[pgpool-general: 6037] Re: Performance

[Mathias Buresch]

Hi,

I have done the same with pgbench now:

### Master
pgbench -c 10 -T 300 -p 5432 -h master test
number of transactions actually processed: 89094
latency average = 33.677 ms
tps = 296.935737 (including connections establishing)
tps = 296.938804 (excluding connections establishing)

### PgPool
pgbench -c 10 -T 300 -p 5433 -h pgpool test
number of transactions actually processed: 67143
latency average = 44.689 ms
tps = 223.769562 (including connections establishing)
tps = 223.776728 (excluding connections establishing)


### Master (1 Client)
pgbench -c 1 -T 300 -p 5432 -h master test
number of transactions actually processed: 52614
latency average = 5.702 ms
tps = 175.379565 (including connections establishing)
tps = 175.381607 (excluding connections establishing)


### PgPool (1 Client)
pgbench -c 1 -T 300 -p 5433 -h pgpool test
number of transactions actually processed: 38840
latency average = 7.724 ms
tps = 129.465732 (including connections establishing)
tps = 129.471359 (excluding connections establishing)


I will try to log what exactly happens in that job I posted before. Is there something I shall look for in detail? Which information you need exactly?


On 07.04.18, 07:19, "Tatsuo Ishii" <ishii at sraoss.co.jp> wrote:

    > I have an import of „core-data“ of a shop system where an application runs some selects and inserts and so. Those are always the same queries.
    > I have run the import several times and I have those timings:
    > 
    > pgpool ---------- 1570.22s
    > master ----------- 166.18s
    > pgpool ---------- 2295.56s
    > pgpool ---------- 2014.09s
    > master ----------- 129.94s
    > 
    > So if I run the queries agains Pgpool it is more than 10x as I would run against Master.
    
    I have run pgbench against pgpool and master. Here are the results:
    
    pgpool:
    
    $ pgpool --version
    pgpool-II version 3.7.2 (amefuriboshi)
    $ pgbench -c 10 -T 300 -p 11000 test
    starting vacuum...end.
    transaction type: <builtin: TPC-B (sort of)>
    scaling factor: 1
    query mode: simple
    number of clients: 10
    number of threads: 1
    duration: 300 s
    number of transactions actually processed: 59781
    latency average = 50.194 ms
    tps = 199.228157 (including connections establishing)
    tps = 199.231150 (excluding connections establishing)
    
    master:
    $ pgbench -c 10 -T 300 -p 11002 test
    starting vacuum...end.
    transaction type: <builtin: TPC-B (sort of)>
    scaling factor: 1
    query mode: simple
    number of clients: 10
    number of threads: 1
    duration: 300 s
    number of transactions actually processed: 67768
    latency average = 44.280 ms
    tps = 225.836885 (including connections establishing)
    tps = 225.838017 (excluding connections establishing)
    
    "tps" = transactions per second. So the nuber larger, the better.
    1 - 199.231150/225.838017 = 0.1179
    
    So pgpool is 11.79% slower than master here.
    
    Pgpool-II's performance is depending how the test was performed like
    other software. If you could provid more details, maybe I could find a
    cause of the difference (a self contaied test case is ideal).
    
    Best regards,
    --
    Tatsuo Ishii
    SRA OSS, Inc. Japan
    English: http://www.sraoss.co.jp/index_en.php
    Japanese:http://www.sraoss.co.jp