Re: [B.A.T.M.A.N.] [RFC] alfred: implement TCP support for server-to-server communication

[Simon Wunderlich <sw@simonwunderlich.de>]

[-- Attachment #1: Type: text/plain, Size: 1541 bytes --]

Hello Hans-Werner,

On Monday 21 March 2016 10:38:55 Hans-Werner Hilse wrote:
> Hi,
> 
> Am 2016-03-21 10:13, schrieb Sven Eckelmann:
> >> +int connect_tcp(struct interface *interface, const struct in6_addr
> >> *dest)
> >> +{
> >> [...]
> >> +}
> >> +
> > 
> > Wouldn't this hang for a while and make the alfred server
> > "unresponsive"
> > when the remote is not reachable at this moment?.
> 
> Yes, it most definitely would, you're right.
> 
> If the route I'm following is generally right, I think I'll work on this
> part (*that* I can make into a different commit). The connect() can be
> made non-blocking, and I would add the socket and the request to send to
> a list (possibly using the transaction list that already exists right
> now).

I think you are on the right track - in the ticket, we discussed whether we do 
TCP or just decrease the fragment size, but doing TCP is probably the more 
reliable option to do.

I agree that we should definitely use nonblocking operation.

I was also thinking that we might want to use TCP in general when the packet 
size reaches a certain threshold (e.g. >3kB), and always use UDP for small 
packets. Then we would also not need the commandline option (unless you think 
we should leave a way to force a mode). Any thoughts on that?

I don't think you have to split your patch further, at least I don't see a 
good way right now. I didn't do a deep review yet, but can do in the next 
iteration with the nonblocking stuff implemented. :)

Thanks a lot for taking care of this!
     Simon

[-- Attachment #2: This is a digitally signed message part. --]
[-- Type: application/pgp-signature, Size: 819 bytes --]