Menu
▾
▴
[Postgres-xc-general] my first results with pgxc 1.1 - part 2
|
From: Hector M. J. <hec...@et...> - 2013-10-03 16:21:42
|
Hi all,
First of all, I would like to thank Mr. Koichi Suzuki for their comments
to my previous post. Was right with regard to the parameter
max_connection. In the new configuration each DataNode has a maximum of
800 (three coordinators with 200 concurrent connections each plus 200 extra)
Regarding the suggestion about stress tool to use (dbt-1), I'm in the
study of their characteristics and peculiarities of installation and use.
When I get my first results with it I publishes in this forum.
I still have details to resolve as is the use of parameter
gtmPxyExtraConfig of pgxc_ctl.conf to include parameters (as
worker_threads = xx, for example) in the gtm_proxy settings.
This was one of the problems detected during the initial deployment
through pgxc_ctl tool.
This is the second post about my impressions with pgxc-1.1
In the previous post exposed them details of the scenario we want to
build as well as configurations necessary to reach a state of
functionality and operability acceptable.
Once you reach that point, we design and execute a set of tests (based
on pgbench) in order to measure the performance of our installation and
know when we reached our goals: 300 (or more) concurrent connections and
increase the number of transactional operations.
The modifications was made mainly on DataNodes config. These changes
weres implemented through datanodeExtraConfig parameter (pgxc_ctl.conf
file) and were as follows:
# ================================================
# Added to all the DataNode postgresql.conf
# Original: datanodeExtraConfig
log_destination = 'syslog'
logging_collector = on
log_directory = 'pg_log'
listen_addresses = '*'
max_connections = 800
work_mem = 100MB
fsync = off
shared_buffers = 5GB
wal_buffers = 1MB
checkpoint_timeout = 5min
effective_cache_size = 12GB
checkpoint_segments = 64
checkpoint_completion_target = 0.9
maintenance_work_mem = 4GB
max_prepared_transactions = 800
synchronous_commit = off
These modifications allowed us to obtain results with increases between
two and three times (in some cases, more than three times) with respect
to the initial results set (number of transactions and tps).
Our other goal, get more than 300 concurrent connections is reached and
can measure up to 355 connections.
During the course of testing measurements were obtained on the
consumption of CPU and memory resources on each of the components of the
cluster (dn01, DN02, DN03, GTM).
For these measurements, use the SAR command parameters:
-R Report memory utilization statistics
-U Report CPU utilization
200 iterations with an interval of 5 seconds (14 test with a duration of
60 seconds divided by 5 seconds is +/- 170 iterations)
The averages obtained by server:
This command was executed on each servers before launch the tests:
sar -u -r 5 200
dn01:
Average: CPU %user %nice %system %iowait %steal
%idle
Average: all 15.79 0.00 18.20 0.44 0.00
65.57
Average: kbmemfree kbmemused %memused kbbuffers kbcached kbcommit
%commit
Average: 12003982 4327934 26.50 44163 1766752 7616567
37.35
dn02:
Average: CPU %user %nice %system %iowait %steal
%idle
Average: all 14.89 0.00 17.37 0.11 0.00
67.62
Average: kbmemfree kbmemused %memused kbbuffers kbcached kbcommit
%commit
Average: 12097661 4234255 25.93 42716 1725394 7609960
37.31
dn03:
Average: CPU %user %nice %system %iowait %steal
%idle
Average: all 16.67 0.00 19.59 0.57 0.00
63.17
Average: kbmemfree kbmemused %memused kbbuffers kbcached kbcommit
%commit
Average: 11603908 4728008 28.95 42955 1708146 7609769
37.31
gtm:
Average: CPU %user %nice %system %iowait %steal
%idle
Average: all 8.54 0.00 24.80 0.12 0.00
66.54
Average: kbmemfree kbmemused %memused kbbuffers kbcached kbcommit
%commit
Average: 3553938 370626 9.44 42358 120419 723856
9.06
The result obtained in each of the tests:
[root@rhelclient ~]# pgbench -c 16 -j 8 -T 60 -h 192.168.97.44 -U
postgres pgbench
starting vacuum...end.
transaction type: TPC-B (sort of)
scaling factor: 100
query mode: simple
number of clients: 16
number of threads: 8
duration: 60 s
number of transactions actually processed: 23680
tps = 394.458636 (including connections establishing)
tps = 394.637063 (excluding connections establishing)
[root@rhelclient ~]# pgbench -S -c 16 -j 8 -T 60 -h 192.168.97.44 -U
postgres pgbench
starting vacuum...end.
transaction type: SELECT only
scaling factor: 100
query mode: simple
number of clients: 16
number of threads: 8
duration: 60 s
number of transactions actually processed: 108929
tps = 1815.247714 (including connections establishing)
tps = 1815.947505 (excluding connections establishing)
[root@rhelclient ~]# pgbench -c 16 -j 16 -T 60 -h 192.168.97.44 -U
postgres pgbench
starting vacuum...end.
transaction type: TPC-B (sort of)
scaling factor: 100
query mode: simple
number of clients: 16
number of threads: 16
duration: 60 s
number of transactions actually processed: 23953
tps = 399.034541 (including connections establishing)
tps = 399.120451 (excluding connections establishing)
[root@rhelclient ~]# pgbench -S -c 16 -j 16 -T 60 -h 192.168.97.44 -U
postgres pgbench
starting vacuum...end.
