MySQL Tokudb engine benchmark

MySQL TokuDB Engine performance benchmark on FusionIO
Environment:
Machine DELL PowerEdge R720 * 2
CPU Intel(R) Xeon(R) CPU E5-2630 v2 @ 2.60GHz * 24
memory 128GB
OS CentOS release 6.5 (Final)
Kernel 2.6.32-431.el6.x86_64 #1 SMP
DB version 10.0.21-MariaDB-log MariaDB Server
raid Internal Raid
File system Ext4 with options
disk FusionIO SX300 1.6TB

Database configuration:
InnoDB:
[client]
port = 3306
socket = /storage/fioa/mysql3306/data/mysql.sock
[mysqld]
port = 3306
server-id = 100011
replicate_ignore_db = test
skip-name-resolve
lower_case_table_names = 1
character-set-server = utf8
replicate_wild_ignore_table = dbmon.%,sys.%,zabbix.%,mysql.%,test.%,bbs.%,slow_query_log.%
tmpdir = /storage/fioa/mysql3306/mysql-tmpdir
key_buffer_size = 64M
sort_buffer_size = 8M
read_buffer_size = 8M
read_rnd_buffer_size = 8M
query_cache_type = 0
max_allowed_packet = 628M
myisam_sort_buffer_size = 128M
tmp_table_size = 1024M

join_buffer_size = 16M
back_log = 500
open_files_limit = 6000
innodb_open_files = 4096
table_open_cache = 768
thread_cache_size = 12
max_connections = 6000
thread_concurrency = 48
datadir = /storage/fioa/mysql3306/data
slow_query_log = 1
long_query_time = 1
slow_query_log_file = /storage/fioa/mysql3306/mysql-slow.log
log-error = /storage/fioa/mysql3306/mysql-error.log
log-warnings
sysdate-is-now
###replication setting
log-slave-updates
sync_binlog = 0
binlog_format = row
log_bin_trust_function_creators = 1
log-bin = /storage/fioa/mysql3306/binlog/mysql-bin
log-slave-updates
relay_log_purge = 0
relay-log = /storage/fioa/mysql3306/binlog/mysql-relay-bin
relay-log-index = /storage/fioa/mysql3306/binlog/mysql-relay-bin.index

relay_log_info_file = /storage/fioa/mysql3306/binlog/relay.info
master-info-file = /storage/fioa/mysql3306/binlog/master.info
#innodb setting
innodb_data_home_dir = /storage/fioa/mysql3306/data
innodb_data_file_path = ibdata0:1024M;ibdata1:1024M;ibdata2:1024M;ibdata3:1024M:autoextend
innodb_log_group_home_dir = /storage/fioa/mysql3306/data
innodb_log_files_in_group = 2
innodb_log_file_size = 512M
innodb_buffer_pool_size = 65536M
innodb_log_buffer_size = 128M
innodb_flush_log_at_trx_commit = 2
innodb_flush_method = O_DIRECT
innodb_thread_concurrency = 6
innodb_file_per_table = 1
transaction-isolation = READ-COMMITTED
[mysqldump]
quick
[myisamchk]
read_buffer = 4M
write_buffer = 4M
[mysql]
default-character-set = utf8

no_auto_rehash
prompt = "R:m:s d> "
pager = "more"
TokuDB :
[client]
port = 3306
[mysqld]
port = 3306
server-id = 100011
replicate_ignore_db = test
skip-name-resolve
lower_case_table_names = 1
character-set-server = utf8
plugin-load=ha_tokudb
tokudb_cache_size =64G
tokudb_directio=1
tokudb_commit_sync = 0
replicate_wild_ignore_table = dbmon.%,sys.%,zabbix.%,mysql.%,test.%,bbs.%,slow_query_log.%
plugin-load=ha_tokudb

read_buffer_size = 8M
read_rnd_buffer_size = 8M
query_cache_type = 0
myisam_sort_buffer_size = 128M
tmp_table_size = 1024M
join_buffer_size = 16M
back_log = 500
open_files_limit = 6000
innodb_open_files = 4096
table_open_cache = 768
thread_cache_size = 12
max_connections = 6000
thread_concurrency = 48
datadir = /storage/fioa/mysql3306/data
slow_query_log = 1
long_query_time = 1
slow_query_log_file = /storage/fioa/mysql3306/mysql-slow.log
log-error = /storage/fioa/mysql3306/mysql-error.log
log-warnings
sysdate-is-now
###replication setting
log-slave-updates

sync_binlog = 0
binlog_format = row
log_bin_trust_function_creators = 1
log-bin = /storage/fioa/mysql3306/binlog/mysql-bin
log-slave-updates
relay_log_purge = 0
relay-log = /storage/fioa/mysql3306/binlog/mysql-relay-bin
relay-log-index = /storage/fioa/mysql3306/binlog/mysql-relay-bin.index
relay_log_info_file = /storage/fioa/mysql3306/binlog/relay.info
master-info-file = /storage/fioa/mysql3306/binlog/master.info
#innodb setting
innodb_data_home_dir = /storage/fioa/mysql3306/data
innodb_data_file_path = ibdata0:1024M;ibdata1:1024M;ibdata2:1024M;ibdata3:1024M:autoextend
innodb_log_group_home_dir = /storage/fioa/mysql3306/data
innodb_log_files_in_group = 2
innodb_log_file_size = 512M
innodb_flush_method = O_DIRECT
innodb_buffer_pool_size = 65536M
innodb_log_buffer_size = 128M
innodb_flush_log_at_trx_commit = 2
innodb_thread_concurrency = 6
innodb_file_per_table = 1
transaction-isolation = READ-COMMITTED
[mysqldump]
quick

[myisamchk]
read_buffer = 4M
write_buffer = 4M
[mysql]
default-character-set = utf8
no_auto_rehash
prompt = "R:m:s d> "
pager = "more"
We create 1000 DWs first to test on both InnoDB and TokuDB engine –separate test from 32 threads to 512 threads

TpmC result on different threads
InnoDB Threads 32 64 128 256 512 1024
TpmC 46376.699 48921.6 47090.801 41429.398 39957.102 33683.102
TokuDB Threads 32 64 128 256 512 1024
TpmC 41694.898 30663.7 22092 13243.9 N/A N/A
TokuDB does not support foreign keys, so we add indexes for queries, even so in a heavy-load environment TokuDB still has too
many system locks that result in a low TpmC.
In this test, TokuDB engine got a poor performance (TpmC decreases rapidly, many connections timeout appear due to system locks
in MySQL when threads up to 512).

Look at total size of these two engines:
InnoDB tpcc1000 DB:
TokuDB tpcc1000 DB:

Simple test table mytest10 size:

Compression ratio of these two engines is nearly 2:1

Mydbtest performance in different threads
We create four tables for this test on four modes (100% insert, 50%insert+50%select, 50%insert+50%update, mix)
mode threads 256 512
insert(innodb) Summary: exec=23768/s, qtps=47669/s Summary: exec=13319/s, qtps=54116/s
insert(tokudb) Summary: exec=23640/s, qtps=47415/s Summary: exec=12818/s, qtps=51969/s
50insert+50sel(innodb) Summary: exec=16606/s, qtps=66803/s Summary: exec=16418/s, qtps=66461/s
50insert+50sel(tokudb) Summary: exec=15324/s, qtps=61667/s Summary: exec=15317/s, qtps=61979/s
50insert+50upd(innodb) Summary: exec=10078/s, qtps=40657/s Summary: exec=10016/s, qtps=40842/s
50insert+50upd(tokudb) Summary: exec=9890/s, qtps=39890/s Summary: exec=9843/s, qtps=40155/s
mix mode(innodb) Summary: exec=7573/s, qtps=61504/s Summary: exec=7728/s, qtps=63939/s
mix mode(tokudb) Summary: exec=6788/s, qtps=55262/s Summary: exec=6833/s, qtps=56753/s
InnoDB still has a better performance than TokuDB when use simple queries to avoid MySQL locks.

option
user xxxxx/xxxx@10.128.6.35:3306:tpcc1000
name tpcc1000
time 10m
log toku_insert100%_512_parallel__%p.log
declare
a int 1 200000000
b int 40000 6000000000
c CHAR 1 20
d CHAR 1 50
e char 1 4
f CHAR 1 5
g char 1 6
h char 3 15
j int 35 555555555
h char 3 15
k date -90 -60
l int 1 49999999
begin
insert into mytest10 values(:a,:b,:c,:d,:e,:f,:g,:h,:j);
--- /*insert into mytest values (:a,:c,:d);*/
--- /*select * from mytest where a=:a;*/
--- /*insert into mytest2 values (:a,:b,:k,:h);*/
--- /*update mytest set a=a-1 where a=:a;*/
--- /*select * from mytest2 where k=:k;*/
end

Conclusion:
Reference:https://www.percona.com/blog/2015/07/24/innodb-vs-tokudb-in-linkbench-benchmark/
Though, it worth remembering, that:
 On a fast expensive storage, TokuDB provides a better compression, which allows to store more data in limited capacity
 TokuDB still writes two time less than InnoDB, that mean twice longer lifetime for SSD (still expensive).
Also looking at the results, I can make the conclusion that InnoDB compression is inefficient in its implementation, as it is not able to get benefits: first, from doing
less reads (well, it helps to get better than uncompressed InnoDB, but not much); and, second, from a fast storage.

In this picture, TokuDB has a better performance on IO slow storage like SATA SAS disk (Author uses M500 SSD disk for this test)
For us data quality is more important than storage cost. We have already used SSD&PCIe cards for all database system. TokuDB
has big advantage for its high compression ratio. I thinks it’s more suitable for log system rather than core system.

MySQL Tokudb engine benchmark

More Related Content

What's hot (20)

Viewers also liked (12)

Similar to MySQL Tokudb engine benchmark (20)

More from Louis liu (20)

Recently uploaded (20)

MySQL Tokudb engine benchmark