Subject | Re: Time sharing |
---|---|
Author | Alexander V.Nevsky |
Post date | 2003-05-21T11:10:10Z |
--- In ib-support@yahoogroups.com, Milan Babuskov <albis@e...> wrote:
Milan, look at
http://www.ibphoenix.com/main.nfs?a=ibphoenix&page=ibp_sem_sm
In my practice configuration of CS server for less than 100
simultaneous users requires only database cashe pages parameter
tuning. Generally it should be less than for SS. If application mainly
selects data, my practice shows cashe about 2048 pages per user is
good if there is enough memory on the server and this does not cause
processes swapping. Perhaps with new in-memory sorting in FB1.5 it is
better to shorten cashe buffers and leave memory to sorting/grouping.
If application mainly changes data, I think default 75 page buffers
per connection for CS will be optimal, to reduce load on lock manager
and interprocess cashe syncronisation.
on SMP machines.
Best regards, Alexander.
> I have the same problem as Valdir, so I will try Classic too. Arethere
> any special things we need to look into if we replace SS withClassic?
Milan, look at
http://www.ibphoenix.com/main.nfs?a=ibphoenix&page=ibp_sem_sm
In my practice configuration of CS server for less than 100
simultaneous users requires only database cashe pages parameter
tuning. Generally it should be less than for SS. If application mainly
selects data, my practice shows cashe about 2048 pages per user is
good if there is enough memory on the server and this does not cause
processes swapping. Perhaps with new in-memory sorting in FB1.5 it is
better to shorten cashe buffers and leave memory to sorting/grouping.
If application mainly changes data, I think default 75 page buffers
per connection for CS will be optimal, to reduce load on lock manager
and interprocess cashe syncronisation.
> Another question: Can I run both SS and CS on same machine?For FB1 - no, for FB1.5 - not sure.
> P.S. I use Linux.On Linux CS advantages are more significant than on Win, especially
on SMP machines.
Best regards, Alexander.