Unit 5
Advanced Features
Syllabus
Instruction Set Issues Profiling
Migration Grids Code optimizations
Garbage Collection
Book
James E. Smith and Ravi Nair, Virtual Machines,
Elsevier, 2005.
Overview
Instruction Set Issues
Profiling
Migration
Grids
Code Optimizations
Garbage Collection
Instruction Set Issues
Instruction Set Issues
Register architectures
Condition codes
Data formats and Arithmetic
Memory address resolution
Memory data alignment
Byte order
Addressing architecture
Instruction Set Issues
Register architectures
Condition codes
Data formats and Arithmetic
Memory address resolution
Memory data alignment
Byte order
Addressing architecture
Profiling
Profiling
Process of collecting instruction and data statistics
for an executing program
Can be used as input to code-optimization process
Static Profiling
Dynamic Profiling
Types of Profiling
Frequency of execution of code regions
Based on control flow predictability
Path profile
Collecting Profiles
Two ways
Instrumentation-based profiling
Targets specific program-related events and counts all
instances of the events being profiled
Hardware instrumentation
Software instrumentation
Sampling-based profiling
Program runs unmodified and at fixed or random intervals, the
program is interrupted and an instance of a program-related
event is captured
Profiling during Interpretation
Two key points to be considered
Source instructions are actually accessed as data
Interpreter routines are the code that is being executed
For block profiling
Profile code should be added to all control transfer
instructions
For edge profiling
Same control transfer instructions are profiled along
with both the PC of the control transfer instruction and
the target are used (defines the specific edge)
Profiling during Interpretation
Profiling Translated Code
Profiling overhead
More memory and time consuming
Ways to reduce
Reduce number of instrumentation points
Using heuristics
And so on
Code Optimization
Code Optimizations
Code Optimizations
Code Optimizations
Code Optimizations
Code Optimizations
Inter-superblock Optimization
Instruction-set-specific
Optimizations
Two examples
Unaligned load optimization
if conversion
Garbage Collection
Garbage Collectors
Mark and sweep
Compacting
Copying
Generational
Incremental and concurrent
Compacting Collectors
Copy Collectors
Generational Collectors
Two sub-heaps
Nursery
Tenured
Incremental and Concurrent
Collectors
Comparison
Migration
VM Migration
Internet Suspend/Resume (ISR)
State Encapsulation
Stanford Collective Project
State of the VM is sent directly to the destination
computer rather than to a DFS
State Encapsulation
Reducing memory state before migration
Balloon program
Reducing the size of the transmitted packet
Copy-on-write
Reducing Start-up time on a Resume
Incremental build-up on demand
Reducing transmission time and bandwidth by
exploiting redundancy in disk blocks
Hashed copy
Migration in Vmotion
Load balancing
To improve the response time of the system through better
utilization of resources
Security
To quarantine a virtual machine that has been attacked
Collocation
To bring communicating virtual machines closer together
Fault tolerance
To move a failing host to another processor
Power management
To move the load away from an overheated processor
Maintenance
To move the load away from some processor while it is upgraded
Migration in Vmotion
Steps
Check if running VM is stable
Baseline copy to the new host
Copy VMs changed state to new host
Activate the VM in new host
Grids
Grid Computing
Characteristics of a grid
Infrastructure
Dependability of Service
Consistency of Service
Pervasive Access
Inexpensive Access
Coordinated Resource Sharing
Dynamic Communities
Comparison with Conventional
VMs
Efficient Utilization of Resources
Sharing of Resources
Distributed Vs Centralized Control
Heterogeneous Nodes
Adaptation of Applications
Portability of Applications
Recap
Instruction Set Issues
Profiling
Migration
Grids
Code Optimizations
Garbage Collection
