slow performance during write (sync on commit)

[Database Administrator]

Hi all,

I'm newbie to BerkeleyDB and testing 4.6.21 on CentOS 5 on 2 Dell
PowerEdge 1950 (8GB RAM each, SAS 6/iR controller). The only
difference is disk:
machine A: 2 x 146GB (15000rpm, RAID 1)
machine B: 2 x 500GB (7200rpm, RAID 1)

Both systems has drive cache enabled as shown in dmesg:

"...
SCSI device sda: drive cache: write back
...."

Some IO performance test tools show result as expected (machine A is a
bit faster than machine B):

- time dd if=/dev/zero of=/tmp/a bs=64k count=100k
- iozone
- bonnie++

However, when I run load test against openldap 2.4.10 using BerkeleyDB
as backend, write performance of machine B is much faster than machine
A (ranging from 8 - 10 times) with the same DB_CONFIG setting. Then I
tried performance tool from Oracle : http://www.oracle.com/technology/pro...files/perf.zip

Test of all following items except the last one show machine A is a
bit faster than machine B:

Data Store (read):
Data Store (bulk read):
Data Store (write):
Transactional Data Store (read):
Transactional Data Store (write, in-memory logging):
Transactional Data Store (write, no-sync on commit):
Transactional Data Store (write, write-no-sync on commit):
Transactional Data Store (write, sync on commit):

Performance of machine B on (write, sync on commit) is almost 2 to 3
times of that of machine A. I've no idea about the reason. Would
anyone please help?

Thank you very much.
Regards,
/ST Wong