Suggestions to improve disk I/O speed

Brian Smith BSmith at
Wed Oct 22 09:50:01 CDT 2003

I'm not sure I understand the problem correctly.

Are you saying that the disks that make up the mirror sdb slow down under
heavy Oracle load, but are fast under light load?
How are you measuring 40 MB/sec?
How are you measuring 10 MB/sec?
Are you using hdparm while the database is being used under heavy load?

It is normal for disks under heavy load to slow down their total transfer
rate due to rapid head movement.   If the disk load consisted of a single
sequential scan, then I would expect to see the maximum transfer rate (40
MB/sec).   But under "heavy load", which in my definition would include
lots of head-seeking, I would expect total data transfer to fall far below
that value.

I would take a look at the database itself.   With 1.5 GB of RAM it should
be possible to cache a lot of data and not require heavy disk load.  Of
course, I don't know the database nor the operations being performed on it
so I could be way off base here.

- Brian

                    Álvaro Palma                                                                                          
                    <apalma at        To:     linux-poweredge at                                          
                    m>                          cc:                                                                       
                    Sent by:                    Subject:     Suggestions to improve disk I/O speed                        
                    n at                                                                                            
                    10/22/2003 10:47 AM                                                                                   

I've a PowerEdge 1400SC, PERC3/DCL and 4x18GB 10Krpm U160 SCSI disk, in 2
mirrors, running RH 7.2 and stock 2.4.20-20.7 custom kernel, with megaraid
2.00.9 replacing original 1.18f, 1.5GB RAM

This server runs Oracle 8i, and the problem is that, in heavy load, the
that contains the data (/dev/sdb, separated from OS and other, but in the
SCSI channel of the controller) slow down its transfer ratio from 40MB/s to
10MB/s or less (measured using hdparm), obviously compromising seriously
performance of the system.

My question is: Do you have any ideas to improve the performance that
involve traumatic changes? I'm trying to get the money to migrate to
and RH 3.0, but meanwhile, I'd like some ideas.

I've think in move the second array to second SCSI channel in the
but really don't know if this need any further settings besides putting the
cable in place. Or maybe this compromises the data? Any suggestions about
can be done, some test that should be run? Or am I forced to move to newer
hardware? I know my problem is not CPU, average load always remain very
really very) low.

Thanks for your patience for such a long question.

Álvaro Palma

Linux-PowerEdge mailing list
Linux-PowerEdge at
Please read the FAQ at or search the list
archives at

More information about the Linux-PowerEdge mailing list