| T.R | Title | User | Personal
Name | Date | Lines |
|---|
| 482.1 | location, location, location ... | ALFSS2::FOLDEVI | DB/DW Tech Support, 343.2368 | Thu Apr 24 1997 23:18 | 14 |
|
Well, the genaral problem with OPS is that you want to achieve as high
a level of "localization of data" as possible, i.e. you do NOT want to
have all requests for data go across the MC. It kind of depends on
whether you're looking for performance or for high availability.
Remember that the MC has a limited throughput (but it's of course much
better in v1.4 than in 1.) If you have no way of partitioning the data
and the data requests you may not achieve the performance you'd expect.
Do you need all nodes to process user requests?
Can you control where the user requests are located?
Can you partition the database over two nodes (or however many you
have)?
- Lars
|
| 482.2 | Any performance improvement? | HGOVC::PATRICKNG | | Fri Apr 25 1997 10:16 | 10 |
| Thanks.
Given more read queries than write transaction, does OPS improves
performance by spread out concurrent query on two nodes?
I think OPS is good for load balancing and high available. If there are
multiple large queries running at the same time, the turnaround time of
each query should be shorten if two nodes are running.
Any comment?
|