[omniORB] NameServer capacity

Pletyak Attila attila.pletyak@anemosky.com
Mon, 19 Feb 2001 17:44:22 +0100


owner-omniorb-list@uk.research.att.com wrote:
> 
> On Monday 19 February, Pletyak Attila wrote:
> 
> >       I just wanted to ask if anyone has experience with the capacity of the
> > NameServer.
> >
> >       We should have 4-500.000 entries concurrently at a given second, but we
> > also need fail over mechanism so as not to lose data.
> 
> The Naming service isn't really designed for that number of objects,
> but as long as you have a powerful machine it should be OK. omniNames
> uses a sequential look-up within naming contexts, though, so you would
> have to use a fairly deep tree structure to get decent performance. If
> you put all 500000 objects in the same naming context, search time
> would be awful.

That is what we are afraid of, actually.

> 
> That said, I think you should question whether you really need to
> store that many object references in the Naming service. What are you
> doing that requires so many named objects?  Normally, you would only
> store the top-level objects in the Naming service, and find other
> objects from those top-level objects.

There are a lot of users in a system, they are continually logging in
and out, but at a given moment we have 400.000 on-line. We should access
them from CGI programs (these objects are giving data to CGI programs to
display), which are known to have a limited lifetime, so those cannot
store the ids of the users locally. 

For a solution we can arrange them into 100 contexts.

But in any case is there a solution to turn off the logging function of
the NameServer? As the checkpoints are making such a big load to the
machine that it cannot really serve requests until the checkpoint
finished.

Thanks.

Attila Pletyak 
Anemo Ltd.

> 
> Cheers,
> 
> Duncan.
> 
> --
>  -- Duncan Grisby  \  Research Engineer  --
>   -- AT&T Laboratories Cambridge          --
>    -- http://www.uk.research.att.com/~dpg1 --