This section provides usage scenarios of memory-usage-analyzer with an example workloads
This is prerequisite for all the examples. This example workload prepares a large memory segment. A portion of this memory segment is accessed (50% or 75% of the total) to create the memory regions which are the more frequently accessed (hot) and less frequently accessed (cold).
cd tests/example
make
./workload
# Sample output
usage ./workload: total_memory(GiB) hot_timer loopsThe parameters:
total_memory: amount of memory used by the workload in GiBhot_timer: time period spend in 50% and 75% hot memory modes in secondsloops: controls the amount of work done, if 0 the workload runs forever
Here's an example run of the workload:
./workload 8 10 1000
# Sample output
[INFO] (main:34) : pid = 20141
[INFO] (main:35) : memory = 8 GiB = 2097152 pages
[INFO] (main:36) : hot timer = 10 sec
[INFO] (main:37) : loops = 1000
[INFO] (main:51) : Done with setup in 3 seconds.
[INFO] (sigalrm_handler:17) : hot pages = 75%
[INFO] (sigalrm_handler:17) : hot pages = 50%
[INFO] (main:68) : Done with test in 21 seconds.There are three different scenarios that are provided here. Please make sure that Prepare the example workload steps are complete.
- Baseline - Running baseline without any memory pressure and collecting the memory usage statistics
- Dynamic Squeezer - Running workload under a Cgroup to create memory pressure, where the memory limit is adjusted dynamically to limit the page fault rate
- Static Squeezer - Running workload under a Cgroup to create memory pressure, where the memory limit to fixed
This is an example of running baseline workload. Before running the workload, system configuration should be performed to enable the data collection. For system configuration details, please refer config
cd tests/example
# Configure the system for data collection
config.py
# Configuration ZSWAP compressor
zswapcompconfig.py -c lzo-rle
memory_usage_analyzer.py ./workload 8 60 10000
# Sample results, actual results may vary
2024-04-01 22:31:19,322 - memoryusageanalyzer - INFO - args = Namespace(cgname='memoryusageanalyzer', cgpath='/sys/fs/cgroup', cmd='./workload', docker=None, options=['8', '60', '10000'], output='profile', outputforce=None, reclaimerconfig=None, sampleperiod=5, verbose=False)
2024-04-01 22:31:19,322 - memoryusageanalyzer - INFO - **** Storing profiling results in profile
2024-04-01 22:31:19,326 - memoryusageanalyzer - INFO - cgroup v2 enabled and memory controller detected
2024-04-01 22:31:19,330 - memoryusageanalyzer - INFO - **** Starting job in cgroup
2024-04-01 22:31:19,332 - memoryusageanalyzer - INFO - **** Starting stats
[INFO] (main:43) : pid = 26557
[INFO] (main:44) : memory = 8 GiB = 2097152 pages
[INFO] (main:45) : hot timer = 60 sec
[INFO] (main:46) : loops = 10000
[INFO] (main:60) : Done with setup in 2 seconds.
[INFO] (sigalrm_handler:25) : hot pages = 75%
[INFO] (sigalrm_handler:25) : hot pages = 50%
[INFO] (sigalrm_handler:25) : hot pages = 75%
[INFO] (main:77) : Done with test in 215 seconds.
2024-04-01 22:34:56,845 - memoryusageanalyzer - INFO - **** Job finished
2024-04-01 22:34:56,846 - memoryusageanalyzer - INFO - **** Stopping stats
**** Closing profile/baseline/stats.csv.gz
profile/baseline
Generating plots profile/baseline/memoryusageanalyzer-plots.html
Loading stats from profile/baseline/stats.csv.gz
Loading stats from profile/baseline/stats.csv.gz
*** Totals without page cache ***
Maximum active+swap memory = 8.051 GiB
Max active memory = 8.0 GiB
Max swap memory = 0.0 GiB
Max page cache = 0.014 GiB
Swap memory = 0.3%
Max inactive anon = 8.005 GiB
Maximum zpool memory = 0.027439 GiB
Maximum zram memory = 0.000011 GiB
incompressible data ratio(zram)= 0.0%
full total memory pressure = 0 us
Average swap memory comp ratio = 1.01
Potential maximum compression savings in percent = 0.0%
Potential median compression savings in percent = 0.0%
Potential compression savings in bytes = 0.000 GiB
Average swap memory comp ratio = 1.01 (including same filled pages)
Major PFs/sec (median, avg, max) = 0.00 0.00 0.00
Total PFs/sec (median, avg, max) = 0.00 9130.44 392608.59
Total PFs (Major+Minor) = 1964235
Major PFs = 0
sampling period = 5.003033671268197The analyzer output is also saved at profile/baseline/analyze.out. The visualization of the memory usage will be at profile/baseline/memoryusageanalyzer-plots.html as shown below.
