[omniORB] Visibroker osagent as naming service for omniORB po ssible?

Ivanov, Roumen Roumen.Ivanov@drkw.com
Wed, 8 Aug 2001 09:25:39 +0200


In fact I don't know what exactly happens, but it happens not only with my
collection of clients (NS clients). We have increased the number of threads
per process to 512 or more. The naming service still "dies" when 200 clients
are trying to read from it every 5 seconds. The other case is a server which
is launching about 50 processes almost at once. Each of them is trying to
read from the NS. In result - the same. When I'm writing "dies" this is some
times crash and some times COMM_FAILURE. The second might look like threads
limit reached, but I suspect it is rather a deadlock, because the changes in
the kernel configuration did not affect it. I will try to find some time and
run NS in a debugger.
Thanks,
Roumen


-----Original Message-----
From:	Duncan Grisby [SMTP:dgrisby@uk.research.att.com]
Sent:	Tuesday, August 07, 2001 12:09
To:	Ivanov, Roumen
Cc:	omniorb-list@uk.research.att.com
Subject:	Re: [omniORB] Visibroker osagent as naming service for
omniORB po ssible? 

On Monday 6 August, "Ivanov, Roumen" wrote:

> We had similar situation here. We are running omniNames 3.0.2. When many
> clients try to connect almost simultaneously (average - 40 per sec.), the
> naming service either dies or gets deadlocked. Looking for a quick
solution,
> we had to extend the reconnect interval limits x 10 in order to keep the
NS
> alive. This might be not the case with other versions or implementations
> though.

omniNames shouldn't become unstable under heavy load. What platform
are you running on?  The worse that should happen is that the process
reaches its limit of threads or file descriptors and starts turning
clients away.

Can you post a trace from -ORBtraceLevel 15 of omniNames dying?

Cheers,

Duncan.

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