SLIDE 1
Incremental Garbage Collection
Part II
Roland Schatz
Incremental Garbage Collection – p.1/22
Baker – Disadvantages
not concurrent
Incremental Garbage Collection – p.2/22
Incremental Garbage Collection Part II Roland Schatz Incremental - - PowerPoint PPT Presentation
Incremental Garbage Collection Part II Roland Schatz Incremental - - PowerPoint PPT Presentation
Incremental Garbage Collection Part II Roland Schatz Incremental Garbage Collection p.1/22 Baker Disadvantages not concurrent Incremental Garbage Collection p.2/22 Baker Disadvantages not concurrent not even threadsafe
SLIDE 2
SLIDE 3
Baker – Disadvantages
not concurrent not even threadsafe
Incremental Garbage Collection – p.2/22
SLIDE 4
Baker – Disadvantages
not concurrent not even threadsafe expensive read trap
Incremental Garbage Collection – p.2/22
SLIDE 5
Baker – Disadvantages
not concurrent not even threadsafe expensive read trap requires hardware support
Incremental Garbage Collection – p.2/22
SLIDE 6
Appel-Ellis-Li Collector
no black ⇒ white pointers
Incremental Garbage Collection – p.3/22
SLIDE 7
Appel-Ellis-Li Collector
no black ⇒ white pointers Baker’s invariant
Incremental Garbage Collection – p.3/22
SLIDE 8
Appel-Ellis-Li Collector
no black ⇒ white pointers Baker’s invariant mutator never sees white pointers
Incremental Garbage Collection – p.3/22
SLIDE 9
Appel-Ellis-Li Collector
no black ⇒ white pointers Baker’s invariant mutator never sees white pointers intercept all pointer reads
Incremental Garbage Collection – p.3/22
SLIDE 10
Appel-Ellis-Li Collector
no black ⇒ white pointers Baker’s invariant mutator never sees white pointers intercept all pointer reads slightly stricter constraint
Incremental Garbage Collection – p.3/22
SLIDE 11
Appel-Ellis-Li Collector
no black ⇒ white pointers Baker’s invariant mutator never sees white pointers intercept all pointer reads slightly stricter constraint mutator only sees black objects
Incremental Garbage Collection – p.3/22
SLIDE 12
Appel-Ellis-Li Collector
no black ⇒ white pointers Baker’s invariant mutator never sees white pointers intercept all pointer reads slightly stricter constraint mutator only sees black objects use page protection hardware
Incremental Garbage Collection – p.3/22
SLIDE 13
Memory Layout
fromspace
potential garbage
start
tospace
scanned objects scanned unscanned objects unscanned free space alloc new objects end
Incremental Garbage Collection – p.4/22
SLIDE 14
Page Protection
black pages
Incremental Garbage Collection – p.5/22
SLIDE 15
Page Protection
black pages contain only tospace pointers
Incremental Garbage Collection – p.5/22
SLIDE 16
Page Protection
black pages contain only tospace pointers no problem
Incremental Garbage Collection – p.5/22
SLIDE 17
Page Protection
black pages contain only tospace pointers no problem white pages
Incremental Garbage Collection – p.5/22
SLIDE 18
Page Protection
black pages contain only tospace pointers no problem white pages mutator never sees white pointers
Incremental Garbage Collection – p.5/22
SLIDE 19
Page Protection
black pages contain only tospace pointers no problem white pages mutator never sees white pointers so we don’t care
Incremental Garbage Collection – p.5/22
SLIDE 20
Page Protection
gray pages
Incremental Garbage Collection – p.6/22
SLIDE 21
Page Protection
gray pages mutator may see gray pointers
Incremental Garbage Collection – p.6/22
SLIDE 22
Page Protection
gray pages mutator may see gray pointers mutator must not see gray objects
Incremental Garbage Collection – p.6/22
SLIDE 23
Page Protection
gray pages mutator may see gray pointers mutator must not see gray objects set to no access
Incremental Garbage Collection – p.6/22
SLIDE 24
Handling Page Faults
Mutator Tospace before the trap Read barrier
Incremental Garbage Collection – p.7/22
SLIDE 25
Handling Page Faults
Mutator Tospace before the trap Read barrier
Incremental Garbage Collection – p.7/22
SLIDE 26
Handling Page Faults
Mutator Tospace before the trap Tospace after the trap Read barrier
Incremental Garbage Collection – p.7/22
SLIDE 27
Handling Page Faults
Mutator Tospace before the trap Tospace after the trap Read barrier
