Re: Can't get Dell PE T420 (Perc H710) perform better than a MacMini with PostgreSQL - Mailing list pgsql-performance
| From | Pietro Pugni |
|---|---|
| Subject | Re: Can't get Dell PE T420 (Perc H710) perform better than a MacMini with PostgreSQL |
| Date | |
| Msg-id | 76EF504C-06AF-471F-8BCD-AFD2B6F85178@gmail.com Whole thread Raw |
| In response to | Re: Can't get Dell PE T420 (Perc H710) perform better than a MacMini with PostgreSQL (Aidan Van Dyk <aidan@highrise.ca>) |
| Responses |
Re: Can't get Dell PE T420 (Perc H710) perform better than
a MacMini with PostgreSQL
(Josh Krupka <jkrupka@gmail.com>)
|
| List | pgsql-performance |
Hi Aidan,
thank you again for your support.
I found an interesting article showing better performance from a Intel i5 vs a Intel Xeon on different Postgres versions: http://blog.pgaddict.com/posts/performance-since-postgresql-7-4-to-9-4-pgbench
I have to say that MacMini has a 2011 CPU ( http://ark.intel.com/it/products/53463/Intel-Core-i7-2635QM-Processor-6M-Cache-up-to-2_90-GHz ) while Dell has two 2013 CPU ( http://ark.intel.com/products/75267/Intel-Xeon-Processor-E5-2640-v2-20M-Cache-2_00-GHz ), both at 2.0Ghz.
NUMA stands for "Non-Uniform-Memory-Access" . It's basically the "label" for systems which have memory attached to different cpu sockets, such that accessing all of the memory from a paritciular cpu thread has different costs based on where the actual memory is located (i.e. on some other socket, or the local socket).
Thanks, good to know.
QPI is the the intel "QuickPath Interconnect". It's the communication path between CPU sockets. Memory ready by one cpu thread that has to come from another cpu socket's memory controller goes through QPI.
Google has lots of info on these, and how they impact performance, etc.
When I’ll get access to BIOS (probably next week or later) I’ll try to disable QPI (if possible). Meanwhile I’ll document on Internet about QPI vs performance.
I’ve done some tests with sysbench on Dell T420 (via apt-get install) and MacMini (I’ve compiled the latest available sources at https://github.com/akopytov/sysbench ).If you want to see how bad the NUMA/QPI is, play with stream to benchmark memory performance.With stream you refer to this: https://sites.utexas.edu/jdm4372/tag/stream-benchmark/ ? Do you suggest me some way to do this kind of tests?Ya, that's the one. I don't have specific tests in mind.
Here are some results with 16GB RAM read and written at 1MB block size (I don’t know if this makes sense, but I’ve no problem in changing these parameters).
T420 - RAM READ - 16GB / 1MB
sh-4.3# sysbench --test=memory --memory-oper=read --memory-block-size=1MB --memory-total-size=16GB run
sysbench 0.4.12: multi-threaded system evaluation benchmark
Running the test with following options:
Number of threads: 1
Doing memory operations speed test
Memory block size: 1024K
Memory transfer size: 16384M
Memory operations type: read
Memory scope type: global
Threads started!
Done.
Operations performed: 16384 (3643025.32 ops/sec)
16384.00 MB transferred (3643025.32 MB/sec)
Test execution summary:
total time: 0.0045s
total number of events: 16384
total time taken by event execution: 0.0031
per-request statistics:
min: 0.00ms
avg: 0.00ms
max: 0.02ms
approx. 95 percentile: 0.00ms
Threads fairness:
events (avg/stddev): 16384.0000/0.00
execution time (avg/stddev): 0.0031/0.00
MacMini - RAM READ - 16GB / 1MB
server:sysbench Pietro$ ./sysbench --test=memory --memory-oper=read --memory-block-size=1MB --memory-total-size=16GB run
sysbench 0.5: multi-threaded system evaluation benchmark
Running the test with following options:
Number of threads: 1
Random number generator seed is 0 and will be ignored
Threads started!
