Description:
Performance of 6.0.10 binary in sysbench OLTP_RW test is notable worse
than in 6.0.9 especially with higher number of threads.
# Test: OLTP_RW
#
# Server 1 - 6.0.9-alpha
# Server 2 - 6.0.10-alpha
#
# Server 1 Server 2
# Thread FALCON FALCON
4 1104.83 939.59
16 841.68 137.01
64 404.43 45.38
128 418.83 30.62
256 412.99 22.25
From sar output it looks that this slowdown was caused by some locking
that was added after 6.0.9 because there is almost no activity at the user side.
CPU %user %nice %system %iowait %steal %idle
Run with 256 threads for 6.0.9/Falcon:
Average: all 48.67 0.00 8.02 0.42 0.00 42.88
Run with 256 threads for 6.0.10/Falcon:
Average: all 2.62 0.00 0.83 0.04 0.00 96.51
Also below is output of oprofile for runs with 256 threads for 6.0.9 and 6.0.10.
Set of functions and their percentage are pretty similar (except SyncObject::grantLocks) but number of samples is differ by a factor of ~20.
6.0.9
-----
samples % symbol name
104987 40.2752 SyncObject::lock(Sync*, LockType, int)
34061 13.0665 SyncObject::unlock(Sync*, LockType)
13599 5.2169 DeferredIndex::scanIndex(IndexKey*, IndexKey*, int, Bitmap*)
11770 4.5152 Index::scanIndex(IndexKey*, IndexKey*, int, Transaction*, Bitmap*)
7323 2.8093 Transaction::visible(Transaction*, unsigned int, int)
5421 2.0796 DeferredIndex::deleteNode(IndexKey*, int)
4914 1.8851 Sync::lock(LockType)
4601 1.7650 my_hash_sort_simple
2657 1.0193 Sync::Sync(SynchronizationObject*, char const*)
2173 0.8336 Sync::~Sync()
1992 0.7642 ptr_compare_0
1955 0.7500 Record::addRef()
1576 0.6046 Bitmap::nextSet(int)
1513 0.5804 Table::fetch(int)
1509 0.5789 Transaction::initialize(Connection*, unsigned int)
1471 0.5643 SerialLogRecord::getInt()
1375 0.5275 Transaction::releaseDependency()
1253 0.4807 SyncObject::grantLocks()
6.0.10
------
4975 37.7037 SyncObject::lock(Sync*, LockType, int)
1602 12.1410 SyncObject::unlock(Sync*, LockType)
590 4.4714 DeferredIndex::scanIndex(IndexKey*, IndexKey*, int, Bitmap*)
496 3.7590 Index::scanIndex(IndexKey*, IndexKey*, int, Transaction*, Bitmap*)
291 2.2054 Transaction::visible(Transaction*, unsigned int, int)
273 2.0690 SyncObject::grantLocks()
240 1.8189 Sync::lock(LockType)
200 1.5157 RecordScavenge::inventoryRecord(Record*)
161 1.2202 DeferredIndex::deleteNode(IndexKey*, int)
146 1.1065 Cache::fetchPage(Dbb*, int, PageType, LockType)
133 1.0080 MemMgr::blockSize(void*)
130 0.9852 Sync::~Sync()
99 0.7503 Sync::Sync(SynchronizationObject*, char const*)
85 0.6442 Bitmap::nextSet(int)
76 0.5760 ptr_compare_0
How to repeat:
- Download and build the latest sysbench from sysbench.sf.net
- prepare database
sysbench \
--test=oltp \
--oltp-table-size=1000000 \
--oltp-dist-type=uniform \
--init-rng=on \
--mysql-user=root \
--mysql-table-engine=falcon \
--mysql-db=test prepare
- run the test
sysbench \
--oltp-table-size=1000000 \
--oltp-dist-type=uniform \
--init-rng=on \
--mysql-user=root \
--mysql-socket=/tmp/mysql.sock \
--mysql-db=test \
--test=oltp \
--max-requests=0 \
--max-time=300 \
--oltp-read-only=off \
--num-threads=<4,16,64,128,256> \
run
