Hypervisorbased Fault tolerance

1
Hypervisor-basedFault tolerance

• Thomas C. Bressoud
• Isis Distributed Systems
• Fred B. Schneider
• Cornell University

2
Outline

• Intro
• Related Works
• Replica Coordination Protocols
• A Prototype System
• Performance of the Prototype
• QA

3
1.Intro

• Intro
• Popular Scheme replicating computing replicas
  coordinating
• Three Problems
• Hardware hardware design cost
• OS difficult complicated OSes
• Software modification on each apps
• Hypervisor software layer to implement virtual
  machines having the same instruction set
  architecture as the hardware

4
1.Intro (cont.)

• Hypervisor advantages
• Multiple OSes coexisting on a single processor
• Isolation mechanism
• Deal with the before 3 problems
• Hardware independency
• OS independency
• App independency
• Two issues in this paper
• Practical Design a protocol
• Performance Prototype experiments

Xiang Xiaojia
Department of Computer Science Slide 4
5
2.Related Works

• Processors implementing replica coordination in
  hardware
• TandemCMJ88
• DECs VAXft 3000
• App level replica coordination
• TandemSS92
• On top of OS
• Fault tolerance under UNIX

Xiang Xiaojia
Department of Computer Science Slide 5
6
2.Related Works (cont)

• NetwareMPN92
• Very similar to us
• Drawbacks
• Modify OS internal
• Proscription of Preemption
• Not transparent
• Failovers are not masked

Xiang Xiaojia
Department of Computer Science Slide 6
7
3. Replica Coordination Protocols

• T-fault tolerant model
• 1 primary, t backups
• Primary works, no backups interact with
  environment
• Primary fails, only one backup takes its place
• Assumptions

Xiang Xiaojia
Department of Computer Science Slide 7
8
3. Replica Coordination Protocols

• Identical Instruction Streams
• VM state memory and registers that change only
  with execution of instructions of that VM
• Two kinds of instructions
• Ordinary instructions its behavior is determined
  by VM state
• Environment instructions on the controry
• Assumptions

Xiang Xiaojia
Department of Computer Science Slide 8
9
3. Replica Coordination Protocols

• Protocol - Primary

Xiang Xiaojia
Department of Computer Science Slide 9
10
3. Replica Coordination Protocols

• Protocol - Backup

Xiang Xiaojia
Department of Computer Science Slide 10
11
3. Replica Coordination Protocols

• Protocol-Primary fail-backup promote

Xiang Xiaojia
Department of Computer Science Slide 11
12
3. Replica Coordination Protocols

• Interaction with an Enviroment
• State of Environment execution of IO
  instructions
• Assumption
• All IO devices are assumed to comply with

Xiang Xiaojia
Department of Computer Science Slide 12
13
4. Prototype

• Hypervisor
• Memory architecture
• Dont support multiple VMs
• Solution single VM
• Privilege levels
• How to arrange for VM instructions
• Mapping VM 0 -gt 1 VM 3 -gt 3

Xiang Xiaojia
Department of Computer Science Slide 13
14
4. Prototype (cont.)

• Replica Coordination in Hypervisor
• TLB problem
• TLB-gtMemory-gtDisk
• Difference in TLB contents become visible ?
• Solution
• Handle pare of TLB miss trap to hypervisor

Xiang Xiaojia
Department of Computer Science Slide 14
15
5.Evaluation

• CPU Intensive Workload

Xiang Xiaojia
Department of Computer Science Slide 15
16
5.Evaluation (cont.)

• IO Intensive Workload

Xiang Xiaojia
Department of Computer Science Slide 16
17
5.Evaluation (cont.)

• Optimizations

Xiang Xiaojia
Department of Computer Science Slide 17
18
QA

• ??!