Hi Richard Thanks for the reply, I know it has some impact unfortunately this database was created by someone who left the company several monsth ago. I wonder, why HP is gonna do 2 data block reads? Does it work this way: - Oracle does index range scan and scans 10 blocks, 10 system calls, HP runs 10 systems calls and read 10 8k blocks. Or how does it work? By the way I am putting this to Direct I/O soon. Convosync and mincache to direct, I recogn it would achieve some improvements. TIA Alex On 11/14/06, Richard J. Goulet <rgoulet@xxxxxxxxxx> wrote:
Alex Having been a long time HP-UX and OnlineJFS user, your making the machine do more work than is needed. Assume that your looking for one row of data in one data block. HP is going to read 2 data blocks of data into memory for you. The Oracle manuals are pretty good at saying that your database block size should be equal to or a multiple of the OS block size. BTW: you've also put dbwr into a quagmire at the same time. Since it's wetting 4K blocks the OS is going to cache those writes till it gets an 8k write to that disk location before it does anything. Now I'm not sure what it does in the event that you never write the remaining 4K of data, but I'd imagine that it would cache the 4k block in swap for a rather long time. Dick Goulet, Senior Oracle DBA 45 Bartlett St Marlborough, Ma 01752, USA *Tel.:* 508.573.1978 |F*ax: * 508.229.2019 | Cell:508.742.5795 RGoulet@xxxxxxxxxx : POWERING TRANSFORMATION ------------------------------ *From:* oracle-l-bounce@xxxxxxxxxxxxx [mailto: oracle-l-bounce@xxxxxxxxxxxxx] *On Behalf Of *amonte *Sent:* Tuesday, November 14, 2006 9:59 AM *To:* oracle-l@xxxxxxxxxxxxx *Subject:* Aligning Oracle Blocks with Filesystem Block Size Hi I have a 9i database running on HP-UX which uses 4k db_block_size. The filesystem used is OnlineJFS (vxfs version 4). The Filesystem block size is 8k (f_bsize 8192) I was wondering what sort of impact can this missalignment cause? Would I/O system perform twice I/O it should? TIA Alex