SimGrid Cloud Broker:
Simulation of Public and Private Clouds
Jonathan Rouzaud-Cornabas CNRS – CC-IN2P3 / LIP (UMR 5668)
SimGrid Cloud Broker
SimGrid
simulates many different distributed systems: clusters, wide-area and local-area networks, peers over DSL connexions, data centers, etc has models are theoretically and experimentally assessed
is scalable exists for 13 years
Simulation of Clouds
From an user point of view
From a cloud provider point of view Evaluates application(s) running on Clouds Evaluates different policies in Cloud middleware Multi-Clouds (private and public)
Architecture
AWS
Amazon Web Services
S3 / EC2 API
All the instance types All the regions of AWS
On-demand and Spot Instances 3 types of storage: local, EBS and S3
Accounting of network, computing and storage resources Models:
Spot instance prices: smart random, file and prediction model Life cycle of VMs
Storage (3 models) Compute
AWS
Amazon AWS Platforms
Private Cloud
Private Cloud
IaaS Computing Simulation
Basic compute API (loosely based on OpenStack) Eucalyptus and OpenStack models (in progress)
Easy-to-use
Easy-to-use simulator
Easy to use platform description
One XML file to describe all the Clouds
SimGrid users: only few modifications to migrate a SimGrid code to SGCB code
Easy to replace and extend: VM image deployment and VM to PM mapping policies
Scalability
Studying the scalability of SGCB
Tested on the simulated AWS platform with all the models activated Mid-term goal: 1 millions VMs on 100,000 Hosts
4 types of scenario (1≤N ≤500)
1 N instancesm1.largeineu 1
2 N instances of each instance types ineu 1 3 N instancesm1.largein all regions
4 N instances of each instance types in all regions
Scalability
Scalability of SGCB
0 500 1000 1500 0 10000 20000 30000 NBVM SimTimeScalability 0 300 600 900 1200 2.5 5.0 7.5 10.0 AMISIZE Dur ation State Boot Deploy Running Shutdown
Scalability Simple bag of tasks
Simple Bag Of Tasks scenario
BoT Characteristics 1,000 Tasks Input data: 1GB Output data: 500MB
Storage Input and/or Output: Local, EBS ou S3 1 master (CeS): m2.2xlarge
Scalability Simple bag of tasks od_ebs_ebs od_ebs_local od_ebs_s3 od_local_ebs od_local_local od_local_s3 od_s3_ebs od_s3_local od_s3_s3 si_ebs_ebs si_ebs_local si_ebs_s3 si_local_ebs si_local_local si_local_s3 si_s3_ebs si_s3_local si_s3_s3
0e+00 5e+05 1e+06 Duration Scenar io StateStart RequestVM SendTasks ReceiveResults TerminateVM End
Scalability Simple bag of tasks od_ebs_ebs od_ebs_local od_ebs_s3 od_local_ebs od_local_local od_local_s3 od_s3_ebs od_s3_local od_s3_s3 si_ebs_ebs si_ebs_local si_ebs_s3 si_local_ebs si_local_local si_local_s3 si_s3_ebs si_s3_local si_s3_s3 0 500 1000 1500 Duration Scenar io StateSendInput ReceiveInput WriteInput ExecuteTask ReadOutput SendOutput ReceiveOutput
Scalability Simple bag of tasks EBS S3 0 500 1000 GBMonth T ype Scenario od_ebs_ebs od_ebs_local od_ebs_s3 od_local_ebs od_local_local od_local_s3 od_s3_ebs od_s3_local od_s3_s3 si_ebs_ebs si_ebs_local si_ebs_s3 si_local_ebs si_local_local si_local_s3 si_s3_ebs si_s3_local si_s3_s3
Scalability Simple bag of tasks
Internet Network VM Intranet Network VM Network S3
0e+00 2e+11 4e+11 6e+11 GB T ype Scenario od_ebs_ebs od_ebs_local od_ebs_s3 od_local_ebs od_local_local od_local_s3 od_s3_ebs od_s3_local od_s3_s3 si_ebs_ebs si_ebs_local si_ebs_s3 si_local_ebs si_local_local si_local_s3 si_s3_ebs si_s3_local si_s3_s3
Scalability Simple bag of tasks od_ebs_ebs od_ebs_local od_ebs_s3 od_local_ebs od_local_local od_local_s3 od_s3_ebs od_s3_local od_s3_s3 si_ebs_ebs si_ebs_local si_ebs_s3 si_local_ebs si_local_local si_local_s3 si_s3_ebs si_s3_local si_s3_s3 0 500 1000 1500 2000 Price Scenar io TypeEBS Network S3 Network VM S3 VM OD VM SI
Scalability Simple bag of tasks
m1.small m1.large m1.xlarge m2.xlarge m2.2xlarge m2.4xlarge c1.medium c1.xlarge
0.0 0.5 1.0 1.5 0 100 200 300 0 100 200 300 0 100 200 300 0 100 200 300 0 100 200 300 0 100 200 300 0 100 200 300 0 100 200 300 Hour Pr ice Region brasil_1 asia_2 asia_1 eu_1 usa_east_1 usa_west_1 usa_west_2
Use case
SGCB usage in the Inria Avalon team
Study different resource reservation and task allocation algorithms for Bag Of Tasks on Clouds
Multi-regions applications and inter-region migration on AWS (Jose Luis Lucas – UCM)
Scientific Workflows on Clouds (Dao Van Toan – Master internship) Applications composed of multiple VMs with security requirements (Arnaud Lefray – Doctorant)
Component based applications on federated PaaS (FP7 PaaSage)
Future work
Future work
Validation of the simulation results with experiment on AWS Finer grain models for AWS
Running the same experimentation on a private Cloud
Study the different between OpenStack and OpenNebula on same hardware and applications
Models for private Clouds
Integrate works around live-migration Multi-core for VMs models in SimGrid Increase the scalability