© 2025, Oracle Corporation and/or its affiliates
Description: Performance of 6.0.10 binary in sysbench OLTP_RW test is notable worse than in 6.0.9 especially with higher number of threads. # Test: OLTP_RW # # Server 1 - 6.0.9-alpha # Server 2 - 6.0.10-alpha # # Server 1 Server 2 # Thread FALCON FALCON 4 1104.83 939.59 16 841.68 137.01 64 404.43 45.38 128 418.83 30.62 256 412.99 22.25 From sar output it looks that this slowdown was caused by some locking that was added after 6.0.9 because there is almost no activity at the user side. CPU %user %nice %system %iowait %steal %idle Run with 256 threads for 6.0.9/Falcon: Average: all 48.67 0.00 8.02 0.42 0.00 42.88 Run with 256 threads for 6.0.10/Falcon: Average: all 2.62 0.00 0.83 0.04 0.00 96.51 Also below is output of oprofile for runs with 256 threads for 6.0.9 and 6.0.10. Set of functions and their percentage are pretty similar (except SyncObject::grantLocks) but number of samples is differ by a factor of ~20. 6.0.9 ----- samples % symbol name 104987 40.2752 SyncObject::lock(Sync*, LockType, int) 34061 13.0665 SyncObject::unlock(Sync*, LockType) 13599 5.2169 DeferredIndex::scanIndex(IndexKey*, IndexKey*, int, Bitmap*) 11770 4.5152 Index::scanIndex(IndexKey*, IndexKey*, int, Transaction*, Bitmap*) 7323 2.8093 Transaction::visible(Transaction*, unsigned int, int) 5421 2.0796 DeferredIndex::deleteNode(IndexKey*, int) 4914 1.8851 Sync::lock(LockType) 4601 1.7650 my_hash_sort_simple 2657 1.0193 Sync::Sync(SynchronizationObject*, char const*) 2173 0.8336 Sync::~Sync() 1992 0.7642 ptr_compare_0 1955 0.7500 Record::addRef() 1576 0.6046 Bitmap::nextSet(int) 1513 0.5804 Table::fetch(int) 1509 0.5789 Transaction::initialize(Connection*, unsigned int) 1471 0.5643 SerialLogRecord::getInt() 1375 0.5275 Transaction::releaseDependency() 1253 0.4807 SyncObject::grantLocks() 6.0.10 ------ 4975 37.7037 SyncObject::lock(Sync*, LockType, int) 1602 12.1410 SyncObject::unlock(Sync*, LockType) 590 4.4714 DeferredIndex::scanIndex(IndexKey*, IndexKey*, int, Bitmap*) 496 3.7590 Index::scanIndex(IndexKey*, IndexKey*, int, Transaction*, Bitmap*) 291 2.2054 Transaction::visible(Transaction*, unsigned int, int) 273 2.0690 SyncObject::grantLocks() 240 1.8189 Sync::lock(LockType) 200 1.5157 RecordScavenge::inventoryRecord(Record*) 161 1.2202 DeferredIndex::deleteNode(IndexKey*, int) 146 1.1065 Cache::fetchPage(Dbb*, int, PageType, LockType) 133 1.0080 MemMgr::blockSize(void*) 130 0.9852 Sync::~Sync() 99 0.7503 Sync::Sync(SynchronizationObject*, char const*) 85 0.6442 Bitmap::nextSet(int) 76 0.5760 ptr_compare_0 How to repeat: - Download and build the latest sysbench from sysbench.sf.net - prepare database sysbench \ --test=oltp \ --oltp-table-size=1000000 \ --oltp-dist-type=uniform \ --init-rng=on \ --mysql-user=root \ --mysql-table-engine=falcon \ --mysql-db=test prepare - run the test sysbench \ --oltp-table-size=1000000 \ --oltp-dist-type=uniform \ --init-rng=on \ --mysql-user=root \ --mysql-socket=/tmp/mysql.sock \ --mysql-db=test \ --test=oltp \ --max-requests=0 \ --max-time=300 \ --oltp-read-only=off \ --num-threads=<4,16,64,128,256> \ run