Repository navigation
Performance
- Raspberry Pi 4b
- ARMv7 Processor rev 3 (v7l) - Broadcom BCM2711B0 (Cortex A-72) - 1.5 GHz Quad core
- 4G ram
- disks
- mmcblk0 disk 119.3G (SSD Samsung EVO Select 128)
- HP XW6600 workstation
- Intel(R) Xeon(R) CPU E5430 @ 2.66GHz Quad core
- 32G ram
- disks
- sda disk 149.1G WDC WD1600AAJS-6
- sdb disk 931.5G WDC WD1001FALS-0
- sdc disk 1.8T ST2000DM008-2FR1
Constants:
- N_THREADS
- N_MESSAGES
- MAX_SLEEP
Logging to a file is configured. N_THREADS threads are created. Each thread will log N_MESSAGES messages. Between writing messages, the threads sleep a random amount of time from 0 to MAX_SLEEP microseconds.
After the output file was generated, it was verified that each message conformed to the output format, and that the correct number of messages were written.
2020-06-13 02:13:45.524 INFO 7213:thread_97 sleeping 750 microseconds, s/n=3994, tid=7213
2020-06-13 02:13:45.524 INFO 7233:thread_117 sleeping 33 microseconds, s/n=3899, tid=7233
2020-06-13 02:13:45.524 INFO 7203:thread_87 sleeping 870 microseconds, s/n=3948, tid=7203
2020-06-13 02:13:45.524 INFO 7233:thread_117 sleeping 301 microseconds, s/n=3900, tid=7233
2020-06-13 02:13:45.524 INFO 7169:thread_53 sleeping 838 microseconds, s/n=3930, tid=7169
2020-06-13 02:13:45.524 INFO 7321:thread_205 sleeping 345 microseconds, s/n=3914, tid=7321
2020-06-13 02:13:45.524 INFO 7256:thread_140 sleeping 222 microseconds, s/n=3837, tid=7256
2020-06-13 02:13:45.524 INFO 7236:thread_120 sleeping 561 microseconds, s/n=3948, tid=7236
2020-06-13 02:13:45.525 INFO 7266:thread_150 sleeping 12 microseconds, s/n=3865, tid=7266
2020-06-13 02:13:45.525 INFO 7233:thread_117 sleeping 241 microseconds, s/n=3901, tid=7233
2020-06-13 02:13:45.525 INFO 7120:thread_4 sleeping 378 microseconds, s/n=3968, tid=7120
2020-06-13 02:13:45.525 INFO 7256:thread_140 sleeping 289 microseconds, s/n=3838, tid=7256
2020-06-13 02:13:45.525 INFO 7266:thread_150 sleeping 6 microseconds, s/n=3866, tid=7266
2020-06-13 02:13:45.525 INFO 7352:thread_236 sleeping 37 microseconds, s/n=3982, tid=7352
2020-06-13 02:13:45.525 INFO 7321:thread_205 sleeping 527 microseconds, s/n=3915, tid=7321
2020-06-13 02:13:45.525 INFO 7324:thread_208 sleeping 317 microseconds, s/n=3879, tid=7324
2020-06-13 02:13:45.525 INFO 7266:thread_150 sleeping 47 microseconds, s/n=3867, tid=7266
2020-06-13 02:13:45.525 INFO 7352:thread_236 sleeping 181 microseconds, s/n=3983, tid=7352
2020-06-13 02:13:45.525 INFO 7213:thread_97 sleeping 241 microseconds, s/n=3995, tid=7213
2020-06-13 02:13:45.525 INFO 7233:thread_117 sleeping 152 microseconds, s/n=3902, tid=7233
2020-06-13 02:13:45.525 INFO 7266:thread_150 sleeping 153 microseconds, s/n=3868, tid=7266
2020-06-13 02:13:45.525 INFO 7256:thread_140 sleeping 46 microseconds, s/n=3839, tid=7256
2020-06-13 02:13:45.525 INFO 7120:thread_4 sleeping 629 microseconds, s/n=3969, tid=7120
Resulting file info:
$ wc /tmp/hive.log
1000500 9004750 97213984 /tmp/hive.log
| param | XW6600 | Raspberry Pi 4b |
|---|---|---|
| N_THREADS | 250 | 250 |
| N_MESSAGES | 4000 | 4000 |
| MAX_SLEEP | 1000 | 1000 |
| Total Messages | 1,000,000 | 1,000,000 |
| real time | 0m15.689s | 0m16.733s |
| user time | 0m9.551s | 0m10.681s |
| sys time | 0m14.218s | 0m17.121s |
| messages/second | 63.7 k | 59.9 k |
| MBytes/sec | 6.2 | 5.8 |
| param | XW6600 | Raspberry Pi 4b |
|---|---|---|
| N_THREADS | 250 | 250 |
| N_MESSAGES | 4000 | 4000 |
| MAX_SLEEP | 1000 | 1000 |
| Total Messages | 1,000,000 | 1,000,000 |
| real time | 0m23.077s | 0m28.207s |
| user time | 0m10.796s | 0m12.486s |
| sys time | 0m19.133s | 0m26.316s |
| messages/second | 43.3 k | 38.0 k |
| MBytes/sec | 4.2 | 3.4 |
-
The time to read and check the output was included in the times listed above.
-
No line buffering on the Raspberry Pi was the only case where the sys time exceeded the real time, meaning it averaged more than 1 core to support the test.
-
1000 threads was initialy used on the xw6600. When it was tried on the Raspberry Pi, it failed for the following reason:
pthread create: Resource temporarily unavailable
I reduced the thread count to 500, and it worked the first time, but not the second. 250 threads works reliably.
- MBytes/sec was an afterthought.