LWN: Comments on "Linux gets DCCP"

Compare contrast with SCTP?

mmikailov — Thu, 17 Jul 2008 14:09:16 +0000

"How does DCCP fit into the picture with SCTP?"
- SCTP could be used for instance for relaible transmission of signaling messages; whereas
DCCP - for transmission of bearer traffic.

Compare contrast with SCTP?

shapr — Sat, 21 Jan 2006 12:27:01 +0000

How does DCCP fit into the picture with SCTP?

What ever happened to SCTP?

AnswerGuy — Fri, 28 Oct 2005 16:32:55 +0000

I never saw an article covering the reasoning and intended purpose of
the SCTP (stream control transport protocol) (RFC2960).

Seeing this one on DCCP brought it to mind (since it's transport oriented,
it sounds like it's role to TCP is analogous to DCCP's role compared to
UDP).

How would someone designing or implementing upper level protocols choose among these lower level ones? Which existing applications and protocols might benefit by being fitted with options to use these lower level protocols when they are available? Is there any standard way to negotiate (something along the lines of, try DCCP first then fall back to UDP)?

This is an interesting article, but I want to see more :)

JimD

DCCP and legacy firewalls.

imcdnzl — Sun, 04 Sep 2005 22:53:19 +0000

What I meant by friendly to firewalls is that it is easy to track sessions (much easier than UDP where there are no direct sessions). That is one of the main reasons why media applications fall back to TCP...

Of course you have to allow protocol 33 through as another poster says which on some equipment may cause problems.

Linux gets DCCP

imcdnzl — Sun, 04 Sep 2005 22:51:29 +0000

Yes you are correct here - the ITU is working with the IETF on DCCP at present.

DCCP and legacy firewalls.

psiren — Thu, 01 Sep 2005 21:34:14 +0000

I'm far from an expert, but having had a quick glance through the RFC, section 19 refers to the protocol number for DCCP being 33 (probably). UDP uses 17, so it will be seen as distinct and different from UDP. Take a quick glance at /etc/protocols to see how many there are (more than you probably realise, not that you use many of them day to day).

Many firewalls support protocols other than the common TCP, UDP and ICMP, so theres no specific reason to think you won't be able to pass the data through. However, there may not be direct support for checking anything inside the packets headers, as the software would need to understand the protocol to extract this information.

DCCP and legacy firewalls.

Duncan — Thu, 01 Sep 2005 16:19:11 +0000

What does DCCP look like to a legacy firewall. You mention that it's
designed to work well with firewalls, but how would I implement "allow"
filters on a default-deny policy firewall, that only understands legacy
protocols? Would DCCP look to it like UDP? IOW, is it UDP with
additional protocol info in what would be the UDP payload, thus recognized
as UDP by legacy routers, or ??? If so, are there NAPT/masquerade
implications similar to those with FTP and various VoIP and security
protocols, or not?

I ask as I run one of those legacy things, one of the first-gen consumer
level NAPT based broadband routers. At some point, I'll likely replace it
with a Linux based appliance and therefore benefit from community firmware
projects, but my old Netgear rt314 has and continues to serve me well, so
why mess with a good thing until I need to?

OTOH, it'll probably be another year or more before there's enough out
there using DCCP in working deployments to be worrisome, particularly if
MSWormOS support lags, and by then I may well have upgraded routers, but
there'll still certainly be others who haven't.

Duncan

Linux gets DCCP

gmaxwell — Thu, 01 Sep 2005 04:32:05 +0000

DCCP isn't really meant to replace TCP.
Take a look at TCP westwood, it's already in Linux and it makes a visable improvement overwireless links.

Linux gets DCCP

showell — Thu, 01 Sep 2005 03:52:45 +0000

This looks like a great improvement if we can use it to replace many of the TCP connections over wireless coms systems. One of the big hassles with TCP over wireless is TCPs slow start backoff mechanism. This leads to bandwidth oscillation and pretty horrible throughputs as TCP was never meant to handle the fast variations in the transmission path that radio introduces (ie lots of errors then really good all in a matter of seconds or milli seconds).CCID3 looks like the solution to this.

There has already been unsuccessful attempts to get the IETF to vary TCPs slow start, maybe this time us wireless network operators will be able to implement something that performs much better for our customers.

Spelling screwup

corbet — Thu, 01 Sep 2005 03:33:52 +0000

The spelling mistake is strange; I remember being extra sure that I had it right. Obviously not, my apologies. Fixed now.

Linux gets DCCP

imcdnzl — Thu, 01 Sep 2005 02:14:48 +0000

One other useful thing to note is that UDP doesn't work well behind firewalls often and DCCP is designed to do that.

For those that would like to learn more about DCCP see http://wlug.org.nz/DCCP or see my research proposal at http://wand.net.nz/~iam4

Ian McDonald (not Ian MacDonald as the article says)