This is an example where dynamic memory-limit is applied to workload through cgroup with page-fault rate as the metric. The memory limit is applied gradually to the workload till the page-fault-rate reaches the threshold. After that, dynamic squeezer adjusts the memory limit to keep the page-fault-rate within the threshold by relaxing the memory-limit. Please refer to the config file for dynamic squeezer in the repo memory-usage-analyzer/src/reclaimer/
With dynamic squeezer
cd tests/example
config.py
zswapcompconfig.py -c lzo-rle
memory_usage_analyzer.py -r ../../src/reclaimer/squeezerdynamicconfig.json ./workload 8 60 10000
# Sample results, actual results may vary
2024-04-01 22:12:13,171 - memoryusageanalyzer - INFO - args = Namespace(cgname='memoryusageanalyzer', cgpath='/sys/fs/cgroup', cmd='./workload', docker=None, options=['8', '60', '10000'], output='profile', outputforce=None, reclaimerconfig='../../src/reclaimer/squeezerdynamicconfig.json', sampleperiod=5, verbose=False)
2024-04-01 22:12:13,171 - memoryusageanalyzer - INFO - **** Storing profiling results in profile.6
2024-04-01 22:12:13,227 - memoryusageanalyzer - INFO - cgroup v2 enabled and memory controller detected
2024-04-01 22:12:13,231 - memoryusageanalyzer - INFO - **** Starting job in cgroup
2024-04-01 22:12:13,234 - memoryusageanalyzer - INFO - **** Starting stats
2024-04-01 22:12:13,237 - memoryusageanalyzer - INFO - **** Starting reclaimer
memory usage = 46,280,704 pf_rate = 0.2 => squeeze percent = 1%
squeeze to 45,817,896 -- ok
[INFO] (main:43) : pid = 18834
[INFO] (main:44) : memory = 8 GiB = 2097152 pages
[INFO] (main:45) : hot timer = 60 sec
[INFO] (main:46) : loops = 10000
[INFO] (main:60) : Done with setup in 2 seconds.
memory usage = 8,645,734,400 pf_rate = 0.0 => squeeze percent = 2%
squeeze to 8,472,819,712 -- ok
memory usage = 8,463,917,056 pf_rate = 16.2 => squeeze percent = 3%
squeeze to 8,209,999,544 -- ok
memory usage = 8,208,674,816 pf_rate = 0.0 => squeeze percent = 4%
squeeze to 7,880,327,823 -- ok
memory usage = 7,880,237,056 pf_rate = 0.0 => squeeze percent = 5%
squeeze to 7,486,225,203 -- ok
memory usage = 7,486,009,344 pf_rate = 0.0 => squeeze percent = 6%
squeeze to 7,036,848,783 -- ok
memory usage = 7,036,510,208 pf_rate = 0.0 => squeeze percent = 7%
squeeze to 6,543,954,493 -- ok
memory usage = 6,543,728,640 pf_rate = 0.0 => squeeze percent = 8%
squeeze to 6,020,230,348 -- ok
memory usage = 6,410,338,304 pf_rate = 43918.6 => squeeze percent = 0%
memory usage = 6,409,936,896 pf_rate = 0.0 => squeeze percent = 1%
squeeze to 6,345,837,527 -- ok
memory usage = 6,409,601,024 pf_rate = 5169.2 => squeeze percent = 2%
squeeze to 6,281,409,003 -- ok
memory usage = 6,409,342,976 pf_rate = 11181.2 => squeeze percent = 0%
memory usage = 6,409,342,976 pf_rate = 0.0 => squeeze percent = 1%
squeeze to 6,345,249,546 -- ok
[INFO] (sigalrm_handler:25) : hot pages = 75%
memory usage = 7,516,831,744 pf_rate = 110002.8 => squeeze percent = 0%
memory usage = 7,516,831,744 pf_rate = 0.0 => squeeze percent = 1%
squeeze to 7,441,663,426 -- ok
memory usage = 7,516,459,008 pf_rate = 6201.4 => squeeze percent = 2%
squeeze to 7,366,129,827 -- ok
memory usage = 7,516,676,096 pf_rate = 16208.0 => squeeze percent = 0%
