Network Problems / Sockets dropped by RS6000

[LHLTech]

Ok.. my first question here...

We have :-

A system running AIX 4.3.3.0

A client/server application (our own) that uses winsock.dll on the client side and the standard sockets interface on the server side.

A problem that manifests itself in the following way. The client side receives a status message of wsa_conn_aborted which I am reliably informed by our programmers is a result of the client software trying to use an already established socket, only to be told by the server side that this socket no longer exists.

Now, all that we can see is that the RS6000 in question is labouring at times, seemingly with a memory and CPU bottleneck. Netstat -D shows that no TCP or UDP packets are being dropped by the interfaces but plenty of IP packets seem to be getting dropped (this is in the protocols section).

So, my question...

Is this what I should expect?! i.e. Would very high CPU usage (or even saturation) cause active sockets to be discarded?

And if so... can anyone point me to some sort of documentary evidence of this so that I can show the required people that it's a deficiency with the RS6000 rather than the client/server app.

If you need more info/better worded questions, please ask.

Regards, LHLTech

IBM Certified Specialist - AIX System Support
Halfway through CATE exams!

 

[MikeLacey]

It doesn't seem likely that AiX (or any version of Unix) would drop sockets like that...

With respect to your application programmers - it's more probable that it's a coding error on their part. Sockets programming is not as easy as some make out, coding for the exceptions is *far* from trivial.

Having said that. Have you talked to IBM about this? If you have Call AiX you could start a query off via that route and it will, eventually, get to the network experts who can help you. You have to be careful going this route though, to make sure that it's always obvious that yours is a technical queryl; that you're not asking them to help you write your application (for which they'll want to charge you) You'll have to show them that their sockets interface does not behave as documented, and that might be quite a trick...

To be honest, it hardly matters whose fault it is - your application programmers will end up coding around it anyway.... <wry smile> Mike
michael.j.lacey@ntlworld.com
Email welcome if you're in a hurry or something -- but post in tek-tips as well please, and I will post my reply here as well.

 

[gileb]

Hi,
As Micheal says, i also think you should see that from the program side - maybe related to SO_REUSEADDR ?