stevepedwards.com/DebianAdmin linux mint IT admin tips info

How Slow Can You Go? Sysbench Comparisons – Memory

stevee@AMDA8 ~ $ sudo sysbench --test=mutex --memory-oper=read run
[sudo] password for stevee:
sysbench 0.4.12: multi-threaded system evaluation benchmark

Running the test with following options:
Number of threads: 1

Doing mutex performance test
Threads started!
Done.
Test execution summary:
total time: 0.0071s
total number of events: 1
total time taken by event execution: 0.0066
per-request statistics:
min: 6.57ms
avg: 6.57ms
max: 6.57ms
approx. 95 percentile: 10000000.00ms

Threads fairness:
events (avg/stddev): 1.0000/0.00
execution time (avg/stddev): 0.0066/0.00

stevee@AMDA8 ~ $ sudo sysbench --test=mutex --memory-oper=write run
sysbench 0.4.12: multi-threaded system evaluation benchmark

Running the test with following options:
Number of threads: 1

Doing mutex performance test
Threads started!
Done.
Test execution summary:
total time: 0.0086s
total number of events: 1
total time taken by event execution: 0.0080
per-request statistics:
min: 8.01ms
avg: 8.01ms
max: 8.01ms
approx. 95 percentile: 10000000.00ms

Threads fairness:
events (avg/stddev): 1.0000/0.00
execution time (avg/stddev): 0.0080/0.00

stevee@hpmint ~ $ sudo sysbench --test=mutex --memory-oper=read run
sysbench 0.4.12: multi-threaded system evaluation benchmark

Running the test with following options:
Number of threads: 1

Doing mutex performance test
Threads started!
Done.
Test execution summary:
total time: 0.0117s
total number of events: 1
total time taken by event execution: 0.0112
per-request statistics:
min: 11.21ms
avg: 11.21ms
max: 11.21ms
approx. 95 percentile: 10000000.00ms

Threads fairness:
events (avg/stddev): 1.0000/0.00
execution time (avg/stddev): 0.0112/0.00

stevee@hpmint ~ $ sudo sysbench --test=mutex --memory-oper=write run
sysbench 0.4.12: multi-threaded system evaluation benchmark

Running the test with following options:
Number of threads: 1

Doing mutex performance test
Threads started!
Done.
Test execution summary:
total time: 0.0116s
total number of events: 1
total time taken by event execution: 0.0108
per-request statistics:
min: 10.81ms
avg: 10.81ms
max: 10.81ms
approx. 95 percentile: 10000000.00ms

Threads fairness:
events (avg/stddev): 1.0000/0.00
execution time (avg/stddev): 0.0108/0.00

stevee@DELLMINT ~ $ sudo sysbench --test=mutex --memory-oper=read run
[sudo] password for stevee:
sysbench 0.4.12: multi-threaded system evaluation benchmark

Running the test with following options:
Number of threads: 1

Doing mutex performance test
Threads started!
Done.
Test execution summary:
total time: 0.0111s
total number of events: 1
total time taken by event execution: 0.0103
per-request statistics:
min: 10.34ms
avg: 10.34ms
max: 10.34ms
approx. 95 percentile: 10000000.00ms

Threads fairness:
events (avg/stddev): 1.0000/0.00
execution time (avg/stddev): 0.0103/0.00

stevee@DELLMINT ~ $ sudo sysbench --test=mutex --memory-oper=write run
sysbench 0.4.12: multi-threaded system evaluation benchmark

Running the test with following options:
Number of threads: 1

Doing mutex performance test
Threads started!
Done.
Test execution summary:
total time: 0.0120s
total number of events: 1
total time taken by event execution: 0.0111
per-request statistics:
min: 11.06ms
avg: 11.06ms
max: 11.06ms
approx. 95 percentile: 10000000.00ms

Threads fairness:
events (avg/stddev): 1.0000/0.00
execution time (avg/stddev): 0.0111/0.00

stevee@minimint ~ $ sudo sysbench --test=mutex --memory-oper=read run
[sudo] password for stevee:
sysbench 0.4.12: multi-threaded system evaluation benchmark

Running the test with following options:
Number of threads: 1

Doing mutex performance test
Threads started!
Done.
Test execution summary:
total time: 0.0132s
total number of events: 1
total time taken by event execution: 0.0126
per-request statistics:
min: 12.62ms
avg: 12.62ms
max: 12.62ms
approx. 95 percentile: 10000000.00ms

Threads fairness:
events (avg/stddev): 1.0000/0.00
execution time (avg/stddev): 0.0126/0.00

stevee@minimint ~ $ sudo sysbench --test=mutex --memory-oper=write run
sysbench 0.4.12: multi-threaded system evaluation benchmark

Running the test with following options:
Number of threads: 1

Doing mutex performance test
Threads started!
Done.
Test execution summary:
total time: 0.0284s
total number of events: 1
total time taken by event execution: 0.0279
per-request statistics:
min: 27.93ms
avg: 27.93ms
max: 27.93ms
approx. 95 percentile: 10000000.00ms

Threads fairness:
events (avg/stddev): 1.0000/0.00
execution time (avg/stddev): 0.0279/0.00

AMDA8 quad coreDell 64 bit dual coreHP Pav 64 bit dual HP Pav 32 bit mono
 Memory DRAM 8GB 6GB 4GB 1GB
reads ratio times slower

write ratio times slower

0.0066/0.0066 = 1

0.008/0.008 = 1 

0.0103/0.0066 = 1.56

0.0111/0.008 = 1.39

0.0112/0.0066 = 1.7

0.0126/0.008 = 1.6

0.0126/0.0066 = 1.9

0.028/0.008 = 3.5 

Added 4/10/16HP  Intel Pent Dual
 Memory DRAM 2GB
reads ratio times slower

write ratio times slower

0.0056/0.0064 = 1.143

 

0.0084/0.0088 = 0.95

Again, the added HP Intel Dual is slightly faster than the quad core for write time!

Some read ratios compared to the quad core are slower than writes and vice-versa here.

The 32 bit Celeron is up to 3.5 times slower in these memory tests.

You can start to imagination the time accumulation for each area combining between compared systems - CPU, memory and Disk IO - making the overall impact of being many times slower for a worst case, from a user experience perspective, as for every clock tick of a CPU that goes by and effective work can not be done due to these combined latencies - let alone considering multiple processes across a whole OS - you can imagine where the slowest bottlenecks for a given process affect the final results system wide.

There is an interesting table on p20 of System Performance that shows lost time from a 3GHz processor's perspective to human terms:cpulosttime.jpg

Level 1 cache - 3 times slower than CPU

Level 2 cache - 9 times slower than CPU

Level 3 cache - 43 times slower than CPU

This puts main memory in perspective to CPU cycles - 6 mins in human 1sec terms to CPU clock ticks - or 6 x 60 =

Main memory - 360 times slower than CPU

The sooner they build larger memory directly onto the CPU chip the better eh?

Comments are closed.

Post Navigation