memory usage = 7,516,622,848 pf_rate = 0.0 => squeeze percent = 1%
squeeze to 7,441,456,619 -- ok
memory usage = 7,516,762,112 pf_rate = 6273.8 => squeeze percent = 2%
squeeze to 7,366,426,869 -- ok
memory usage = 7,516,549,120 pf_rate = 15865.8 => squeeze percent = 0%
memory usage = 7,516,549,120 pf_rate = 0.0 => squeeze percent = 1%
squeeze to 7,441,383,628 -- ok
memory usage = 7,516,581,888 pf_rate = 6191.6 => squeeze percent = 2%
squeeze to 7,366,250,250 -- ok
memory usage = 7,516,581,888 pf_rate = 15659.6 => squeeze percent = 0%
memory usage = 7,516,581,888 pf_rate = 0.0 => squeeze percent = 1%
squeeze to 7,441,416,069 -- ok
memory usage = 7,516,712,960 pf_rate = 6257.4 => squeeze percent = 2%
squeeze to 7,366,378,700 -- ok
[INFO] (sigalrm_handler:25) : hot pages = 50%
memory usage = 7,516,700,672 pf_rate = 15766.8 => squeeze percent = 0%
memory usage = 7,516,700,672 pf_rate = 0.0 => squeeze percent = 1%
squeeze to 7,441,533,665 -- ok
memory usage = 7,441,297,408 pf_rate = 0.0 => squeeze percent = 2%
squeeze to 7,292,471,459 -- ok
memory usage = 7,291,830,272 pf_rate = 0.0 => squeeze percent = 3%
squeeze to 7,073,075,363 -- ok
memory usage = 7,138,152,448 pf_rate = 7081.8 => squeeze percent = 4%
squeeze to 6,852,626,350 -- ok
memory usage = 6,911,389,696 pf_rate = 4600.8 => squeeze percent = 5%
squeeze to 6,565,820,211 -- ok
memory usage = 6,565,621,760 pf_rate = 0.0 => squeeze percent = 6%
squeeze to 6,171,684,454 -- ok
memory usage = 6,409,621,504 pf_rate = 21559.4 => squeeze percent = 0%
memory usage = 6,409,515,008 pf_rate = 0.0 => squeeze percent = 1%
squeeze to 6,345,419,857 -- ok
memory usage = 6,409,342,976 pf_rate = 5136.4 => squeeze percent = 2%
squeeze to 6,281,156,116 -- ok
memory usage = 6,409,166,848 pf_rate = 11203.6 => squeeze percent = 0%
memory usage = 6,409,166,848 pf_rate = 0.0 => squeeze percent = 1%
squeeze to 6,345,075,179 -- ok
[INFO] (sigalrm_handler:25) : hot pages = 75%
memory usage = 7,516,372,992 pf_rate = 109975.2 => squeeze percent = 0%
memory usage = 7,516,372,992 pf_rate = 0.0 => squeeze percent = 1%
squeeze to 7,441,209,262 -- ok
memory usage = 7,516,700,672 pf_rate = 6243.2 => squeeze percent = 2%
squeeze to 7,366,366,658 -- ok
memory usage = 7,516,553,216 pf_rate = 13308.0 => squeeze percent = 0%
memory usage = 7,516,323,840 pf_rate = 0.0 => squeeze percent = 1%
squeeze to 7,441,160,601 -- ok
memory usage = 7,516,733,440 pf_rate = 6277.2 => squeeze percent = 2%
squeeze to 7,366,398,771 -- ok
memory usage = 7,516,246,016 pf_rate = 14325.8 => squeeze percent = 0%
memory usage = 7,516,241,920 pf_rate = 0.0 => squeeze percent = 1%
squeeze to 7,441,079,500 -- ok
memory usage = 7,516,504,064 pf_rate = 6300.6 => squeeze percent = 2%
squeeze to 7,366,173,982 -- ok
memory usage = 7,516,651,520 pf_rate = 13345.0 => squeeze percent = 0%
memory usage = 7,516,651,520 pf_rate = 0.0 => squeeze percent = 1%
squeeze to 7,441,485,004 -- ok
memory usage = 7,516,610,560 pf_rate = 6166.6 => squeeze percent = 2%
squeeze to 7,366,278,348 -- ok
[INFO] (sigalrm_handler:25) : hot pages = 50%
memory usage = 7,516,684,288 pf_rate = 14280.6 => squeeze percent = 0%
memory usage = 7,516,479,488 pf_rate = 0.0 => squeeze percent = 1%
squeeze to 7,441,314,693 -- ok
[INFO] (main:77) : Done with test in 251 seconds.
