Привет!
Очень много в интернете позитива/негатива об EBS. Так же не мало тестов на их производительность. Я решил сам провести несколько тестов и сделать небольшое исследование по этому поводу.
Итак, в тесте участвовала машинка m1.large к которой были примаунчены диски:
Были проведены несколько тестов:
Результаты вывода в консоль:
1. Сравнения скорости чтения/записи через dd, Mb/s:
2. Количество запросов в секунду через sysbench, Requests/s:
Итак, похоже быстрее всех работает именно Ephemeral, как и ожидалось, но IOPS к нему самый низкий.
Скорость соединения со стандартными EBS выше, чем к оптимизированным, даже с 2к IOPS. Но количество IOPS больше у оптимизированных EBS, хотя RAID по IOPS всех победил.
В итоге, как и следовало ожидать, по количеству операций в секунду и простоте настроек оптимизированные EBS лучше, хотя использование рейдов из них будет ещё быстродейственнее.
Если вам важна скорость обращения к файлу — используйте ephemeral или стандартные EBS.
Очень много в интернете позитива/негатива об EBS. Так же не мало тестов на их производительность. Я решил сам провести несколько тестов и сделать небольшое исследование по этому поводу.
Итак, в тесте участвовала машинка m1.large к которой были примаунчены диски:
- EBS Standard, 100 GB
- EBS IO-1 500 IOPS, 100 GB
- EBS IO-1 1000 IOPS, 100 GB
- EBS IO-1 2000 IOPS, 200 GB
- 8xEBS Standard, 30 GB, RAID 10
- Ephemeral, 450 GB
Были проведены несколько тестов:
# hdparm -tT /dev/xcdX
# dd if=/dev/zero of=tempfile bs=5M count=1024 conv=fdatasync,notrunc
# echo 3 > /proc/sys/vm/drop_caches
# dd if=tempfile of=/dev/null bs=5M count=1024
# sysbench --num-threads=16 --test=fileio --file-total-size=5G --file-test-mode=rndrw prepare
# sysbench --num-threads=16 --test=fileio --file-total-size=5G --file-test-mode=rndrw run
Результаты вывода в консоль:
EBS Standard, 100 GB
# hdparm -tT /dev/xvdj
/dev/xvdj:
Timing cached reads: 4866 MB in 2.00 seconds = 2428.53 MB/sec
Timing buffered disk reads: 242 MB in 3.00 seconds = 80.54 MB/sec
Timing cached reads: 5146 MB in 2.00 seconds = 2579.25 MB/sec
Timing buffered disk reads: 294 MB in 3.01 seconds = 97.59 MB/sec
Timing cached reads: 4870 MB in 2.00 seconds = 2440.55 MB/sec
Timing buffered disk reads: 306 MB in 3.00 seconds = 101.89 MB/sec
# dd if=/dev/zero of=tempfile bs=5M count=1024 conv=fdatasync,notrunc
1024+0 records in
1024+0 records out
5368709120 bytes (5.4 GB) copied, 161.222 s, 33.3 MB/s
# echo 3 > /proc/sys/vm/drop_caches
# dd if=tempfile of=/dev/null bs=5M count=1024
1024+0 records in
1024+0 records out
5368709120 bytes (5.4 GB) copied, 86.4683 s, 62.1 MB/s
# sysbench --num-threads=16 --test=fileio --file-total-size=5G --file-test-mode=rndrw run
sysbench 0.4.12: multi-threaded system evaluation benchmark
Running the test with following options:
Number of threads: 16
Extra file open flags: 0
128 files, 40Mb each
5Gb total file size
Block size 16Kb
Number of random requests for random IO: 10000
Read/Write ratio for combined random IO test: 1.50
Periodic FSYNC enabled, calling fsync() each 100 requests.
Calling fsync() at the end of test, Enabled.
Using synchronous I/O mode
Doing random r/w test
Threads started!
Done.
