[omniORB] omniORB timeout problems on systems with two network interface cards

[Nikola Radovanovic]

Hi,
we have omniORB timeout problems on systems with two network interface 
cards (NIC)

We have system consisting out of 2 PCs running windows XP Pro 32 bit, 
each with two NIC's.
Our software uses omniORB-4.1.1 with following params:
    resetTimeOutOnRetries        1
    verifyObjectExistsAndType    0
    traceLevel                    25
    
also, DisableDHCPMediaSense is set to 1 in 
HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Services\Tcpip\Parameters
    
We wanted to see how omniORB behaves in cases when we plug out one 
(primary) network cable.

Main test scenario:
On one PC runs CORBA server, binded to both NICs via ORBendpoints passed 
through command line.
We than pull out primary network cable on server PC and try to run 
client on second PC. Client will try to obtain server object reference 
from omniNames and call some (simple) method on it.
We have tried different timeout configurations and get the following 
results:

1.    all timeouts are disabled (this includes connection and both 
(global and per reference) call timeouts, per thread call timeout is 
disabled in omniORB configuration)
    When opening connection to omniNames it takes (roughly) 5 secs 
(probably due to scanGranularity which is 5 secs by default) for omniORB 
on client side to figure out that connection to omniNames via primary 
connection can not be established. Switch rope occurs and connection 
will be established via secondary network. Reference is obtained via 
that connection. When method is called on that server reference object, 
it takes roughly 5 sesc for omniORB on client side to figure out that 
connection to server via primary connection can not be established. 
Switch rope will occur and connection is established via secondary 
network. After that, method call on server reference will be executed 
successfully.
    
2.     connection timeout is set to value less than 5 secs (e.g. 2 or 3 
secs); global and per reference call timeouts are disabled
    Behavior is as expected: it takes specified amount of time for 
client to figure out that connection to server via primary connection 
could not be established.
    Beside that, behavior is like in previous scenario.
    
3.    global call timeout is set to 10 secs, connection timeout is 3 
secs and per reference call timeout is not specified
    Connection timeout would not be detected (3 secs as expected) but 
global call timeout (10 secs) will expire.
    This is the pattern when global call timeout is set to less than 20 
secs. For global call timeout greater than 20 secs, time to detect 
failure while opening connection via primary LAN is roughly 20 secs.
    In short, in this scenario connection timeout is totally ignored and 
it takes roughly 20 secs to detect failure to open connection (4x more 
than in scenario 1)
    
4.    per reference call timeout is set to 10 secs, connection timeout 
is 3 secs and global call timeout is not specified
    It takes roughly 3 secs (connection timeout occurs) to connect to 
omniNames and acquire server object reference. Per reference call 
timeout on that reference is set to 10 secs. When method is called on 
that reference connection failure will not be detected, instead call 
timeout will expire. If per reference call timeout is set to value 
greater than 20 secs, time to detect failure while opening connection 
via primary LAN is roughly 20 secs.
    In short, in this scenario connection timeout is totally ignored on 
reference with per reference call timeout set.

thanks for the replies
    Telvent DMS