Re: OPS/OLTP Bad Idea???

I'd love to, but given the weather in the UK at present perhaps it would be best done on-site ;)

The main performance problem with OPS is partitioning the data in such a way that each machine is responsible for a distinct sub-set of the data. If this can be done perfectly a very small number of static DLM locks can be used to cover all the data and the overheads are small.

If perfect partitioning is not possible, then static locking tends to lead to 'pinging', one machine/instance being forced to
write buffered blocks to disc so that it can release the DLM lock covering those blocks to another machine.

To reduce the static lock problem, you can use dynamic or fine-grain locking, but there is then the overhead of each buffered block requiring the dynamic locating or allocating of a block - which isn't necessarily particularly fast.

There are several options, though, for getting to high availability depending on how realistic the demand for 365 x 24 is, and how much spare money is available for 'redundant' hardware.

--

Jonathan Lewis
Yet another Oracle-related web site: http://www.jlcomp.demon.co.uk

Chakravarthy KM Nalamotu wrote in message <7upuq7$df6$1_at_news.hawaii.edu>...
>We are currently investigating High Availability (365X7X24)
>option for an OLTP application.
>We heard some emphatic NO's for OPS configuration.
>The argument was that OPS is a good solution for Data
>Warehouse solutions but not for OLTP. Since OPS
>adds a lot of overhead managing locks, and adds
>extra I/O to the system.
>Can someone share their experiences in configuring
>a Highly Available OLTP system on Oracle DB.
Received on Fri Oct 22 1999 - 11:03:25 CDT