[Sis-dtn] [EXT] Re: BP Sockets [was RE: IP networking within DTN-based protocol architectures]

Tue Apr 23 23:52:57 UTC 2024

Hi, Marc.  A lot of the motivation for a “BP Sockets” API is that it could potentially minimize the cost of modifying IP applications that are already based on POSIX sockets I/O to use BP instead – you’d just modify a little bit of socket code and presto! your application now sends/receives bundles.  Is the TAPS API lexically close enough to the POSIX sockets API to preserve this advantage?  Or would the “AF_BUNDLE implementation delegating actual behavior to a userspace BPA implementation” that Brian references be a better way to accomplish that?

Scott

From: SIS-DTN <sis-dtn-bounces at mailman.ccsds.org> On Behalf Of Marc Blanchet via SIS-DTN
Sent: Thursday, March 21, 2024 12:45 AM
To: Brian J. Sipos <Brian.Sipos at jhuapl.edu>
Cc: Dr. Keith L Scott via SIS-DTN <sis-dtn at mailman.ccsds.org>
Subject: Re: [Sis-dtn] [EXT] Re: BP Sockets [was RE: IP networking within DTN-based protocol architectures]

Le 21 mars 2024 à 17:15, Sipos, Brian J. via SIS-DTN <sis-dtn at mailman.ccsds.org <mailto:sis-dtn at mailman.ccsds.org> > a écrit :

All,

I meant to post this earlier to the mailing list instead of individual people.

Another thought related to kernel vs. userspace implementations of protocols are the pattern described in [3], specifically the design in Figure 3 where a kernel module implements the socket API for a new address family but delegates all behavior to a userspace daemon. This is akin to an AF_BUNDLE implementation delegating actual behavior to a userspace BPA implementation (as others have mentioned).

Finally, there is also an IETF Transport Services (TAPS) 

I said exactly that on this mailing list a while ago. One looking at defining a new Transport API should really look into TAPS. TAPS is a generic API for any transport: tcp, quic, udp, sctp, … It has generic transport semantics and a modern approach (all asynchronous).

effort to create a socket API [4] that has similar logic and behavior to POSIX sockets but presumably not directly tied to kernel implementation so easier to adapt to new AFs and protocols. It looks preliminary, but could be an interesting avenue for prototyping these ideas without getting into kernel module stuff.

It is more than preliminary. A well-known implementation is Apple, in production, in iOS and MacOSX: i.e. in your hands. I developed networking apps using it, and it is very modern. See: https://developer.apple.com/documentation/network . TAPS is an IETF working group that has almost completed its work. See: https://datatracker.ietf.org/wg/taps/documents/ (most documents are in or near IESG queue for publication). 

Regards, Marc.

[3] https://www.usenix.org/system/files/conference/usenixsecurity18/sec18-o_neill.pdf

[4] https://www.ietf.org/archive/id/draft-dreibholz-taps-neat-socketapi-14.html

From: Sipos, Brian J. 
Sent: Wednesday, March 20, 2024 1:09 AM
To: 'sburleig.sb at gmail.com <mailto:sburleig.sb at gmail.com> ' <sburleig.sb at gmail.com <mailto:sburleig.sb at gmail.com> >; Jeremy.Mayer at dlr.de <mailto:Jeremy.Mayer at dlr.de> ; Felix.Flentge at esa.int <mailto:Felix.Flentge at esa.int> ; vint at google.com <mailto:vint at google.com> 
Subject: RE: [EXT] Re: [Sis-dtn] BP Sockets [was RE: IP networking within DTN-based protocol architectures]

All,

I agree that the POSIX socket interface provides all of the needed mechanisms do the job of datagram sending/receiving. One thing that I think is an important aspect of BP agent--application interface is the ability for the application to influence sourced metadata and have visibility into delivered metadata (e.g. in addition to the primary block EIDs, things like the Lifetime value or Hop Count values). Since there is a similar need in IP land there are mechanisms like IP_PKTINFO and “control data” like IP_TOS / IP_RECVTOS or IP_TTL / IP_RECVTTL with recvmsg [1].

All of that is when used with SOCK_DGRAM -type sockets sending/receiving ADUs as the msg data. A separate interface with SOCK_RAW -type sockets could have the entire encoded bundle present as the msg data. Similar to using (AF_INET, SOCK_RAW) sockets this allows full visibility into header contents but puts a quite large burden on the application to properly handle and process that data.