2024-04-01 22:16:27,252 - memoryusageanalyzer - INFO - **** Job finished
2024-04-01 22:16:27,252 - memoryusageanalyzer - INFO - **** Stopping stats
memory usage = 29,151,232 pf_rate = 50.2 => squeeze percent = 2%
squeeze to 28,568,207 -- ok
**** Closing profile.6/dynamic/stats.csv.gz
2024-04-01 22:16:28,527 - memoryusageanalyzer - INFO - **** Stopping squeezer
Generating plots profile.6/dynamic/memoryusageanalyzer-plots.html
**** Squeezer exiting
profile.6/dynamic
Loading stats from profile.6/dynamic/stats.csv.gz
Loading stats from profile.6/dynamic/stats.csv.gz
*** Totals without page cache ***
Maximum active+swap memory = 8.044 GiB
Max active memory = 8.0 GiB
Max swap memory = 4.3 GiB
Max page cache = 0.014 GiB
Swap memory = 53.1%
Max inactive anon = 8.005 GiB
Maximum zpool memory = 4.187023 GiB
Maximum zram memory = 0.000011 GiB
incompressible data ratio(zram)= 0.0%
full total memory pressure = 5578899 us
Average swap memory comp ratio = 1.00
Potential maximum compression savings in percent = 0.0%
Potential median compression savings in percent = 0.1%
Potential compression savings in bytes = 0.000 GiB
Average swap memory comp ratio = 1.00 (including same filled pages)
Major PFs/sec (median, avg, max) = 6254.82 11343.99 107242.28
Total PFs/sec (median, avg, max) = 7407.01 19192.46 392421.31
Total PFs (Major+Minor) = 4800634
Major PFs = 2837486
sampling period = 5.002625856399536
A memory usage report is generated after the profiling runs are complete. The default results path is profile* and we can also override the default path using the -o option.
This is an example where static memory-limit is applied to workload through cgroup. The fixed memory limit on Cgroup can be a percentage of the maximum baseline memory usage.
Please refer to the static squeezer config file in memory-usage-analyzer/src/reclaimer/squeezerstaticconfig.json. In this example, there are multiple runs - baseline runs followed by memory-limit runs with memory-limit set as 2%, 4%, 8% and 16% of the maximum baseline memory usage.
With static squeezer
cd tests/example
config.py
zswapcompconfig.py -c lzo-rle
memory_usage_analyzer.py -r ../../src/reclaimer/squeezerstaticconfig.json ./workload 8 60 10000
# Sample results for baseline
2024-04-01 21:44:48,272 - memoryusageanalyzer - INFO - args = Namespace(cgname='memoryusageanalyzer', cgpath='/sys/fs/cgroup', cmd='./workload', docker=None, options=['8', '60', '10000'], output='profile', outputforce=None, reclaimerconfig='../../src/reclaimer/squeezerstaticconfig.json', sampleperiod=5, verbose=False)
2024-04-01 21:44:48,273 - memoryusageanalyzer - INFO - **** Storing profiling results in profile.5
2024-04-01 21:44:48,328 - memoryusageanalyzer - INFO - cgroup v2 enabled and memory controller detected
2024-04-01 21:44:48,332 - memoryusageanalyzer - INFO - **** Starting job in cgroup
2024-04-01 21:44:48,334 - memoryusageanalyzer - INFO - **** Starting stats
2024-04-01 21:44:48,337 - memoryusageanalyzer - INFO - **** Starting reclaimer
[INFO] (main:43) : pid = 9787
[INFO] (main:44) : memory = 8 GiB = 2097152 pages
[INFO] (main:45) : hot timer = 60 sec
[INFO] (main:46) : loops = 10000
[INFO] (main:60) : Done with setup in 2 seconds.
[INFO] (sigalrm_handler:25) : hot pages = 75%
[INFO] (sigalrm_handler:25) : hot pages = 50%
[INFO] (sigalrm_handler:25) : hot pages = 75%
[INFO] (main:77) : Done with test in 225 seconds.
