[hipl-users] Re: some recent bug fixes

  • From: Miika Komu <miika@xxxxxx>
  • To: hipl-users@xxxxxxxxxxxxx
  • Date: Fri, 2 Feb 2007 21:23:52 +0200 (EET)

On Fri, 2 Feb 2007, Tomi Hautakoski wrote:

On Sun, 21 Jan 2007, Miika Komu wrote:
#175: initiating from multiple source HITs when hosts are in diff
-----------------------------------------------------------------

Hopefully I finally got this fixed. I believe this occurred because adding IPsec SPs is a asynchronous operation in the current code. All of the communications with hip daemon are not synchronous (i.e. there is always a return message) and I hope this delays the connection set-up enough for the SPs to be completed.

This bug does not occur on my equipment anymore. Tomi, please try pingtest.sh on your machine and report if it works for you also? Thx.

First, sorry for the delay(other deadlines). I've done testing with hipl-189(100-200 BEX's, both WLAN and Ethernet) and this fix looks very promising when compared to older releases. Only ~5 times hipd got stuck and IIRC all of them were situations where 1 node was doing BEX's to 2-3 other nodes at the same time. Is there a limit on how many BEX's one node can do parallel?

Good to hear! The socket buffers currently queue messages and there daemon is processing them single-threaded. It makes debugging also easier. There has been some discussions in making the daemon multithreaded, but it is quite a big effort and I don't know when this is actually going to happen.

We haven't really done much of stress testing yet. Do you know whether the problems occurred in triggering of base exchange or receiving of control packets? I guess it could be possible to increase netlink and/or input socket buffer size. There is already some #if 0 code in hipd/init.c:hipd_init in case you have time to try it out.

--
Miika Komu                                       http://www.iki.fi/miika/

Other related posts: