FlashCache benchmarks on tpcc-like workload

Binary package of FlashCache for CentOS 5.5, Kernel 2.6.18-194 is there http://www.percona.com/downloads/TESTING/FlashCache/centos-2.6.18-194.el5.tar.gz

• Hardware: Dell PowerEdge R900
• IO subsystems
  • RAID10: 8 disks SAS 2.5” 15K
  • SSD: Intel X25-M 160GB ( 2 gen ) ( Model Number: INTEL SSDSA2M160G2GC, Firmware Revision: 2CV102HA )
  • FlashCache: build over SSD-2
• Filesystem XFS, builds as: mkfs.xfs -f -d su=16384,sw=16 /dev/sda2
  • mounted with -o nobarrier,swalloc option
• deadline IO scheduler for all devices
• InnoDB files layout: ibdata1 and ib_logfile* are placed on separate RAID partition ( not on FlashCache or SSD)
• MySQL version: Percona Server/XtraDB 5.1.45 rel10.1

Benchmark: tpcc-mysql workload, 1000W ( about 100GB of data ). Buffer pool 22GB, for Cache partition allocated 70GB ( flashcache_create -b 4k -s 70g cachedev /dev/sdb /dev/sda2 )

There were next storage configurations tested:

• Tables on RAID ( raid )
• Tables on SSD ( ssd )
• FlashCache is configured for 20% dirty pages
• FlashCache is configured for 80% dirty pages

[mysqld]
basedir=/usr/local/mysql
user=root
socket=/var/lib/mysql/mysql.sock
server_id=1
local_infile=1
tmpdir=/mnt/fio
datadir=/mnt/fio320
skip-grant-table
log_slow_queries=ON
slow_query_log_file=slow.log
innodb_buffer_pool_size=24G
innodb_data_file_path=ibdata1:10M:autoextend
innodb_file_per_table=1
innodb_flush_log_at_trx_commit=1
innodb_fast_checksum=1
innodb_log_buffer_size=8M
innodb_log_files_in_group=2
innodb_log_file_size=1900M
innodb_thread_concurrency=0
innodb_flush_method             = O_DIRECT
innodb_write_io_threads=4
innodb_read_io_threads=4
innodb_io_capacity=500
innodb_max_dirty_pages_pct=90
max_connections=3000
query_cache_size=0
skip-name-resolve
table_cache=10000
[mysql]
socket=/var/lib/mysql/mysql.sock
[client]
socket=/var/lib/mysql/mysql.sock

set -u
set -x
set -e
DR="/bench"
BD="/mnt/fio/back"
WT=10
RT=7200
ROWS=80000000
log2="/data/log"
ENAME="bp_22_7200_cachdev_80pct_lru"
mkdir -p $ENAME
function waitm {
while [ true ]
do
mysql -e "set global innodb_max_dirty_pages_pct=0" mysql
wt=`mysql -e "SHOW ENGINE INNODB STATUS\G" | grep "Modified db pages" | sort -u | awk '{print $4}'`
if [[ "$wt" -lt 100 ]] ;
then
mysql -e "set global innodb_max_dirty_pages_pct=90" mysql
break 
fi
echo "mysql pages $wt"
sleep 10
done
}
for par in  22
do
/usr/local/mysql/libexec/mysqld --defaults-file=/etc/my.opt.cnf --basedir=/usr/local/mysql --datadir=$DR  --innodb_data_home_dir=$log2 --innodb_log_group_home_dir=$log2 --innodb_thread_concurrency=0 --innodb_buffer_pool_size=${par}GB  &
sleep 60
mysql -e "select count(*) from order_line" tpcc
mysql -e "select count(*) from customer" tpcc
iostat -dxm 5 260 >> $ENAME/iostat.5.txt &
vmstat  5 260 >> $ENAME/vmstat.5.txt &
./tpcc_start localhost tpcc root "" 1000 16 10 7200 | tee -a $ENAME/res1.out
waitm
mysqladmin  shutdown
done