Operations performed: 6000 Read, 4000 Write, 12800 Other = 22800 Total
Read 93.75Mb Written 62.5Mb Total transferred 156.25Mb (69.816Mb/sec)
4468.25 Requests/sec executed
Test execution summary:
total time: 2.2380s
total number of events: 10000
total time taken by event execution: 1.5942
per-request statistics:
min: 0.01ms
avg: 0.16ms
max: 65.00ms
approx. 95 percentile: 0.02ms
Threads fairness:
events (avg/stddev): 625.0000/249.48
execution time (avg/stddev): 0.0996/0.03
EBS IO-1 500 IOPS, 100 GB
# hdparm -tT /dev/xvdh
/dev/xvdh:
Timing cached reads: 4314 MB in 2.00 seconds = 2161.08 MB/sec
Timing buffered disk reads: 72 MB in 3.05 seconds = 23.57 MB/sec
Timing cached reads: 3646 MB in 2.00 seconds = 1826.09 MB/sec
Timing buffered disk reads: 76 MB in 3.04 seconds = 25.01 MB/sec
Timing cached reads: 4346 MB in 2.00 seconds = 2175.61 MB/sec
Timing buffered disk reads: 76 MB in 3.03 seconds = 25.12 MB/sec
# dd if=/dev/zero of=tempfile bs=5M count=1024 conv=fdatasync,notrunc
1024+0 records in
1024+0 records out
5368709120 bytes (5.4 GB) copied, 271.993 s, 19.7 MB/s
# echo 3 > /proc/sys/vm/drop_caches
# dd if=tempfile of=/dev/null bs=5M count=1024
1024+0 records in
1024+0 records out
5368709120 bytes (5.4 GB) copied, 182.106 s, 29.5 MB/s
# sysbench --num-threads=16 --test=fileio --file-total-size=5G --file-test-mode=rndrw run
sysbench 0.4.12: multi-threaded system evaluation benchmark
Running the test with following options:
Number of threads: 16
Extra file open flags: 0
128 files, 40Mb each
5Gb total file size
Block size 16Kb
Number of random requests for random IO: 10000
Read/Write ratio for combined random IO test: 1.50
Periodic FSYNC enabled, calling fsync() each 100 requests.
Calling fsync() at the end of test, Enabled.
Using synchronous I/O mode
Doing random r/w test
Threads started!
Done.
Operations performed: 6000 Read, 4000 Write, 12800 Other = 22800 Total
Read 93.75Mb Written 62.5Mb Total transferred 156.25Mb (16.794Mb/sec)
1074.78 Requests/sec executed
Test execution summary:
total time: 9.3042s
total number of events: 10000
total time taken by event execution: 0.2975
per-request statistics:
min: 0.01ms
avg: 0.03ms
max: 30.70ms
approx. 95 percentile: 0.02ms
Threads fairness:
events (avg/stddev): 625.0000/553.34
execution time (avg/stddev): 0.0186/0.02
EBS IO-1 1000 IOPS, 100 GB
# hdparm -tT /dev/xvdf
/dev/xvdf:
Timing cached reads: 5090 MB in 2.00 seconds = 2550.81 MB/sec
Timing buffered disk reads: 104 MB in 3.03 seconds = 34.30 MB/sec
Timing cached reads: 5000 MB in 2.00 seconds = 2505.62 MB/sec
Timing buffered disk reads: 98 MB in 3.10 seconds = 31.64 MB/sec
Timing cached reads: 5046 MB in 2.01 seconds = 2507.34 MB/sec
Timing buffered disk reads: 98 MB in 3.04 seconds = 32.19 MB/sec
# dd if=/dev/zero of=tempfile bs=5M count=1024 conv=fdatasync,notrunc
1024+0 records in
1024+0 records out
5368709120 bytes (5.4 GB) copied, 167.252 s, 32.1 MB/s
# echo 3 > /proc/sys/vm/drop_caches
# dd if=tempfile of=/dev/null bs=5M count=1024
1024+0 records in
1024+0 records out
5368709120 bytes (5.4 GB) copied, 126.793 s, 42.3 MB/s
# sysbench --num-threads=16 --test=fileio --file-total-size=5G --file-test-mode=rndrw run
sysbench 0.4.12: multi-threaded system evaluation benchmark
Running the test with following options:
Number of threads: 16
Extra file open flags: 0
128 files, 40Mb each
5Gb total file size
Block size 16Kb
Number of random requests for random IO: 10000
Read/Write ratio for combined random IO test: 1.50
Periodic FSYNC enabled, calling fsync() each 100 requests.
Calling fsync() at the end of test, Enabled.
Using synchronous I/O mode
Doing random r/w test
Threads started!
Done.
