Implementing Garbage Collection for Active Objects on Top of Erlang - PowerPoint PPT Presentation

Implementing Garbage Collection for Active Objects on Top of Erlang Sigmund Hansen Master thesis autumn 2014

Problem ◮ ABS is a modeling language for distributed systems ◮ ABS has an Erlang back end ◮ Back end cannot collect garbage processes

Goals The garbage collector should be: ◮ Fast It should not considerably slow down simulations ◮ Comprehensive It should collect most or all garbage ◮ Correct It should not collect resources required later

ABS Overview ◮ Executable models ◮ Functional subset ◮ Object-oriented ◮ Concurrency: ◮ Active objects ◮ Scheduling in Concurrent Object Groups (COGs) ◮ RPC with message passing

Erlang Overview ◮ Concurrency: ◮ Actor model ◮ Message passing ◮ Fault-tolerant ◮ Functional

Garbage Collection Overview Automatic memory management for dynamically allocated memory. ◮ Reference counting ◮ Tracing ◮ Object graph traversal ◮ De-allocation of unmarked objects

Active Object Collection ◮ Kept alive by associated process ◮ Must kill the process ◮ High-level implementation

Choice of Algorithm ◮ Reference counting cannot collect cyclic garbage. ◮ Killing processes, not collecting memory, i.e. no moving collectors ◮ Greater risk with non-standard collection algorithms Mark and sweep that stops the world implemented.

Stopping the World ◮ No interrupts ◮ Stop by messaging ◮ Tasks and scheduling must stop

Blocking Operations ◮ Blocking on futures ◮ Blocking on object instantiation ◮ Inappropriate COG state during block

COG State Machine (Before) Not running new task, token, [runnable task found] [no runnable task found] task state changed, task error task error Running None found new task, task state changed

COG State Machine (After) ok Not0running stop0world new0task, task0state0changed, resume0world [no0runnable0task0found] task0error, [reference0count0!=00, inc/dec0reference0count Resume0state0=0Not0running] None0found stop0world stop0world Stopped token, get0references, [runnable0task0found] task0error inc/dec0reference0count Resume0state resume0world resume0world [reference0count0!=00, stop0world [reference0count0=00] Resume0state0=0Blocked] Blocked new0task, stop0world task0state0change, Blocked0task inc/dec0reference0count running0task0state blocked0task0state changed0to0blocked changed0to0runnable Running new0task, task0state0change, inc/dec0reference0count

Stopping Running Tasks ◮ Will reduce waiting ◮ COG messages task ◮ Task then blocks

Asynchronous Method Calls on Inactive Objects ◮ Task initialization is synchronous ◮ Waits until object is activated ◮ Respond to GC messages while waiting

Tasks Created While the World is Stopping ◮ Task is created by a future ◮ Stopped COGs do not receive task creation messages ◮ Futures temporarily become roots and keep parameters

Future State Machine (Before) init Await_start Task Parameters Loop completed Task Started Result error Await_completion

Future State Machine (After) init Await_start get_references Started Wait Parameters get_references Started Task Task complete error die Loop Result

A Complete View of COGs and Futures Why is it required?

Marking Phase ◮ Messages all grays in parallel ◮ B N + 1 = B N ∪ G N ◮ G N + 1 = References ( G N ) \ B N + 1 ◮ Sweeps when gray set is empty

Sweeping ◮ Identifies white objects ◮ Identifies white futures ◮ Sends message to kill them ◮ Resume message sent to COGs

Collecting COGs with Reference Counting ◮ Counts objects left in group ◮ COG stops if world resumes and it is empty ◮ Messages garbage collector

Static analysis ◮ Types that may contain references ◮ Reaching unawaited futures ◮ Other potential analyses: ◮ Loops with synchronization points ◮ Deep functions

Triggering Garbage Collection When should garbage be collected: ◮ On time? ◮ Number of objects and futures? ◮ Process ratio?