On the other hand, implementations of recent transports like QUIC have specifically opted out of in-kernel processing with an associated POSIX socket API. Just something to consider from another recent transport protocol development. Putting implementation into kernel space is quite a large hurdle to clear. But also nothing says you cannot implement a ‘prototype’ userspace API modeled after the sockets API like `bp_socket()` `bp_sendmsg()`, etc.

[1] https://man7.org/linux/man-pages/man7/ip.7.html

[2] https://stackoverflow.com/questions/16165542/is-it-possible-to-read-the-ttl-ip-header-field-when-receiving-udp-packets

From: SIS-DTN <sis-dtn-bounces at mailman.ccsds.org <mailto:sis-dtn-bounces at mailman.ccsds.org> > On Behalf Of sburleig.sb--- via SIS-DTN
Sent: Monday, March 11, 2024 10:22 AM
To: Jeremy.Mayer at dlr.de <mailto:Jeremy.Mayer at dlr.de> ; Felix.Flentge at esa.int <mailto:Felix.Flentge at esa.int> ; vint at google.com <mailto:vint at google.com> 
Cc: sis-dtn at mailman.ccsds.org <mailto:sis-dtn at mailman.ccsds.org> 
Subject: [EXT] Re: [Sis-dtn] BP Sockets [was RE: IP networking within DTN-based protocol architectures]

APL external email warning: Verify sender sis-dtn-bounces at mailman.ccsds.org <mailto:sis-dtn-bounces at mailman.ccsds.org> before clicking links or attachments

Right, the service interface in the CCSDS BP specification -- like all service specifications -- is abstract; it defines the properties of a BP API for the purpose of explaining the required behavior of a Bundle Protocol Agent.  Any number of existing and future BP implementations with very different APIs can be wholly conformant to that specification.  But a BP sockets specification would enable any application that currently communicates over POSIX (BSD) sockets to use BP instead of UDP without extensive code changes; the syntax is already in place.  That’s the compelling advantage of the idea.

Scott

From: Jeremy.Mayer at dlr.de <mailto:Jeremy.Mayer at dlr.de>  <Jeremy.Mayer at dlr.de <mailto:Jeremy.Mayer at dlr.de> > 
Sent: Monday, March 11, 2024 3:48 AM
To: Felix.Flentge at esa.int <mailto:Felix.Flentge at esa.int> ; sburleig.sb at gmail.com <mailto:sburleig.sb at gmail.com> ; vint at google.com <mailto:vint at google.com> 
Cc: sis-dtn at mailman.ccsds.org <mailto:sis-dtn at mailman.ccsds.org> 
Subject: RE: BP Sockets [was RE: [Sis-dtn] IP networking within DTN-based protocol architectures]

Hi,

In the abstract, I’d say so, but (let’s be honest) many developers tend to ignore the abstract notion of the service interface and prefer to have a tangible set of API operations.

Thanks,

Jeremy

From: Felix Flentge <Felix.Flentge at esa.int <mailto:Felix.Flentge at esa.int> > 
Sent: Montag, 11. März 2024 11:46
To: sburleig.sb at gmail.com <mailto:sburleig.sb at gmail.com> ; Mayer, Jeremy <Jeremy.Mayer at dlr.de <mailto:Jeremy.Mayer at dlr.de> >; vint at google.com <mailto:vint at google.com> 
Cc: sis-dtn at mailman.ccsds.org <mailto:sis-dtn at mailman.ccsds.org> 
Subject: BP Sockets [was RE: [Sis-dtn] IP networking within DTN-based protocol architectures]

Hi,

I like the idea but isn’t this already covered (partially) by the CCSSDS BP service interface, eg:

*        Bind – register EID in passive state

*        Listen – register EID in active state

*        Connect – sets a default destination for future send.request

*        Send – send.request 

*        Recv – poll.request

Regards,

Felix

From: SIS-DTN <sis-dtn-bounces at mailman.ccsds.org <mailto:sis-dtn-bounces at mailman.ccsds.org> > On Behalf Of sburleig.sb--- via SIS-DTN
Sent: Saturday, March 9, 2024 6:56 PM
To: Jeremy.Mayer at dlr.de <mailto:Jeremy.Mayer at dlr.de> ; vint at google.com <mailto:vint at google.com> 
Cc: sis-dtn at mailman.ccsds.org <mailto:sis-dtn at mailman.ccsds.org> 
Subject: Re: [Sis-dtn] IP networking within DTN-based protocol architectures

Thanks for that analysis, Jeremy.  We’ve been talking about at least since STINT several years ago, and I really think it’s an idea whose time has come.

