List Info

Thread: RE: BoF request: STUN Control Usage




RE: BoF request: STUN Control Usage
country flaguser name
United States
2007-06-08 15:47:58
> Any idea how many of the deployed SIP servers use UDP?


There is some information from SIPit 20,
https://www.sip
it.net/SIPit20_Summary but of course that
lacks market share information.

> In my 
> understanding, more and more implementations were
switching 
> to TCP because SIP messages were hitting up against the
1500 
> byte MTU limit for UDP. 

Based on that SIPit information, 71% claimed to work fine
with fragmented UDP packets and 10% didn't know.  I guess
the
remainder (19%) know they break with fragmented UDP.  I
dunno
what those implementations plan to do -- maybe choose a 
better IP stack?

Additionally, under high traffic rates from a single IP 
address (such as a bunch of SIP clients behind a single
NAPT), 
draft-heffner-frag-harmful describes problems with
reassembly
which I doubt are sufficiently appreciated by those running
SIP over UDP.

> Maybe the wireless guys are using UDP?  I never really
considered
> using the UDP implementation myself just because I
didn't see the
> point.  I guess the point would be better performance,
but the added
> complication doesn't seem worth it.
> 
> I've even heard talk of excluding UDP altogether if we
were 
> to write SIP over again.  I bring this up because the
double 
> CRLF keep alive for TCP with SIP outbound is a really
trivial 
> addition.  If almost everyone's using TCP, then your 
> reasoning above for STUN control breaks down. 

To summarize, I believe you're saying if UDP keepalives
create 
too much load on your server or network, just use TCP.

-d


_______________________________________________
Behave mailing list
Behaveietf.org
https:/
/www1.ietf.org/mailman/listinfo/behave

Re: BoF request: STUN Control Usage
user name
2007-06-08 16:03:55


On 6/8/07, Dan Wing < dwingcisco.com">dwingcisco.com> wrote:
> Any idea how many of the deployed SIP servers use UDP?

There is some information from SIPit 20,
https://www.sipit.net/SIPit20_Summary but of course that
lacks market share information.

Interesting. ; So it looks like most implementations support both. ; I wonder how many of those actually use the UDP implementation over the TCP in terms of what clients actually connect with. ; I know I wouldn';t, but I wonder.


Additionally, under high traffic rates from a single IP
address (such as a bunch of SIP clients behind a single NAPT),
draft-heffner-frag-harmful describes problems with reassembly
which I doubt are sufficiently appreciated by those running
SIP over UDP.

I would think there would be many issues along these lines.&nbsp; It has to be one of the thorniest parts of any implementation in any case.


To summarize, I believe you're saying if UDP keepalives create
too much load on your server or network, just use TCP.

Basically.&nbsp; Put a slightly different way, if no one's using UDP in practice anyway, then the TCP keep alive mechanism doesn't seem too onerous.&nbsp; There unfortunately seems to be little way to get a solid answer to that question, though, so I'm not sure where to go with it.

Thanks.

-Adam


Re: BoF request: STUN Control Usage
country flaguser name
Netherlands
2007-06-11 16:07:11
On 8-jun-2007, at 22:47, Dan Wing wrote:

> Based on that SIPit information, 71% claimed to work
fine
> with fragmented UDP packets and 10% didn't know.  I
guess the
> remainder (19%) know they break with fragmented UDP.  I
dunno
> what those implementations plan to do -- maybe choose
a
> better IP stack?

<delurk>

A problem with fragmentation is that the port numbers are
only in the  
first fragment so fragmented packets often have trouble
getting by  
firewalls.

</delurk>


_______________________________________________
Behave mailing list
Behaveietf.org
https:/
/www1.ietf.org/mailman/listinfo/behave

[1-3]

about | contact  Other archives ( Real Estate discussion Medical topics )