managing the performance of clouds and cloud based ......• what cloud infrastructure and cloud...
TRANSCRIPT
IEEE CQR 2012 - Clark
Managing the Performance of Clouds and Cloud Based Applications
Alan Clark, Telchemy
IEEE CQR 2012
IEEE CQR 2012 - Clark
Outline
• Model for Cloud based Applications
• Cloud Application Performance
• Typical Cloud SLA
• What Cloud Infrastructure and Cloud
Application providers need to measure
IEEE CQR 2012 - Clark
Why run applications in the Cloud?
Time
Required Capacity
Available Capacity
Tran
sact
ions
Flexible/ Elastic Resources – Computing on Demand
Cost savings
Cost savings
IEEE CQR 2012 - Clark
Storage Resources Storage
Resources
Compute Resources Compute Resources
Model of a Cloud Based Application
Compute Resources
Storage Resources
Broker/ Resource Manager
Application
Distributed Computation Layer
Application User Network
IaaS (Infrastructure as a Service)
Cloud Based Application
Resource Requests
Storage Resources Storage
Resources Network
Access BW
VM VM VM VM VM
IEEE CQR 2012 - Clark
Storage Resources Storage
Resources
Compute Resources Compute Resources
Cloud Model
Compute Resources
Storage Resources
Broker/ Resource Manager
Application
Distributed Computation Layer
Application User Network
IaaS (Infrastructure as a Service)
Cloud Based Application
Resource Requests
Storage Resources Storage
Resources Network
Access BW
Measured Cloud Application Performance
Measurable Application QoE for Users Impacted by load, available resources..
IEEE CQR 2012 - Clark
Storage Resources Storage
Resources
Compute Resources Compute Resources
Cloud Model
Compute Resources
Storage Resources
Broker/ Resource Manager
Application
Distributed Computation Layer
Application User Network
IaaS (Infrastructure as a Service)
Cloud Based Application
Resource Requests
Storage Resources Storage
Resources Network
Access BW
Implicit or Explicit Cloud Application Performance Model
Relates number of users, user driven activity to computing resources required
Determines when to request add/drop resources based on predicted load
Objective – least usage of cloud resources required to provide acceptable performance to users
Measured Cloud Application Performance
Measurable Application QoE for Users Impacted by load, available resources..
IEEE CQR 2012 - Clark
Flexing Resources
Time
Performance model drives resource requests Tr
ansa
ctio
ns
Actual Application Capacity
IEEE CQR 2012 - Clark
Flexing Resources – potential issues
Time
Tran
sact
ions
Resource Allocation Delay
Prediction of resource requirements may be too aggressive or conservative
Billing period – may be 5 minutes or 1 hour
Degraded Performance Unnecessary
cost $$
IEEE CQR 2012 - Clark
Storage Resources Storage
Resources
Compute Resources Compute Resources
Cloud Infrastructure – typical SLA
Compute Resources
Storage Resources
Broker/ Resource Manager
IaaS (Infrastructure as a Service)
Storage Resources Storage
Resources Network
Access BW
Typical SLA – Availability
Amazon EC2 SLA - > 99.5% availability over 365 days Verizon Terremark – service credit for each 15 minute outage
Typical performance metrics (not part of SLA)
CPU capacity allocated Percentage usage of CPU capacity
IEEE CQR 2012 - Clark
What should a Cloud SLA guarantee?
Application
Distributed Computation Layer
Cloud Based Application
Cloud SLA - Service Availability (Short/ Long Term) - Dynamic Resource Availability - Resource Allocation Response Time - Resource Allocation Accuracy
Elastic Computing service – should have SLA that supports Elastic Computing - Do resources get allocated - Delay to provide resources - Are correct resources being provided
Availability guarantees should be short term – e.g. minutes of outage per hour
IEEE CQR 2012 - Clark
What should the Application Provider measure?
Application
Distributed Computation Layer
Cloud Based Application
Performance of Cloud API - Availability - Resource Requests
- Failure rate - Response time
- Resource Usage
Performance seen by user - Availability - Response time - Transaction time - User QoE
Performance Model - Model accuracy - Model tuning
Measured by Cloud Application
IEEE CQR 2012 - Clark
Summary
• Model for Cloud based Applications
• Cloud Application Performance
• Typical Cloud SLA
• What Cloud Infrastructure and Cloud Application
providers need to measure
• Related information – ANSA – (1985-95 Esprit funded research program) – NIST Cloud Computing Definition – DMTF Cloud Management WG – ATIS Cloud Services Forum – ITU-T Cloud Focus Group – Amazon CloudWatch – OpenStack
IEEE CQR 2012 - Clark
About Telchemy
• Leading provider of technology for real time analysis of VoIP, Videoconferencing, IPTV and Video Streaming services
• Pioneered distributed performance management model for real time services, using endpoint based agents. Over 100 million agents deployed in IP phones, media gateways, CPE routers, MTA’s, set top boxes, soft clients……
• Performance Management solutions for IP Centrex/ Hosted PBX, Residential triple play, Mobile, Enterprise, Government