>>> As you say, the sum of those 9 page caches will be significant in the consumption of available RAM. Additionally, each connection to the SS process spawns a thread that eats the same amount of RAM as the main process
>
> At 08:43 a.m. 20/02/2013, Thomas Steinmaurer wrote:
>
>> Are we talking about the page cache here only? I'm curious, because this
>> would mean we are close to the CS/SC architecture then. ;-)
>
> No, not closer than it ever was. SS has one shared cache per open database. CS/SC have one cache per connection. The OP has 9 databases per customer. He has n customers (we don't know how many, do we?) So, in 2GB of RAM, SS will be maintaining 9 * n caches of unspecified sizes as well as (potentially) 9 * n * m SS threads (each circa 3.5 MB), where m is the total number of client connections.

I am/was confused about: "Additionally, each connection to the SS
process spawns a thread that eats the same amount of RAM as the main
process"

3.5 MB per SS thread is different to whatever "same amount of RAM as the
main process" means. Anyway ... ;-)

>>> although there's not necessarily a 1:1 correspondence as the server will reallocate a recently de-allocated thread if one is available. My point was that RAM usage will accumulate towards a tipping point on an under-resourced system like this one.
>>
>> I fully agree, although I doubt that getting CPU bound is the result of
>> running out of RAM.
>
> 50% is CPU-bound? The main symptom of the OP seemed to be the inability of the server to accept new connections.

If it is a 2 CPU/core environment (which we don't know), one of them is
fully utilized.

In case of SS, this could mean, a single database hogs a single CPU/core
and it can't get beyond 50% CPU usage, as SS, even in 2.5, isn't able to
fully utilize a SMP environment for a single database.

But with 9 * <nr_of_agencies> databases, chances are good that CPUS
aren't that bored.

> Actually, it would be interesting to know whether this 32-bit box has SMP. The OP also implied that Oracle and a webserver were hosted on the same box...

Right and we have to less information on the environment anyway.


--
With regards,
Thomas Steinmaurer
http://www.upscene.com/