Scott

From: Jeremy.Mayer at dlr.de <mailto:Jeremy.Mayer at dlr.de>  <Jeremy.Mayer at dlr.de <mailto:Jeremy.Mayer at dlr.de> > 
Sent: Saturday, March 9, 2024 8:01 AM
To: vint at google.com <mailto:vint at google.com> ; sburleig.sb at gmail.com <mailto:sburleig.sb at gmail.com> 
Cc: Tomaso.deCola at dlr.de <mailto:Tomaso.deCola at dlr.de> ; sis-dtn at mailman.ccsds.org <mailto:sis-dtn at mailman.ccsds.org> 
Subject: RE: [Sis-dtn] IP networking within DTN-based protocol architectures

Hi,

That’s a fair point. Generally, I’d say that BP-POSIX could work with the following:

*        Bind – works: sets source EID.

*        Listen – works; sets source EID and opens reception. 

*        Connect – UDP style, sets default destination address.

*        Send – only valid for connected bundlesocks. 

*        Sendto/sendmsg – works.

*        Recv – works.

*        Recvfrom/recvmsg – works, probably should support “wildcard” EIDs, at least for services. 

*        Getaddrinfo and friends – not required

*        Select – Works, highly implementation dependent.

BSD sockets really do work for everything!

Thanks,

Jeremy

From: Vint Cerf <vint at google.com <mailto:vint at google.com> > 
Sent: Samstag, 9. März 2024 16:51
To: Scott Burleigh <sburleig.sb at gmail.com <mailto:sburleig.sb at gmail.com> >
Cc: Mayer, Jeremy <Jeremy.Mayer at dlr.de <mailto:Jeremy.Mayer at dlr.de> >; de Cola, Tomaso <Tomaso.deCola at dlr.de <mailto:Tomaso.deCola at dlr.de> >; Dr. Keith L Scott via SIS-DTN <sis-dtn at mailman.ccsds.org <mailto:sis-dtn at mailman.ccsds.org> >
Subject: Re: [Sis-dtn] IP networking within DTN-based protocol architectures

Yes generally speaking

On Sat, Mar 9, 2024, 10:43 < <mailto:sburleig.sb at gmail.com> sburleig.sb at gmail.com> wrote:

I think we get closer to BSD/POSIX if we use UDP as a model.  The attenuated UDP notion of connect() would work for BP, I think.

Scott

From: SIS-DTN <sis-dtn-bounces at mailman.ccsds.org <mailto:sis-dtn-bounces at mailman.ccsds.org> > On Behalf Of Jeremy Mayer via SIS-DTN
Sent: Saturday, March 9, 2024 12:35 AM
To: vint at google.com <mailto:vint at google.com> ; Tomaso.deCola at dlr.de <mailto:Tomaso.deCola at dlr.de> 
Cc: sis-dtn at mailman.ccsds.org <mailto:sis-dtn at mailman.ccsds.org> 
Subject: Re: [Sis-dtn] IP networking within DTN-based protocol architectures

Hi,

I’m tempted to agree. I’ve seen limited utility in IP->BP encap, as (other than some unidirectional UDP-based applications), the assumptions made in IP networks (especially TCP) do not scale well to DTN environments. The two largest issues I see are:

A.     The assumption that relatively low-latency and bidirectional communication is available.

B.     The layering of stream and message-based semantics.

Almost every TCP-ish application which I’ve seen run over BP winds up with exceptionally low performance, unless some sort of PEP is integrated, which adds complexity, failure points, and fragility. 

+1 for the notion of a standard API though, with the additional caveat that we should standardize nomenclature across implementations. We could likely reuse about 80% of the BSD/POSIX interface for BP, with the notable exception of connect.