Incremental Garbage Collection – p.7/22
SLIDE 28
Summary
advantages
Incremental Garbage Collection – p.8/22
SLIDE 29
Summary
advantages concurrent scanning possible
Incremental Garbage Collection – p.8/22
SLIDE 30
Summary
advantages concurrent scanning possible cheaper read trap
Incremental Garbage Collection – p.8/22
SLIDE 31
Summary
advantages concurrent scanning possible cheaper read trap problems
Incremental Garbage Collection – p.8/22
SLIDE 32
Summary
advantages concurrent scanning possible cheaper read trap problems relies on virtual memory
Incremental Garbage Collection – p.8/22
SLIDE 33
Summary
advantages concurrent scanning possible cheaper read trap problems relies on virtual memory global lock is bottleneck
Incremental Garbage Collection – p.8/22
SLIDE 34
Summary
advantages concurrent scanning possible cheaper read trap problems relies on virtual memory global lock is bottleneck flip is still stop-the-world
Incremental Garbage Collection – p.8/22
SLIDE 35
Nettle’s Replicating Collector
read barrier is still expensive
Incremental Garbage Collection – p.9/22
SLIDE 36
Nettle’s Replicating Collector
read barrier is still expensive use write barrier instead
Incremental Garbage Collection – p.9/22
SLIDE 37
Nettle’s Replicating Collector
Mutator Fromspace forward ptr
- riginal
Tospace header word replica
GC
getHeader read write
Incremental Garbage Collection – p.10/22
SLIDE 38
Nettle’s Replicating Collector
Mutator Fromspace forward ptr
- riginal
Tospace header word replica
GC
Mutation Log getHeader read write
Incremental Garbage Collection – p.10/22
SLIDE 39
Huelsbergen-Laurus Collector
distinguish mutable/immutable objects
Incremental Garbage Collection – p.11/22
SLIDE 40
Huelsbergen-Laurus Collector
distinguish mutable/immutable objects e.g. ML
Incremental Garbage Collection – p.11/22
SLIDE 41
Huelsbergen-Laurus Collector
distinguish mutable/immutable objects e.g. ML immutable objects
Incremental Garbage Collection – p.11/22
SLIDE 42
Huelsbergen-Laurus Collector
distinguish mutable/immutable objects e.g. ML immutable objects replicating collection
Incremental Garbage Collection – p.11/22
SLIDE 43
Huelsbergen-Laurus Collector
distinguish mutable/immutable objects e.g. ML immutable objects replicating collection mutable objects
Incremental Garbage Collection – p.11/22
SLIDE 44
Huelsbergen-Laurus Collector
distinguish mutable/immutable objects e.g. ML immutable objects replicating collection mutable objects classic copying collection
Incremental Garbage Collection – p.11/22
SLIDE 45
Doligez-Leroy Collector
Problem: heavy synchronization
Incremental Garbage Collection – p.12/22
SLIDE 46
Doligez-Leroy Collector
Problem: heavy synchronization seperate thread-local data
Incremental Garbage Collection – p.12/22
SLIDE 47
Doligez-Leroy Collector
Problem: heavy synchronization seperate thread-local data young generation
Incremental Garbage Collection – p.12/22
SLIDE 48
Doligez-Leroy Collector
Problem: heavy synchronization seperate thread-local data young generation thread-local
Incremental Garbage Collection – p.12/22
SLIDE 49
Doligez-Leroy Collector
Problem: heavy synchronization seperate thread-local data young generation thread-local immutable
Incremental Garbage Collection – p.12/22
SLIDE 50
Doligez-Leroy Collector
Problem: heavy synchronization seperate thread-local data young generation thread-local immutable
- ld generation
Incremental Garbage Collection – p.12/22
SLIDE 51
Doligez-Leroy Collector
Problem: heavy synchronization seperate thread-local data young generation thread-local immutable
- ld generation
the rest
Incremental Garbage Collection – p.12/22
SLIDE 52
Thread Local Heaps
thread 1 thread 2 thread 3 young generation
- ld generation
Global variables
Incremental Garbage Collection – p.13/22
SLIDE 53
In-Place Garbage Collection
uncooperative environment
Incremental Garbage Collection – p.14/22
SLIDE 54
In-Place Garbage Collection
uncooperative environment relaxed consistency requirements
Incremental Garbage Collection – p.14/22
SLIDE 55
In-Place Garbage Collection
uncooperative environment relaxed consistency requirements Fragmentation!