2024-04-01 21:48:36,162 - memoryusageanalyzer - INFO - **** Job finished
2024-04-01 21:48:36,162 - memoryusageanalyzer - INFO - **** Stopping stats
**** Closing profile.5/staticsweep/stats.csv.gz
2024-04-01 21:48:38,648 - memoryusageanalyzer - INFO - **** Stopping squeezer
Generating plots profile.5/staticsweep/memoryusageanalyzer-plots.html
2024-04-01 21:58:57,581 - memoryusageanalyzer - INFO - cgroup v2 enabled and memory controller detected
2024-04-01 21:58:57,585 - memoryusageanalyzer - INFO - **** Starting job in cgroup
2024-04-01 21:58:57,587 - memoryusageanalyzer - INFO - **** Starting stats
2024-04-01 21:58:57,589 - memoryusageanalyzer - INFO - **** Starting reclaimer
[INFO] (main:43) : pid = 14857
[INFO] (main:44) : memory = 8 GiB = 2097152 pages
[INFO] (main:45) : hot timer = 60 sec
[INFO] (main:46) : loops = 10000
[INFO] (main:60) : Done with setup in 5 seconds.
[INFO] (sigalrm_handler:25) : hot pages = 75%
[INFO] (sigalrm_handler:25) : hot pages = 50%
[INFO] (sigalrm_handler:25) : hot pages = 75%
[INFO] (main:77) : Done with test in 238 seconds.
2024-04-01 22:03:00,810 - memoryusageanalyzer - INFO - **** Job finished
2024-04-01 22:03:00,811 - memoryusageanalyzer - INFO - **** Stopping stats
**** Closing profile.5/staticsweep/memory_8/stats.csv.gz
2024-04-01 22:03:02,872 - memoryusageanalyzer - INFO - **** Stopping squeezer
Generating plots profile.5/staticsweep/memory_8/memoryusageanalyzer-plots.html
**** Squeezer exiting
profile.5/staticsweep
Loading stats from profile.5/staticsweep/stats.csv.gz
Loading stats from profile.5/staticsweep/stats.csv.gz
*** Totals without page cache ***
Maximum active+swap memory = 8.038 GiB
Max active memory = 8.0 GiB
Max swap memory = 0.0 GiB
Max page cache = 0.014 GiB
Swap memory = 0.0%
Max inactive anon = 8.005 GiB
Maximum zpool memory = 0.000000 GiB
Maximum zram memory = 0.000011 GiB
incompressible data ratio(zram)= 0.0%
full total memory pressure = 0 us
Major PFs/sec (median, avg, max) = 0.00 0.00 0.00
Total PFs/sec (median, avg, max) = 0.00 8636.03 388621.24
Total PFs (Major+Minor) = 1944351
Major PFs = 0
sampling period = 5.003200546900431
**** Using max_limit from baseline run: 8630243328.0 = 8.038 GiB
**** Initial memory limit = 7939823861
**** Squeezer exiting
# Sample output from 8% memory reduction as a reference. Other memory limit runs will have similar outputs
profile.5/staticsweep/memory_8
Loading stats from profile.5/staticsweep/memory_8/stats.csv.gz
Loading stats from profile.5/staticsweep/memory_8/stats.csv.gz
*** Totals without page cache ***
Maximum active+swap memory = 7.395 GiB
Max active memory = 6.1 GiB
Max swap memory = 1.3 GiB
Max page cache = 0.014 GiB
Swap memory = 17.3%
Max inactive anon = 6.751 GiB
Maximum zpool memory = 1.277714 GiB
Maximum zram memory = 0.000011 GiB
incompressible data ratio(zram)= 0.0%
full total memory pressure = 1204429 us
Average swap memory comp ratio = 1.00
Potential maximum compression savings in percent = 0.1%
Potential median compression savings in percent = 0.1%
Potential compression savings in bytes = 0.005 GiB
Average swap memory comp ratio = 1.00 (including same filled pages)
Major PFs/sec (median, avg, max) = 0.00 6550.13 130470.80
Total PFs/sec (median, avg, max) = 0.00 14762.16 415849.90
Total PFs (Major+Minor) = 3544846
Major PFs = 1572886
sampling period = 5.002720882495244There are multiple directories that are created for each memory-limit runs. Here is one example from 8% memory-limit run.
A memory usage report is generated after the profiling runs are complete. The default results path is profile* and we can also override the default path using the -o option.