transaction type: SELECT only
scaling factor: 100
query mode: simple
number of clients: 16
number of threads: 16
duration: 60 s
number of transactions actually processed: 127142
tps = 2118.825088 (including connections establishing)
tps = 2119.318006 (excluding connections establishing)
[root@rhelclient ~]# pgbench -c 96 -j 8 -T 60 -h 192.168.97.44 -U
postgres pgbench
starting vacuum...end.
transaction type: TPC-B (sort of)
scaling factor: 100
query mode: simple
number of clients: 96
number of threads: 8
duration: 60 s
number of transactions actually processed: 95644
tps = 1592.722011 (including connections establishing)
tps = 1595.906611 (excluding connections establishing)
[root@rhelclient ~]# pgbench -S -c 96 -j 8 -T 60 -h 192.168.97.44 -U
postgres pgbench
starting vacuum...end.
transaction type: SELECT only
scaling factor: 100
query mode: simple
number of clients: 96
number of threads: 8
duration: 60 s
number of transactions actually processed: 580728
tps = 9675.754717 (including connections establishing)
tps = 9695.954649 (excluding connections establishing)
[root@rhelclient ~]# pgbench -c 64 -j 32 -T 60 -h 192.168.97.44 -U
postgres pgbench
starting vacuum...end.
transaction type: TPC-B (sort of)
scaling factor: 100
query mode: simple
number of clients: 64
number of threads: 32
duration: 60 s
number of transactions actually processed: 72239
tps = 1183.511659 (including connections establishing)
tps = 1184.529232 (excluding connections establishing)
[root@rhelclient ~]# pgbench -S -c 64 -j 32 -T 60 -h 192.168.97.44 -U
postgres pgbench
starting vacuum...end.
transaction type: SELECT only
scaling factor: 100
query mode: simple
number of clients: 64
number of threads: 32
duration: 60 s
number of transactions actually processed: 388861
tps = 6479.326642 (including connections establishing)
tps = 6482.532350 (excluding connections establishing)
[root@rhelclient ~]# pgbench -c 64 -j 64 -T 60 -h 192.168.97.44 -U
postgres pgbench
starting vacuum...end.
transaction type: TPC-B (sort of)
scaling factor: 100
query mode: simple
number of clients: 64
number of threads: 64
duration: 60 s
number of transactions actually processed: 61663
tps = 1026.636406 (including connections establishing)
tps = 1027.679280 (excluding connections establishing)
[root@rhelclient ~]# pgbench -S -c 64 -j 64 -T 60 -h 192.168.97.44 -U
postgres pgbench
starting vacuum...end.
transaction type: SELECT only
scaling factor: 100
query mode: simple
number of clients: 64
number of threads: 64
duration: 60 s
number of transactions actually processed: 369321
tps = 6151.931064 (including connections establishing)
tps = 6155.611035 (excluding connections establishing)
[root@rhelclient ~]# pgbench -c 104 -j 8 -T 60 -h 192.168.97.44 -U
postgres pgbench
starting vacuum...end.
transaction type: TPC-B (sort of)
scaling factor: 100
query mode: simple
number of clients: 104
number of threads: 8
duration: 60 s
number of transactions actually processed: 80479
tps = 1337.396423 (including connections establishing)
tps = 1347.248687 (excluding connections establishing)
[root@rhelclient ~]# pgbench -S -c 104 -j 8 -T 60 -h 192.168.97.44 -U
postgres pgbench
starting vacuum...end.
transaction type: SELECT only
scaling factor: 100
query mode: simple
number of clients: 104
number of threads: 8
duration: 60 s
number of transactions actually processed: 587109
tps = 9782.401960 (including connections establishing)
tps = 9805.111450 (excluding connections establishing)
[root@rhelclient ~]# pgbench -c 300 -j 10 -T 60 -h 192.168.97.44 -U
postgres pgbench
starting vacuum...end.
transaction type: TPC-B (sort of)
scaling factor: 100
query mode: simple
number of clients: 300
number of threads: 10
duration: 60 s
number of transactions actually processed: 171351
tps = 2849.021939 (including connections establishing)
tps = 2869.345032 (excluding connections establishing)
[root@rhelclient ~]# pgbench -S -c 300 -j 10 -T 60 -h 192.168.97.44 -U
postgres pgbench
starting vacuum...end.
transaction type: SELECT only
scaling factor: 100
query mode: simple
number of clients: 300
number of threads: 10
duration: 60 s
number of transactions actually processed: 1177464
tps = 19613.592584 (including connections establishing)
tps = 19716.537285 (excluding connections establishing)
[root@rhelclient ~]#
Our new goal is to provide to our pgxc cluster of characteristics of
High Availability adding slave servers for DataNodes (and perhaps for
coordinators servers)
When we get this environment, run again the same set of tests above,
plus new ones, such as fault recovery tests simulating the loss of
cluster components, etc..
Ok, that's all for now.
Thank you very much. This is a great project and I'm very glad I found it.
Thanks again,
Hector M. Jacas |