Data Collection ◮ Garbage collector state changes ◮ Number of objects and futures swept ◮ Memory usage and number of objects, futures and COGs

Test Cases ◮ Ping pong Basic test with cyclic garbage ◮ Sequences Unstoppable loop ◮ Prime Sieve Long-running tasks ◮ MapReduce - Indexing Real-world test case

Ping Pong - Time Collection scheme Real time Relative Relative without idle Before GC 1274 ms 100.0 % 100.0 % Never collect 1257 ms 98.6 % 93.5 % Always collect 1854 ms 145.5 % 311.3 % Collect on count 1293 ms 101.4 % 106.7 % Collect on time 1261 ms 98.9 % 94.9 % + Stop running 1265 ms 99.2 % 96.5 % Collect on either 1290 ms 101.2 % 105.5 % + Stop running 1298 ms 101.8 % 108.4 %

Ping Pong - No GC 750 500 Total 250 variable 0 cogs objects futures 18 Memory in MiB 16 14 0 ms 100 ms 200 ms 300 ms 400 ms 500 ms Time

Ping Pong - Always collecting 300 200 Total 100 variable 0 cogs 22.5 objects futures 20.0 Memory in MiB 17.5 15.0 12.5 0 ms 1220 ms 2440 ms 3660 ms 4880 ms 6100 ms Time

Ping Pong - By count 60 40 Total 20 variable 0 cogs objects futures 15.5 15.0 Memory in MiB 14.5 14.0 13.5 0 ms 140 ms 280 ms 420 ms 560 ms 700 ms Time

Ping Pong - Timed 200 150 Total 100 50 variable 0 cogs 16.5 objects futures 16.0 15.5 Memory in MiB 15.0 14.5 14.0 13.5 0 ms 100 ms 200 ms 300 ms 400 ms 500 ms Time

Infinite Ping Pong - No GC 150000 100000 Total 50000 variable 0 cogs 1200 objects futures 900 Memory in MiB 600 300 0 0 ms 31120 ms 62240 ms 93360 ms 124480 ms 155600 ms Time

Infinite Ping Pong - Timed 8000 6000 Total 4000 2000 variable 0 cogs objects futures 100 Memory in MiB 50 0 ms 11900 ms 23800 ms 35700 ms 47600 ms 59500 ms Time

Sequences - Time Collection scheme Real time Relative Relative without idle Before GC 1395 ms 100.0 % 100.0 % Never collect 1439 ms 103.2 % 111.2 % Always collect 39159 ms 2807.2 % 9662.1 % Collect on count 16801 ms 1204.4 % 4001.0 % Collect on time 1622 ms 116.3 % 157.5 % + Stop running 1530 ms 109.7 % 134.1 % Collect on either 17054 ms 1222.6 % 4065.1 % + Stop running 12614 ms 904.3 % 2940.8 %

Sequences - No GC 5000 4000 3000 Total 2000 1000 variable 0 cogs 70 objects futures 60 50 Memory in MiB 40 30 20 0 ms 720 ms 1440 ms 2160 ms 2880 ms 3600 ms Time

Sequences - Timed 5000 4000 3000 Total 2000 1000 variable 0 cogs objects 60 futures 50 Memory in MiB 40 30 20 0 ms 860 ms 1720 ms 2580 ms 3440 ms 4300 ms Time

Sequences - Timed stopping running processes 600 400 Total 200 variable 0 cogs objects futures 24 Memory in MiB 20 16 0 ms 700 ms 1400 ms 2100 ms 2800 ms 3500 ms Time

Prime Sieve - Time Collection scheme Real time Relative Relative without idle Before GC 1448 ms 100.0 % 100.0 % Never collect 1453 ms 100.3 % 101.1 % Always collect 1903 ms 131.4 % 201.6 % Collect on count 1482 ms 102.4 % 107.6 % Collect on time 1454 ms 100.4 % 101.3 % + Stop running 1542 ms 106.5 % 120.9 % Collect on either 1472 ms 101.6 % 105.3 % + Stop running 1545 ms 106.7 % 121.6 %

Indexing - Time Collection scheme Real time Relative Relative without idle Before GC 1356 ms 100.0 % 100.0 % Never collect 1373 ms 101.2 % 104.7 % Always collect 5043 ms 371.8 % 1134.2 % Collect on count 1408 ms 103.9 % 114.7 % Collect on time 1398 ms 103.1 % 111.7 % + Stop running 1405 ms 103.6 % 113.9 % Collect on either 1410 ms 104.0 % 115.3 % + Stop running 1434 ms 105.7 % 121.9 %

Conclusion Timed trigger gives acceptable results. ◮ Acceptably fast ◮ Mark and sweep is complete ◮ Measures taken stopping the world to ensure correctness

Discussion Testing with infinite loops have discouraging results. Future work to lower overhead may be needed: ◮ Optimize stopping the world ◮ Optimize marking ◮ Concurrent collection ◮ Change of algorithm