Thanks,

Jeremy

From: SIS-DTN <sis-dtn-bounces at mailman.ccsds.org <mailto:sis-dtn-bounces at mailman.ccsds.org> > On Behalf Of Vint Cerf via SIS-DTN
Sent: Freitag, 8. März 2024 12:31
To: de Cola, Tomaso <Tomaso.deCola at dlr.de <mailto:Tomaso.deCola at dlr.de> >
Cc: sis-dtn at mailman.ccsds.org <mailto:sis-dtn at mailman.ccsds.org> 
Subject: Re: [Sis-dtn] IP networking within DTN-based protocol architectures

The more I think about this, the less I like trying to manage off-Earth IP networks remotely. I think I would much rather see native applications running over BP. In any case we need to develop standard APIs for BP, such as the Sockets interface for Linux/Unix/Posix. 

We do need to figure out how IP-based and BP-based applications interwork. What do Domain Names mean in the BP context for example. 

v

On Fri, Mar 8, 2024 at 4:24 AM Tomaso de Cola via SIS-DTN < <mailto:sis-dtn at mailman.ccsds.org> sis-dtn at mailman.ccsds.org> wrote:

Dear All,

as mentioned yesterday during our DTN weekly, during the past Fall meeting a presentation was given by University of Nanjing with the support of Chinese agency about IP-networking within DTN-based protocol stacks, especially for planetary network segments (i.e. what is often referred to as IP-islands). 

The link to that presentation is as follows:

https://cwe.ccsds.org/sis/docs/SIS-DTN/Meeting%20Materials/2023/Fall%202023%20SIS-DTN/Possible%20BOF%20Discussion.pptx?d=w44d31f0b3be649e4bf24ec4ae50479a1

The proposal made there was for a BOF (i.e. towards a WG formation), which I see however too premature. On the contrary, I’d be more in favour for a SIG (Special Interest Group), happening outside the DTN WG.

As discussed yesterday, key points to clarify/agree are:

1.      Is the proposal sound and meaningful to be addressed?

2.      Is appropriate for the SIS area or should be better addressed at architectural level in SEA (maybe it can be even a cross-area initiative)?

3.      In the case of SIS, is there significant support to start this? In other words, which agencies have interest in contributing to this activity?

Thank you for sharing comments/objections/suggestions.

Have a nice weekend,

Tomaso

————————————————————————

Deutsches Zentrum für Luft- und Raumfahrt (DLR)
German Aerospace Center
Institute of Communications and Navigation | Satellite Networks | Oberpfaffenhofen | 82234 Wessling | Germany

Tomaso de Cola, Ph.D. | Integrated Satellite Systems (INS) Team Leader
Telefon  <tel:+49%208153%20282156> +49 8153 28-2156 | Telefax   <tel:+49%208153%20282844> +49 8153 28-2844 |  <mailto:sandro.scalise at dlr.de> tomaso.decola at dlr.de
 <http://www.dlr.de/kn/institut/abteilungen/san> http://www.dlr.de/kn/institut/abteilungen/san

_______________________________________________
SIS-DTN mailing list
 <mailto:SIS-DTN at mailman.ccsds.org> SIS-DTN at mailman.ccsds.org
 <https://mailman.ccsds.org/cgi-bin/mailman/listinfo/sis-dtn> https://mailman.ccsds.org/cgi-bin/mailman/listinfo/sis-dtn

-- 

Please send any postal/overnight deliveries to:

Vint Cerf

Google, LLC

1900 Reston Metro Plaza, 16th Floor

Reston, VA 20190

+1 (571) 213 1346

until further notice

This message is intended only for the recipient(s) named above. It may contain proprietary information and/or protected content. Any unauthorised disclosure, use, retention or dissemination is prohibited. If you have received this e-mail in error, please notify the sender immediately. ESA applies appropriate organisational measures to protect personal data, in case of data privacy queries, please contact the ESA Data Protection Officer ( <mailto:dpo at esa.int> dpo at esa.int). 

_______________________________________________
SIS-DTN mailing list
SIS-DTN at mailman.ccsds.org <mailto:SIS-DTN at mailman.ccsds.org> 
https://mailman.ccsds.org/cgi-bin/mailman/listinfo/sis-dtn

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.ccsds.org/pipermail/sis-dtn/attachments/20240423/09691c77/attachment-0001.htm>