[sctp-dev] Problem in receiving notifications on SCTP channel events

[Michal Lefler]

Chris.
You are right.
We did manage to get the required notifications within a short while
after playing with the different OS net.sctp. params (such as
net.sctp.path_max_retrans etc.).
Thanks,
Michal.

-----Original Message-----
From: Chris Hegarty [mailto:chris.hegarty at oracle.com] 
Sent: Friday, June 24, 2011 6:41 PM
To: Michal Lefler
Cc: sctp-dev at openjdk.java.net
Subject: Re: [sctp-dev] Problem in receiving notifications on SCTP
channel events

Hi Michal,

Notification of these events is being passed by the native SCTP stack to

the Java implementation. If they are not coming in a timely manner then 
it may be that there is a property in the OS controlling the heartbeat 
timeout that could help. What platform are you running on?

Another question, are you saying that the notification is slow because 
of the fact that the selector doesn't fire? or is it just that you 
happen to be using this model rather then blocking. Do you know if the 
same delay is seen when using blocking SCTP channels?

I'm sorry but I haven't done much testing of this type of scenario.

-Chris.

On 06/24/11 02:51 PM, Michal Lefler wrote:
> Hi.
>
> We use SCTP channel, with asynchronous reads by a selector.
>
> We are using it as we saw in the SCTP example:
>
> private boolean read(SelectionKey k) {
>
> final SctpChannel ch = ...
>
> ...
>
> MessageInfo messageInfo = ch.receive(bb, ch, associationHandler);
>
> ...
>
> In our associationHandler implementation, we override all the methods
in
> "AbstractNotificationHandler",
>
> and for now, we simply write a log line for each.
>
> The problem is that we don't get any immediate notification when
network
> interfaces are disconnected.
>
> It becomes very problematic when we have no more connected network
> interfaces left, and we don't get any notification for that.
>
> We do get few notifications after about a minute, which is far too
late
> for us.
>
> We also don't get any other indication that anything is wrong.
>
> It doesn't seem as if the SCTP implementation notices that anything is
> wrong at all.
>
> When the network interfaces are back, the system is not recovering.
>
> Is there a way we can get any immediate notification (or any other
> indication, such as an exception) when network interfaces are
disconnected?
>
> Thanks, Michal.
>
> *Michal Lefler*
>
> *Chief Architect* / *Traffix Systems*
>
> *T* +972(0)9-788-9237, *M* +972(0)52-534-0905
>
> *Skype *michal.lefler
>
> *We invite you to join the** LinkedIn Diameter discussion group
> <http://www.linkedin.com/groups?mostPopular=&gid=1787697>***
>
> www.traffixsystems.com <http://www.traffixsystems.com/>
>
> Description: Description: Description: sig 1
> <http://www.traffixsystems.com/>
>