Operations performed: 16384 ( 5484.50 ops/sec)
16384.00 MB transferred (5484.50 MB/sec)
General statistics:
total time: 2.9873s
total number of events: 16384
total time taken by event execution: 2.9836s
response time:
min: 0.18ms
avg: 0.18ms
max: 0.24ms
approx. 95 percentile: 0.19ms
Threads fairness:
events (avg/stddev): 16384.0000/0.00
execution time (avg/stddev): 2.9836/0.00
T420 - RAM WRITE - 16GB / 1MB
sh-4.3# sysbench --test=memory --memory-oper=write --memory-block-size=1MB --memory-total-size=16GB run
sysbench 0.4.12: multi-threaded system evaluation benchmark
Running the test with following options:
Number of threads: 1
Doing memory operations speed test
Memory block size: 1024K
Memory transfer size: 16384M
Memory operations type: write
Memory scope type: global
Threads started!
Done.
Operations performed: 16384 ( 8298.97 ops/sec)
16384.00 MB transferred (8298.97 MB/sec)
Test execution summary:
total time: 1.9742s
total number of events: 16384
total time taken by event execution: 1.9723
per-request statistics:
min: 0.12ms
avg: 0.12ms
max: 0.25ms
approx. 95 percentile: 0.12ms
Threads fairness:
events (avg/stddev): 16384.0000/0.00
execution time (avg/stddev): 1.9723/0.00
MacMini - RAM WRITE - 16GB / 1MB
server:sysbench Pietro$ ./sysbench --test=memory --memory-oper=write --memory-block-size=1MB --memory-total-size=16GB run
sysbench 0.5: multi-threaded system evaluation benchmark
Running the test with following options:
Number of threads: 1
Random number generator seed is 0 and will be ignored
Threads started!
Operations performed: 16384 ( 5472.90 ops/sec)
16384.00 MB transferred (5472.90 MB/sec)
General statistics:
total time: 2.9937s
total number of events: 16384
total time taken by event execution: 2.9890s
response time:
min: 0.18ms
avg: 0.18ms
max: 0.32ms
approx. 95 percentile: 0.19ms
Threads fairness:
events (avg/stddev): 16384.0000/0.00
execution time (avg/stddev): 2.9890/0.00
T420 - CPU
sh-4.3# sysbench --test=cpu run
sysbench 0.4.12: multi-threaded system evaluation benchmark
Running the test with following options:
Number of threads: 1
Doing CPU performance benchmark
Threads started!
Done.
Maximum prime number checked in CPU test: 10000
Test execution summary:
total time: 13.0683s
total number of events: 10000
total time taken by event execution: 13.0674
per-request statistics:
min: 1.30ms
avg: 1.31ms
max: 1.44ms
approx. 95 percentile: 1.35ms
Threads fairness:
events (avg/stddev): 10000.0000/0.00
execution time (avg/stddev): 13.0674/0.00
MacMini - CPU
server:sysbench Pietro$ ./sysbench --test=cpu run
sysbench 0.5: multi-threaded system evaluation benchmark
Running the test with following options:
Number of threads: 1
Random number generator seed is 0 and will be ignored
Primer numbers limit: 10000
Threads started!
General statistics:
total time: 11.5728s
total number of events: 10000
total time taken by event execution: 11.5703s
response time:
min: 1.15ms
avg: 1.16ms
max: 2.17ms
approx. 95 percentile: 1.17ms
Threads fairness:
events (avg/stddev): 10000.0000/0.00
execution time (avg/stddev): 11.5703/0.00
It returns the following output:A more simple "overview" might be "numactl —hardware”
sh-4.3# numactl --hardware
available: 2 nodes (0-1)
node 0 cpus: 0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30
node 0 size: 64385 MB
node 0 free: 56487 MB
node 1 cpus: 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31
node 1 size: 64508 MB
node 1 free: 62201 MB
node distances:
node 0 1
0: 10 20
1: 20 10
Thank you so much for your help, really appreciate it.
Best regards,
Pietro
pgsql-performance by date: