Re: What is better asychronous IO or multiple db_writers?

From: Steve Adams <steveadams_at_acslink.net.au>
Date: 1998/01/23
Message-ID: <34c7fddf.3204708@news.syd.aone.net.au>#1/1

Hi Jonathan,

It is a generic unix thing. You find it in most of the unix internals books, but I think the best explanation is in Maurice Bach's. Basically, it is designed to avoid the risk of process A requesting a direct read of say blocks 1,2 and 3 and then blocks on the physical I/O, then after the physcial I/O has completed and the blocks are in the buffer cache, but berfore process A wakes up, DBWR overwrites say block 2, with the result that process A gets a potentially inconsistent image of the data. Inode locking prevents this risk.

Hope this helps,
Regards,
Steve Adams
steveadams_at_acslink.net.au

---

On Thu, 22 Jan 1998 15:25:04 GMT, "Jonathan Lewis" <jonathan_at_jlcomp.demon.co.uk> wrote:

>
>Steve,
>
>> On the other hand, the management overheads of the
>> buffer cache are significant. Mostly importantly, all readers take
 a
>> shared lock on the datafile inode which delays dbwr from taking its
>> exclusive lock on the inode to write. This can significantly impact
>> dbwr performance.
>
>Is this a generic Unix thing, or a particular feature of HP-UX ?
>
>

---

My real email address is: steveadams_at_acslink.net.au Received on Fri Jan 23 1998 - 00:00:00 CST