RDS: no freely available protocol specification?

RDS: no freely available protocol specification?

[Juliusz Chroboczek]

Dear all,

I am a little worried about the Reliable Datagram Sockets (RDS), which are
currently in the process of being merged into the kernel.  I've checked the
latest Interned-Drafts index, I've tried searching the web, and I cannot
find a freely available protocol specification.  (I've been unable to check
the documentation section of the oss.oracle.com site, which requires
a password with no way to register visible to my tired eyes.)

I am wondering whether we wish to encourage the use of proprietary
(undocumented) protocols, and I wish that the kernel crowd should really
try to put some pressure on the Oracle guys to get the RDS protocol spec
published as an RFC, or at the very least submitted as an Internet Draft,
before it is released as part of the Linux kernel.

                                        Juliusz Chroboczek

Re: RDS: no freely available protocol specification?

[Andy Grover]

Juliusz Chroboczek wrote:
> I am a little worried about the Reliable Datagram Sockets (RDS), which are
> currently in the process of being merged into the kernel.  I've checked the
> latest Interned-Drafts index, I've tried searching the web, and I cannot
> find a freely available protocol specification.  (I've been unable to check
> the documentation section of the oss.oracle.com site, which requires
> a password with no way to register visible to my tired eyes.)

Look in net-next tree, Documentation/networking/rds.txt.

> I am wondering whether we wish to encourage the use of proprietary
> (undocumented) protocols, and I wish that the kernel crowd should really
> try to put some pressure on the Oracle guys to get the RDS protocol spec
> published as an RFC, or at the very least submitted as an Internet Draft,
> before it is released as part of the Linux kernel.

I thought "rough consensus and running code" was the IETF credo? We've
provided running code, some documentation, and an active maintainer and
mailing list. If you're interested in getting an RFC drafted for RDS,
then I invite you to start work on one. I also will accept patches to
improve the documentation.

See how this works?

Regards -- Andy

Re: RDS: no freely available protocol specification?

[Juliusz Chroboczek]

Thanks for your reply, Andy.

>> I've tried searching the web, and I cannot find a freely available
>> protocol specification.

> Look in net-next tree, Documentation/networking/rds.txt.

I cannot find a protocol description, let alone a full specification, in
this document.

>> I am wondering whether we wish to encourage the use of proprietary
>> (undocumented) protocols,

> I thought "rough consensus and running code" was the IETF credo?

I'm not the IETF, but it is my understanding that rough consensus is
achieved by writing protocol specifications, and submitting them as
Internet-Drafts.  (While some Internet-Drafts are destined to become
standards track RFCs after IETF consensus has been reached, many are only
submitted in order to stimulate discussion, and some only get submitted so
that they may become Informational RFCs.)

>From what I can gather from the very sparse documentation, RDS in its
current state cannot possibly achieve IETF consensus, for multiple reasons
(many of which I happen to disagree with -- but I'm not the IETF).  Still,
if Oracle find it useful, I'd like to see it published as an informational
RFC.

I'd have no objection to RDS being included in the Linux kernel without
IETF consensus, as long as the protocol is fully documented.

> If you're interested in getting an RFC drafted for RDS, then I invite you
> to start work on one. I also will accept patches to improve the
> documentation.
>
> See how this works?

Andy, please don't bully me.

                                        Juliusz

Re: RDS: no freely available protocol specification?

[David Miller]

From: Juliusz Chroboczek <Juliusz.Chroboczek@pps.jussieu.fr>
Date: Tue, 10 Mar 2009 23:07:45 +0100

> I am wondering whether we wish to encourage the use of proprietary
> (undocumented) protocols, and I wish that the kernel crowd should really
> try to put some pressure on the Oracle guys to get the RDS protocol spec
> published as an RFC, or at the very least submitted as an Internet Draft,
> before it is released as part of the Linux kernel.

There is nothing proprietary about a protocol for which you have
the source code to a full implementation.

And having an RFC or even a draft for a protocol is not a prerequesite
for it being included in the Linux kernel.

So put your bad-juju back in your pocket, kthx.