1/22 workshop modcs 2012.2 dependability investigation of i/o bound services on cloud computing...
TRANSCRIPT
1/22
Workshop MODCS 2012.2
Dependability Investigation of
I/O Bound Services on
Cloud Computing
Erico Augusto Cavalcanti Guedes
2/22Erico Augusto C. GuedesMODCS - Modelling of Distributed and Concurrent Systems
Summary
1. Introduction
2. Virtualization
3. “What is the metter?”
4. PMRAV
5. Expirements
6. Discussion and future work
3/22Erico Augusto C. GuedesMODCS - Modelling of Distributed and Concurrent Systems
1. Introduction
• Cloud computing presents dependability problems on I/O bound services:
4/22Erico Augusto C. GuedesMODCS - Modelling of Distributed and Concurrent Systems
1. Introduction
• Cloud computing presents dependability problems on I/O bound services:
5/22Erico Augusto C. GuedesMODCS - Modelling of Distributed and Concurrent Systems
Summary
1. Introduction
2. Virtualization
3. “What is the metter?”
4. PMRAV
5. Expirements
6. Discussion and future work
6/22Erico Augusto C. GuedesMODCS - Modelling of Distributed and Concurrent Systems
2. Virtualization
7/22Erico Augusto C. GuedesMODCS - Modelling of Distributed and Concurrent Systems
2. Virtualization
8/22Erico Augusto C. GuedesMODCS - Modelling of Distributed and Concurrent Systems
2. Virtualization
• Server Virtualization
• Full Virtualization
• Virtual Box, VMWare Virtualization, KVM*
• Paravirtualization
• Xen
• No concensus
• OS virtualization
• Hardware-assisted virtualization
9/22Erico Augusto C. GuedesMODCS - Modelling of Distributed and Concurrent Systems
Summary
1. Introduction
2. Virtualization
3. “What is the metter?”
4. PMRAV
5. Expirements
6. Discussion and future work
10/22Erico Augusto C. GuedesMODCS - Modelling of Distributed and Concurrent Systems
3. “What is the metter?”
• Web cache proxy systems does not present acceptable availability and reliability when provided on VMs.
11/22Erico Augusto C. GuedesMODCS - Modelling of Distributed and Concurrent Systems
3. “What is the metter?”
• Reliability and Availability
12/22Erico Augusto C. GuedesMODCS - Modelling of Distributed and Concurrent Systems
3. “What is the metter?”
• Reliability and Availability
13/22Erico Augusto C. GuedesMODCS - Modelling of Distributed and Concurrent Systems
3. “What is the metter?”
• Reliability and Availability
14/22Erico Augusto C. GuedesMODCS - Modelling of Distributed and Concurrent Systems
3. “What is the metter?”
• Web cache proxy systems does not present acceptable availability and reliability when provided on VMs.
• How we arrived at this conclusion?
• Monitoring the system under analysis
15/22Erico Augusto C. GuedesMODCS - Modelling of Distributed and Concurrent Systems
Summary
1. Introduction
2. Virtualization
3. “What is the metter?”
4. PMRAV
5. Expirements
6. Discussion and future work
16/22Erico Augusto C. GuedesMODCS - Modelling of Distributed and Concurrent Systems
4. PMRAV
• Protocolo para Monitoramento de Recursos em Ambientes Virtuais
• Aim to:
• Provide a unified methodology to monitor:
• VMs resources
• PMs resources
• CPU, RAM, I/O, network
• Service metrics
• Web Cache Proxy System: Document Hit Ratio,(DHR) , Byte Hit Ratio(BHR), Response Time(RT).
17/22Erico Augusto C. GuedesMODCS - Modelling of Distributed and Concurrent Systems
4. PMRAV
• Protocolo para Monitoramento de Recursos em Ambientes Virtuais
• netpipes
18/22Erico Augusto C. GuedesMODCS - Modelling of Distributed and Concurrent Systems
4. PMRAV
• Protocolo para Monitoramento de Recursos em Ambientes Virtuais
• Scenario
19/22Erico Augusto C. GuedesMODCS - Modelling of Distributed and Concurrent Systems
Summary
1. Introduction
2. Virtualization
3. “What is the metter?”
4. PMRAV
5. Expirements
6. Discussion and future work
20/22Erico Augusto C. GuedesMODCS - Modelling of Distributed and Concurrent Systems
5. Expirements• PMs:
21/22Erico Augusto C. GuedesMODCS - Modelling of Distributed and Concurrent Systems
5. Expirements• PMs:
22/22Erico Augusto C. GuedesMODCS - Modelling of Distributed and Concurrent Systems
5. Expirements• VMs:
PM running VM VM
23/22Erico Augusto C. GuedesMODCS - Modelling of Distributed and Concurrent Systems
5. Expirements• VMs:
PM running VM VM
24/22Erico Augusto C. GuedesMODCS - Modelling of Distributed and Concurrent Systems
Summary
1. Introduction
2. Virtualization
3. “What is the metter?”
4. PMRAV
5. Expirements
6. Discussion and future work
25/22Erico Augusto C. GuedesMODCS - Modelling of Distributed and Concurrent Systems
6. Discussion and Future Work• Tunning KVM
• CPU pinning
• NUMA x SMP
• Cache policies: write-through, write-back, none
• virtio driver
• Measure MTTF and MTBF
• Raj Jain, about measurements.
26/22
Workshop MODCS 2012.2
Dependability Investigation of
I/O Bound Services on
Cloud Computing
Erico Augusto Cavalcanti Guedes