Incremental Garbage Collection – p.14/22
SLIDE 56
Four-Color Abstraction
black: scanned
Incremental Garbage Collection – p.15/22
SLIDE 57
Four-Color Abstraction
black: scanned gray: marked, not scanned
Incremental Garbage Collection – p.15/22
SLIDE 58
Four-Color Abstraction
black: scanned gray: marked, not scanned white: not marked
Incremental Garbage Collection – p.15/22
SLIDE 59
Four-Color Abstraction
black: scanned gray: marked, not scanned white: not marked dead-white: free objects
Incremental Garbage Collection – p.15/22
SLIDE 60
Four-Color Implementation
fromspace
potential garbage
start
tospace
scanned objects scanned unscanned objects unscanned free space alloc new objects end
Incremental Garbage Collection – p.16/22
SLIDE 61
Baker’s Treadmill
new objects Tospace Fromspace free list scan top free bottom
Incremental Garbage Collection – p.17/22
SLIDE 62
Baker’s Treadmill
new objects Tospace Fromspace free list scan top free bottom
Incremental Garbage Collection – p.17/22
SLIDE 63
Baker’s Treadmill
new objects Tospace Fromspace free list scan top free bottom
Incremental Garbage Collection – p.17/22
SLIDE 64
Baker’s Treadmill
new objects Tospace Fromspace free list s c a n top free bottom
Incremental Garbage Collection – p.17/22
SLIDE 65
Baker’s Treadmill
new objects Tospace Fromspace free list s c a n top f r e e bottom
Incremental Garbage Collection – p.17/22
SLIDE 66
Baker’s Treadmill
new objects Tospace Fromspace free list scan top f r e e bottom
Incremental Garbage Collection – p.17/22
SLIDE 67
Baker’s Treadmill
new objects Tospace Fromspace free list scan top f r e e bottom
Incremental Garbage Collection – p.17/22
SLIDE 68
Baker’s Treadmill
new objects Tospace Fromspace free list scan top f r e e bottom
Incremental Garbage Collection – p.17/22
SLIDE 69
Baker’s Treadmill
new objects Tospace Fromspace scan top free bottom
Incremental Garbage Collection – p.17/22
SLIDE 70
Baker’s Treadmill
Fromspace free list s c a n t
- p
free bottom
Incremental Garbage Collection – p.17/22
SLIDE 71
Baker’s Treadmill
Tospace Fromspace free list s c a n top free bottom
Incremental Garbage Collection – p.17/22
SLIDE 72
Baker’s Treadmill
Tospace Fromspace free list scan t
- p
free bottom
Incremental Garbage Collection – p.17/22
SLIDE 73
Baker’s Treadmill
new Tospace Fromspace free list scan t
- p
f r e e bottom
Incremental Garbage Collection – p.17/22
SLIDE 74
Advantages
compared to incremental mark/sweep
Incremental Garbage Collection – p.18/22
SLIDE 75
Advantages
compared to incremental mark/sweep easy allocation
Incremental Garbage Collection – p.18/22
SLIDE 76
Advantages
compared to incremental mark/sweep easy allocation easy to “blacken” object
Incremental Garbage Collection – p.18/22
SLIDE 77
Advantages
compared to incremental mark/sweep easy allocation easy to “blacken” object easy flip
Incremental Garbage Collection – p.18/22
SLIDE 78
Advantages
compared to incremental mark/sweep easy allocation easy to “blacken” object easy flip compared to copying collector
Incremental Garbage Collection – p.18/22
SLIDE 79
Advantages
compared to incremental mark/sweep easy allocation easy to “blacken” object easy flip compared to copying collector write barrier
Incremental Garbage Collection – p.18/22
SLIDE 80
Advantages
compared to incremental mark/sweep easy allocation easy to “blacken” object easy flip compared to copying collector write barrier equal or better memory utilization
Incremental Garbage Collection – p.18/22
SLIDE 81
Advantages
compared to incremental mark/sweep easy allocation easy to “blacken” object easy flip compared to copying collector write barrier equal or better memory utilization compared to both
Incremental Garbage Collection – p.18/22
SLIDE 82
Advantages
compared to incremental mark/sweep easy allocation easy to “blacken” object easy flip compared to copying collector write barrier equal or better memory utilization compared to both depth first traversal for free
Incremental Garbage Collection – p.18/22
SLIDE 83
Problems
heterogenous objects
Incremental Garbage Collection – p.19/22
SLIDE 84
Problems
heterogenous objects multiple Treadmills
Incremental Garbage Collection – p.19/22
SLIDE 85
Problems
heterogenous objects multiple Treadmills memory fragmentation
Incremental Garbage Collection – p.19/22
SLIDE 86
Improvements
- ne Treadmill for big objects
Incremental Garbage Collection – p.20/22
SLIDE 87
Improvements
- ne Treadmill for big objects
seperate Treadmills for small objects
Incremental Garbage Collection – p.20/22
SLIDE 88
Improvements
- ne Treadmill for big objects
seperate Treadmills for small objects free page migration between Treadmills
Incremental Garbage Collection – p.20/22
SLIDE 89
Improvements
- ne Treadmill for big objects
seperate Treadmills for small objects free page migration between Treadmills page-wise collection for fast migration
Incremental Garbage Collection – p.20/22
SLIDE 90
Improvements
- ne Treadmill for big objects
seperate Treadmills for small objects free page migration between Treadmills page-wise collection for fast migration free page remapping for large allocations
Incremental Garbage Collection – p.20/22
SLIDE 91
Improvements
- ne Treadmill for big objects
seperate Treadmills for small objects free page migration between Treadmills page-wise collection for fast migration free page remapping for large allocations page-filling allocation
Incremental Garbage Collection – p.20/22
SLIDE 92
Questions?
Incremental Garbage Collection – p.21/22
SLIDE 93
Thanks for your attention!
Incremental Garbage Collection – p.22/22