Hello,
I just installed Percona XtraDB Cluster on 3 nodes, it’s running on latest 5.7 centos 7 package. I try to do some performance testing and I didnt found much results to compare if my results are good or not.
I got
Number of threads: 32
OLTP test statistics:
queries performed:
read: 418194
write: 119459
other: 59706
total: 597359
transactions: 29835 (495.78 per sec.)
read/write requests: 537653 (8934.31 per sec.)
other operations: 59706 (992.15 per sec.)
ignored errors: 36 (0.60 per sec.)
reconnects: 0 (0.00 per sec.)
General statistics:
total time: 60.1784s
total number of events: 29835
total time taken by event execution: 1921.2193s
response time:
min: 13.63ms
avg: 64.39ms
max: 656.78ms
approx. 95 percentile: 211.55ms
Threads fairness:
events (avg/stddev): 932.3438/478.22
execution time (avg/stddev): 60.0381/0.05
Is it good or not? Is somewhere some results to compare? I know that there will be different results on different db version and HW, but just for some basic judgement if that result is good or not it should be enough.
Being a clustered product lot of things depends on how things are configured and what kind of workload you are running.
(and of-course as you pointed HW plays and important role too).
I assume you are running basic sysbench oltp workload with all default values.
Generally sysbench or any performance suite helps in understanding performance in relative perspective.
So next question is “does sysbench resembles your production workload ?”
You can try to execute your product workload and see if that you are hitting your planned targets.
This is my sysbench command:
for each in 1 4 8 16 32 64; do sysbench --test=/usr/share/doc/sysbench/tests/db/oltp.lua --mysql-host=10.5.3.25 --mysql-port=3306 --db-driver=mysql --oltp-table-size=10000000 --mysql-db=sbtest --mysql-user=sbtest --mysql-password=******* --max-time=60 --max-requests=0 --oltp-read-only --oltp-skip-trx --oltp-nontrx-mode=select --num-threads=$each run; sleep 10; done
I got best results for 32 threads, so I mentioned it. Yes, its standard oltp test.
I agree that performance could be very relative, BUT I think that if you got some numbers, you can at least say if my TPS is terribly low or its some average? I want to know if some similar sysbench results are somewhere to compare or if someone can run sysbench oltp test on own cluster and share results. Thats all.
I build that cluster for third person and I am not responsible for product workload, so I need to have some performance measures with some standard benchmark tool like sysbench. I need it for case that their code is not so well and they will have bad results on every DB.