Operations performed: 6001 Read, 4000 Write, 12800 Other = 22801 Total
Read 93.766Mb Written 62.5Mb Total transferred 156.27Mb (37.871Mb/sec)
2423.73 Requests/sec executed
Test execution summary:
total time: 4.1263s
total number of events: 10001
total time taken by event execution: 0.8466
per-request statistics:
min: 0.01ms
avg: 0.08ms
max: 22.31ms
approx. 95 percentile: 0.02ms
Threads fairness:
events (avg/stddev): 625.0625/318.25
execution time (avg/stddev): 0.0529/0.02
EBS IO-1 2000 IOPS, 200 GB
# hdparm -tT /dev/xvdi
/dev/xvdi:
Timing cached reads: 4846 MB in 2.00 seconds = 2428.51 MB/sec
Timing buffered disk reads: 90 MB in 3.02 seconds = 29.80 MB/sec
Timing cached reads: 5122 MB in 2.05 seconds = 2503.64 MB/sec
Timing buffered disk reads: 100 MB in 3.07 seconds = 32.56 MB/sec
Timing cached reads: 4330 MB in 2.04 seconds = 2123.52 MB/sec
Timing buffered disk reads: 102 MB in 3.05 seconds = 33.41 MB/sec
# dd if=/dev/zero of=tempfile bs=5M count=1024 conv=fdatasync,notrunc
1024+0 records in
1024+0 records out
5368709120 bytes (5.4 GB) copied, 161.549 s, 33.2 MB/s
# echo 3 > /proc/sys/vm/drop_caches
# dd if=tempfile of=/dev/null bs=5M count=1024
1024+0 records in
1024+0 records out
5368709120 bytes (5.4 GB) copied, 155.51 s, 34.5 MB/s
# sysbench --num-threads=16 --test=fileio --file-total-size=5G --file-test-mode=rndrw prepare
sysbench 0.4.12: multi-threaded system evaluation benchmark
128 files, 40960Kb each, 5120Mb total
Creating files for the test...
[root@ip-10-98-91-92 4]# sysbench --num-threads=16 --test=fileio --file-total-size=5G --file-test-mode=rndrw run
sysbench 0.4.12: multi-threaded system evaluation benchmark
Running the test with following options:
Number of threads: 16
Extra file open flags: 0
128 files, 40Mb each
5Gb total file size
Block size 16Kb
Number of random requests for random IO: 10000
Read/Write ratio for combined random IO test: 1.50
Periodic FSYNC enabled, calling fsync() each 100 requests.
Calling fsync() at the end of test, Enabled.
Using synchronous I/O mode
Doing random r/w test
Threads started!
Done.
Operations performed: 6000 Read, 4000 Write, 12801 Other = 22801 Total
Read 93.75Mb Written 62.5Mb Total transferred 156.25Mb (74.645Mb/sec)
4777.28 Requests/sec executed
Test execution summary:
total time: 2.0932s
total number of events: 10000
total time taken by event execution: 1.0015
per-request statistics:
min: 0.01ms
avg: 0.10ms
max: 10.10ms
approx. 95 percentile: 0.02ms
Threads fairness:
events (avg/stddev): 625.0000/177.29
execution time (avg/stddev): 0.0626/0.02
8xEBS Standard, 30 GB, RAID 10
# hdparm -tT /dev/md0
/dev/md0:
Timing cached reads: 3964 MB in 1.93 seconds = 2048.84 MB/sec
Timing buffered disk reads: 230 MB in 3.53 seconds = 65.13 MB/sec
Timing cached reads: 3994 MB in 1.99 seconds = 2002.16 MB/sec
Timing buffered disk reads: 398 MB in 3.00 seconds = 132.64 MB/sec
Timing cached reads: 4334 MB in 2.03 seconds = 2138.00 MB/sec
Timing buffered disk reads: 302 MB in 3.02 seconds = 99.84 MB/sec
# dd if=/dev/zero of=tempfile bs=5M count=1024 conv=fdatasync,notrunc
1024+0 records in
1024+0 records out
5368709120 bytes (5.4 GB) copied, 113.234 s, 47.4 MB/s
# echo 3 > /proc/sys/vm/drop_caches
# dd if=tempfile of=/dev/null bs=5M count=1024
1024+0 records in
1024+0 records out
5368709120 bytes (5.4 GB) copied, 97.9346 s, 54.8 MB/s
# sysbench --num-threads=16 --test=fileio --file-total-size=5G --file-test-mode=rndrw run
sysbench 0.5: multi-threaded system evaluation benchmark
Running the test with following options:
Number of threads: 16
Random number generator seed is 0 and will be ignored
Extra file open flags: 0
128 files, 40Mb each
5Gb total file size
Block size 16Kb
Number of IO requests: 10000
Read/Write ratio for combined random IO test: 1.50
Periodic FSYNC enabled, calling fsync() each 100 requests.
Calling fsync() at the end of test, Enabled.
Using synchronous I/O mode
Doing random r/w test
Threads started!
Operations performed: 5998 reads, 4002 writes, 12800 Other = 22800 Total
Read 93.719Mb Written 62.531Mb Total transferred 156.25Mb (87.287Mb/sec)
5586.40 Requests/sec executed
General statistics:
total time: 1.7901s
total number of events: 10000
total time taken by event execution: 1.1625s
response time:
min: 0.01ms
avg: 0.12ms
max: 8.99ms
approx. 95 percentile: 0.03ms
Threads fairness:
events (avg/stddev): 625.0000/171.83
execution time (avg/stddev): 0.0727/0.01
Ephemeral, 450 GB
# hdparm -tT /dev/xvdb
/dev/xvdb:
Timing cached reads: 4048 MB in 2.00 seconds = 2027.97 MB/sec
Timing buffered disk reads: 1794 MB in 3.12 seconds = 575.84 MB/sec
Timing cached reads: 4854 MB in 2.00 seconds = 2432.18 MB/sec
Timing buffered disk reads: 1830 MB in 3.00 seconds = 609.94 MB/sec
Timing cached reads: 3434 MB in 2.00 seconds = 1719.73 MB/sec
Timing buffered disk reads: 770 MB in 3.13 seconds = 245.97 MB/sec
# dd if=/dev/zero of=tempfile bs=5M count=1024 conv=fdatasync,notrunc
1024+0 records in
1024+0 records out
5368709120 bytes (5.4 GB) copied, 95.9093 s, 56.0 MB/s
# echo 3 > /proc/sys/vm/drop_caches
# dd if=tempfile of=/dev/null bs=5M count=1024
1024+0 records in
1024+0 records out
5368709120 bytes (5.4 GB) copied, 55.5027 s, 96.7 MB/s
# sysbench --num-threads=16 --test=fileio --file-total-size=5G --file-test-mode=rndrw run
sysbench 0.4.12: multi-threaded system evaluation benchmark
Running the test with following options:
Number of threads: 16
Extra file open flags: 0
128 files, 40Mb each
5Gb total file size
Block size 16Kb
Number of random requests for random IO: 10000
Read/Write ratio for combined random IO test: 1.50
Periodic FSYNC enabled, calling fsync() each 100 requests.
Calling fsync() at the end of test, Enabled.
Using synchronous I/O mode
Doing random r/w test
Threads started!
Done.
Operations performed: 6000 Read, 4000 Write, 12800 Other = 22800 Total
Read 93.75Mb Written 62.5Mb Total transferred 156.25Mb (11.263Mb/sec)
720.82 Requests/sec executed
Test execution summary:
total time: 13.8731s
total number of events: 10000
total time taken by event execution: 0.1603
per-request statistics:
min: 0.01ms
avg: 0.02ms
max: 2.34ms
approx. 95 percentile: 0.02ms
Threads fairness:
events (avg/stddev): 625.0000/117.83
execution time (avg/stddev): 0.0100/0.00
Обработка результатов
1. Сравнения скорости чтения/записи через dd, Mb/s:
2. Количество запросов в секунду через sysbench, Requests/s:
Итоги
Итак, похоже быстрее всех работает именно Ephemeral, как и ожидалось, но IOPS к нему самый низкий.
Скорость соединения со стандартными EBS выше, чем к оптимизированным, даже с 2к IOPS. Но количество IOPS больше у оптимизированных EBS, хотя RAID по IOPS всех победил.
В итоге, как и следовало ожидать, по количеству операций в секунду и простоте настроек оптимизированные EBS лучше, хотя использование рейдов из них будет ещё быстродейственнее.
Если вам важна скорость обращения к файлу — используйте ephemeral или стандартные EBS.