origin and other members of the BonFIRE consortium 2010
BonFIRE 28
Technical overview
Develop a web-app for test purposes, and generate an image inside a VM to be instantiated at any time. Create a dummy load generator. Define load scenarios. An application must be developed in order to: Collect /store/process KPI (on demand or periodically). Check, in all scenarios, the SLAs accomplishment. Manage rules of SLAs. Generate changes over the testbed scheme. Define a testbed in the cloud with some minimum initial requirements eg: Number of VM Geographical location of resources. Network topology Identify the KPIs for each load scenario Set the upper and lower thresholds of the initial testbed scheme by load tests. Define the set of KPIs values that trigger the release or the enlargement of the resources deployed. Using a time diagram of a load model, resize the testbed resources in advance. With the information collected about KPI in a defined period of time generate a load model inferring the ideal deployment.
29
Fraunhofer Institute FOKUS, ATOS Origin and other members of the BonFIRE consortium 2010
BonFIRE 29
Improvements to
30
Fraunhofer Institute FOKUS, ATOS Origin and other members of the BonFIRE consortium 2010
Experiment Objectives
To optimize the distribution of web-apps containers. To investigate new scalability policies. To be able to predict the ideal deployment under the upcoming load. Learn how to resize dynamically the number of resources deployed, from a load model of a web application, inside a cloud federation infrastructure. Develop algorithms and procedures to avoid overprovisioning resources and to satisfy SLAs with web-app users. To study the concept of infinite elasticity in the cloud.
31
Fraunhofer Institute FOKUS, ATOS Origin and other members of the BonFIRE consortium 2010
BonFIRE 31
EXPERIMENT -MANAGER
KPI
LB OPERATIONS WS AS WS AS DB WEB-APP WS AS
CLOUD-MANAGER
MONITORING
VM-REPOSITORY
32
Fraunhofer Institute FOKUS, ATOS Origin and other members of the BonFIRE consortium 2010
BonFIRE 32
A cloud ready to run VMs offering the various web app containers and databases. Definition of different network schemes. Access to resources on-demand. Federation with other clouds providers. A service to finely monitor the VM containers. Access to geographically separated sites.
33
Fraunhofer Institute FOKUS, ATOS Origin and other members of the BonFIRE consortium 2010
BonFIRE 33
Comprehensive test of BonFIRE. Better understanding of BonFIRE elasticity capabilities. Monitoring process. Users interface (OVF, portal) Dynamic testbed configuration. Creation of scalability policies. Simulation of realistic scenarios with geogr. sep. sites. Improvement of some features of the BonFIRE facility:
Load balancing. Evolution of web traffic generators.
34
Fraunhofer Institute FOKUS, ATOS Origin and other members of the BonFIRE consortium 2010
BonFIRE 34
Agenda BonFIREs Experimental Facility Architecture BonFIREs Infrastructure, available Resources BonFIREs Experiment Support Examples of current BonFIRE Experiments Demo of BonFIRE Experiment Setup and Monitoring Procedure
35
Fraunhofer Institute FOKUS, ATOS Origin and other members of the BonFIRE consortium 2010
BonFIRE 35
Demo
This Demo
Client-server application
Server VM on EPCC BonFIRE site
Monitoring
VM on EPCC BonFIRE site Display CPU load Configure and monitor application metric
Fraunhofer Institute FOKUS, ATOS Origin and other members of the BonFIRE consortium 2010
Iperf Server
Monitoring
38
Fraunhofer Institute FOKUS, ATOS Origin and other members of the BonFIRE consortium 2010
BonFIRE 38
Set up experiment
What is an experiment?
Collection of Compute, Storage and Network resources, each associated with a BonFIRE site Software implementing the experiment logic
40
Fraunhofer Institute FOKUS, ATOS Origin and other members of the BonFIRE consortium 2010
BonFIRE 40
Create VMs
42
Fraunhofer Institute FOKUS, ATOS Origin and other members of the BonFIRE consortium 2010
BonFIRE 42
Monitoring
Set up Monitoring
BonFIRE base images include Zabbix monitoring client Zabbix monitoring server deployed as a separate VM
44
Fraunhofer Institute FOKUS, ATOS Origin and other members of the BonFIRE consortium 2010
BonFIRE 44
Cross-site Elasticity
46
Fraunhofer Institute FOKUS, ATOS Origin and other members of the BonFIRE consortium 2010
BonFIRE 46
Controlled network
Emulated Control
Virtual Wall: Large-scale testbed in IBBT Based on Emulab software Add a private network with configured parameters:
> > > > > > > > > private_network = experiment.networks.submit( :location => ibbt, :name => "network-experiment#{experiment['id']}", :bandwidth => 1000, :latency => 0, :size => 24, :lossrate => 0, :address => "192.168.0.0" )
48
Fraunhofer Institute FOKUS, ATOS Origin and other members of the BonFIRE consortium 2010
BonFIRE 48
Controlled network
To interconnect with FEDERICA during the summer
Advanced network experimentation, supported by GEANT3 Ability to control routing on FEDERICA slices Possibility to route EPCC-PSNC traffic through FEDERICA
49
Fraunhofer Institute FOKUS, ATOS Origin and other members of the BonFIRE consortium 2010
BonFIRE 49
Summary
Experiment Descriptors
Portal Restfully JSON DSL (OVF on the way)
Advanced monitoring
Zabbix on all VMs Infrastructure monitoring
BonFIRE 51
+ =
W: www.bonfire-project.eu E: bonfire@bonfire-project.eu
53
Fraunhofer Institute FOKUS, ATOS Origin and other members of the BonFIRE consortium 2010
BonFIRE 53