2009-09-19 17:07:36 UTC
hi,
I would like to ask what can cause the difference of the result of "read" and "write" operation in global and sequential mode of "memory" benchmark in SysBench.
I have tried several "memory-block-size" and I realize that "read" operation is quicker than "write". Is this result coming from the effect of data prefetching of cache? or coming from the restriction of write buffer? or what else can cause?
I show one example in read and write below. read is twice quicker than write there. I use intel core 2 duo and fedora 7.
read operation
$ sysbench --num-threads=100 --test=memory --memory-block-size=8M --memory-scope=global --memory-total-size=50G --memory-oper=read run
sysbench v0.4.8: multi-threaded system evaluation benchmark
Running the test with following options:
Number of threads: 100
Doing memory operations speed test
Memory block size: 8192K
Memory transfer size: 51200M
Memory operations type: read
Memory scope type: global
Threads started!
Done.
Operations performed: 6400 ( 255.66 ops/sec)
51200.00 MB transferred (2045.30 MB/sec)
Test execution summary:
total time: 25.0330 s
total number of events: 6400
total time taken by event execution: 64.4097
per-request statistics:
min: 0.0073s
avg: 0.0101s
max: 0.6164s
approx. 95 percentile: 0.0077s
Threads fairness:
events (avg/stddev): 64.0000/0.63
execution time (avg/stddev): 0.6441/0.31
write operation
$ sysbench --num-threads=100 --test=memory --memory-block-size=8M --memory-scope=global --memory-total-size=50G --memory-oper=write run
sysbench v0.4.8: multi-threaded system evaluation benchmark
Running the test with following options:
Number of threads: 100
Doing memory operations speed test
Memory block size: 8192K
Memory transfer size: 51200M
Memory operations type: write
Memory scope type: global
Threads started!
Done.
Operations performed: 6400 ( 120.93 ops/sec)
51200.00 MB transferred (967.47 MB/sec)
Test execution summary:
total time: 52.9216 s
total number of events: 6400
total time taken by event execution: 2886.7675
per-request statistics:
min: 0.0100s
avg: 0.4511s
max: 1.4056s
approx. 95 percentile: 0.8500s
Threads fairness:
events (avg/stddev): 64.0000/2.04
execution time (avg/stddev): 28.8677/0.99
