Tenho um HDD externo (não ssd) que informa:
Sector Sizes: 512 bytes logical, 4096 bytes physical
Devo usar --sector-size 512ou --sector-size 4096com cryptosetup LuksFormat? Usando parâmetros padrão (não tenho certeza se ele tenta detectar automaticamente ou usar 512 sempre) acabou com 512.
Se for importante, a configuração será[disk]->[gpt]->[physical partition]->[LUKSv2]->[btrfs]
Fui em frente e usei ambas as maneiras de coletar benchmarks:
512
# sysbench fileio prepare
2147483648 bytes written in 34.28 seconds (59.75 MiB/sec).
# sysbench fileio --file-test-mode=rndrw run
Extra file open flags: (none)
128 files, 16MiB each
2GiB total file size
Block size 16KiB
Number of IO requests: 0
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
Initializing worker threads...
Threads started!
File operations:
reads/s: 14.42
writes/s: 9.62
fsyncs/s: 31.25
Throughput:
read, MiB/s: 0.23
written, MiB/s: 0.15
General statistics:
total time: 12.4778s
total number of events: 562
Latency (ms):
min: 0.00
avg: 17.84
max: 365.73
95th percentile: 84.47
sum: 10024.98
Threads fairness:
events (avg/stddev): 562.0000/0.00
execution time (avg/stddev): 10.0250/0.00
# sysbench fileio --file-test-mode=seqrewr run
Extra file open flags: (none)
128 files, 16MiB each
2GiB total file size
Block size 16KiB
Periodic FSYNC enabled, calling fsync() each 100 requests.
Calling fsync() at the end of test, Enabled.
Using synchronous I/O mode
Doing sequential rewrite test
Initializing worker threads...
Threads started!
File operations:
reads/s: 0.00
writes/s: 815.93
fsyncs/s: 1045.19
Throughput:
read, MiB/s: 0.00
written, MiB/s: 12.75
General statistics:
total time: 10.0488s
total number of events: 18576
Latency (ms):
min: 0.00
avg: 0.54
max: 433.66
95th percentile: 0.02
sum: 9996.57
Threads fairness:
events (avg/stddev): 18576.0000/0.00
execution time (avg/stddev): 9.9966/0.00
##4096
# sysbench fileio prepare
2147483648 bytes written in 28.85 seconds (70.99 MiB/sec).
# sysbench --test=fileio --file-test-mode=rndrw run
File operations:
reads/s: 26.17
writes/s: 17.45
fsyncs/s: 58.35
Throughput:
read, MiB/s: 0.41
written, MiB/s: 0.27
General statistics:
total time: 11.4636s
total number of events: 1041
Latency (ms):
min: 0.00
avg: 9.63
max: 370.49
95th percentile: 44.98
sum: 10021.97
Threads fairness:
events (avg/stddev): 1041.0000/0.00
execution time (avg/stddev): 10.0220/0.00
# sysbench fileio --file-test-mode=seqrewr run
File operations:
reads/s: 0.00
writes/s: 1229.02
fsyncs/s: 1574.44
Throughput:
read, MiB/s: 0.00
written, MiB/s: 19.20
General statistics:
total time: 10.0071s
total number of events: 27929
Latency (ms):
min: 0.00
avg: 0.36
max: 471.49
95th percentile: 0.02
sum: 9997.75
Threads fairness:
events (avg/stddev): 27929.0000/0.00
execution time (avg/stddev): 9.9978/0.00
Para arquivos artificiais pequenos, 4K parece ser duas vezes mais rápido para leituras e 0,25 mais rápido para gravações.
Outra edição. Fiz o mesmo teste, mas copiando 110G de um pacote de arquivos variando de 2,5Mb (95%tile) a 300Mb.
Tanto a unidade formatada em 512 quanto a formatada em 4096 levou 20m30s +-4s
