Self-Collecting Mutators are Self-Compacting An analysis of memory - - PowerPoint PPT Presentation

Self-Collecting Mutators are Self-Compacting

An analysis of memory consumption in real-periodic programs Stephanie Stroka

Embedded Software Engineering University of Salzburg

January 25, 2011

Motivation Periodic Memory Consumption Analysis Implementation Results Outlook

Outline

1 Motivation 2 Periodic Memory Consumption Analysis 3 Implementation 4 Results 5 Outlook

Stephanie Stroka Self-Collecting Mutators are Self-Compacting

Motivation Periodic Memory Consumption Analysis Implementation Results Outlook

The ideal allocator

Fast memory allocation and deallocation No ”wasted” memory space

Internal fragmentation External fragmentation

Predictable time and memory consumption Fast compile-time

Stephanie Stroka Self-Collecting Mutators are Self-Compacting

Motivation Periodic Memory Consumption Analysis Implementation Results Outlook

The problem

Fast allocation ֒ → wasting memory Less fragmentation ֒ → slow allocation OptimalMemoryConsumption ∼

1 AllocationTime

Predictable memory consumption ֒ → unreasonable run-time (memory copying) ֒ → unreasonable compile-time (program analysis?!)

Stephanie Stroka Self-Collecting Mutators are Self-Compacting

Motivation Periodic Memory Consumption Analysis Implementation Results Outlook

Typical program characteristics [1]

time memory malloc() free()

Figure 1: Periodic memory allocation and deallocation Stephanie Stroka Self-Collecting Mutators are Self-Compacting

Motivation Periodic Memory Consumption Analysis Implementation Results Outlook

Memory consumption in Garbage Collectors

time memory GC

Figure 2: Garbage Collector loses periodic memory characteristics Stephanie Stroka Self-Collecting Mutators are Self-Compacting

Motivation Periodic Memory Consumption Analysis Implementation Results Outlook

Periodic memory allocation in the heap

Bumppointer Long- Living- Memory Long- Living- Memory Mid- Living- Memory malloc() Bumppointer Long- Living- Memory Long- Living- Memory Mid- Living- Memory malloc() Short- Living- Memory Short- Living- Memory Short- Living- Memory Bumppointer Long- Living- Memory Long- Living- Memory Mid- Living- Memory malloc()

Figure 3: Periodic memory allocation and deallocation in the heap Stephanie Stroka Self-Collecting Mutators are Self-Compacting

Motivation Periodic Memory Consumption Analysis Implementation Results Outlook

Typical program characteristics [2]

time memory malloc() free()

Figure 4: Periodic memory allocation with growing list Stephanie Stroka Self-Collecting Mutators are Self-Compacting

Motivation Periodic Memory Consumption Analysis Implementation Results Outlook

Semi-periodic memory allocation in the heap

Bumppointer Long- Living- Memory Long- Living- Memory Mid- Living- Memory malloc() Bumppointer Long- Living- Memory Long- Living- Memory Mid- Living- Memory malloc() Short- Living- Memory Short- Living- Memory Bumppointer Long- Living- Memory Long- Living- Memory Mid- Living- Memory malloc() Mid- Living- Memory Mid- Living- Memory

Figure 5: Periodic memory allocation with growing list in the heap Stephanie Stroka Self-Collecting Mutators are Self-Compacting

Motivation Periodic Memory Consumption Analysis Implementation Results Outlook

Analyzing techniques

To analyze periodic memory consumption, we need... an allocator that provides information about fragmentation ֒ → Compact-Fit a mutator that provides time information ֒ → Self-Collecting Mutator

Stephanie Stroka Self-Collecting Mutators are Self-Compacting

Motivation Periodic Memory Consumption Analysis Implementation Results Outlook

Binding memory life-time to a clock

time memory tick() t1 t2 t3 t4 t5 t6 t7 t8 refresh(7) refresh(4) refresh(2) tick() tick() tick() tick() tick() tick() tick() refresh(0)

Figure 6: Periodic memory allocation with self-collecting mutators Stephanie Stroka Self-Collecting Mutators are Self-Compacting

Motivation Periodic Memory Consumption Analysis Implementation Results Outlook

Implementation

SCM eager collection ֒ → all expired objects will be deallocated at scm tick() SCM‘s malloc function uses cf malloc() instead of ptmalloc2() Example program with scm malloc(), scm refresh() and scm tick()

stm stm mtm ltm stm stm mtm ltm stm stm mtm ltm tick() tick() tick() tick() tick() tick() tick()

Figure 7: Analyzing memory consumption behaviour with growing #ticks Stephanie Stroka Self-Collecting Mutators are Self-Compacting

Motivation Periodic Memory Consumption Analysis Implementation Results Outlook

Results

Runtime increases with growing # ticks ֒ → tick(), refresh() and compaction() Memory consumption decreases with # ticks ֒ → max. fragmentation = #free() ֒ → ...unlikely, because of periodic characteristic Compaction causes compaction

Long- Living- Memory Long- Living- Memory Mid- Living- Memory Mid- Living- Memory Mid- Living- Memory Short- Living- Memory Short- Living- Memory Short- Living- Memory Short- Living- Memory

Figure 8: Compaction causes compaction Stephanie Stroka Self-Collecting Mutators are Self-Compacting

Motivation Periodic Memory Consumption Analysis Implementation Results Outlook

Results

1 tick

k = 1, compaction = 160 k = 2, compaction = 520 k = 3, compaction = 0 k = 4, compaction = 0 ...

3 ticks

k = 1, compaction = 282 k = 2, compaction = 12 k = 3, compaction = 0 k = 4, compaction = 0

Stephanie Stroka Self-Collecting Mutators are Self-Compacting

Motivation Periodic Memory Consumption Analysis Implementation Results Outlook

Outlook

Analyze k for non-trivial programs Find optimal middle ground for memory consumption and compaction overhead Formalize relation of k and refresh()/tick()

Stephanie Stroka Self-Collecting Mutators are Self-Compacting