[sctp-dev] Performance degradation when changing from primary to backup path
Bruno Sousa
bmsousa at dei.uc.pt
Mon May 6 06:44:11 PDT 2013
Dear SCTP users and developers
I'm Bruno Sousa and I'm working with SCTP due to its multihoming support.
Currently I've an application for file transfer using SCTP and JDK7.
I've two linux servers multihomed where each one has two ethernet
interfaces.
File transfer starts with primary path in /eth0/ interface, and after a
certain period I've programmed failures using iptables, to drop packet
on server B coming from server A (srv_A):
/iptables -I INPUT -i eth0 -s IP_eth0_srv_A -j DROP //
//iptables -A OUTPUT -d IP_eth0_srv_A -j DROP /
This works and SCTP detects failures on primary path in eth0, as expected.
I've implement handlers to receive and treat SCTP events, namely:
/*PeerAddressChangeNotification.AddressChangeEvent.ADDR_UNREACHABLE*/
With which I'm performing the change of Primary addresses, as follows:
/try{//
// auxChannel.setOption(SctpStandardSocketOptions.SCTP_PRIMARY_ADDR,
remote.toInetSocketAddr());//
//
////Change remote primary address//
//auxChannel.setOption(SctpStandardSocketOptions.SCTP_SET_PEER_PRIMARY_ADDR,
auLocal );//
//}catch(Exception ex){ //
// // Treat Exception//
//}/
The code works for changing primary addresses, as I've checked in
//proc/net/sctp/assocs/.
Nonetheless, the problem, is that after changing address, file transfer
get very slow on backup path.
Using Wireshark I've checked that SCTP DATA messages and respective ACKs
are not sent 'sequentially' as with primary path.
SCTP Data is sent and only after a certain time it is sent again.
Is there any reason for such performance degradation?
What can I do, to improve performance?
Thanks in advance,
Bruno Sousa
My Kernel is: 2.6.32-5 and I've also tried on 3.2.0
Lksctp-tools: 1.0.11+dfsg-1 and 1.0.11+dfsg-2
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mail.openjdk.java.net/pipermail/sctp-dev/attachments/20130506/97450802/attachment.html
More information about the sctp-dev
mailing list