Re: [DMM] WGLC #1 for draft-ietf-dmm-ondemand-mobility-01

[Dave Dolson <[e-mail]>]

>From an application developer's point of view, I don't think there should be a distinction about roaming between providers or with the same provider.
It should work with WiFi, mobile, even wired Ethernet.
E.g., as I unplug my laptop from work, use the 3G stick on the train, WiFi in the coffee shop and plug in wired Ethernet at home. I could have a fixed address as well as several temporary (no guarantee) addresses.

I'm not saying all of those access technologies need to have implementation of all types of addresses.
One might only get a fixed address from one of those providers, all the others being no-guarantee.

-Dave

________________________________________
From: Moses, Danny [[e-mail]]
Sent: Sunday, February 21, 2016 10:05 AM
To: Dave Dolson; [e-mail]
Subject: RE: [DMM] WGLC #1 for draft-ietf-dmm-ondemand-mobility-01

Hi Dave,

Regarding the term "Nomadic" IP address:
Actually, this draft is about enhancements to the Socket API that are useful in relation with movement of the mobile host. Its intent was to notify the access network as to the type of IP session continuity it requires upon movement between LANs. The idea is to reduce the network support for session continuity (via PMIP, GTP, etc) when it is not needed. This draft is not dealing with devices migrating from one service provider to another as other issues should be handled in such scenario.

"Local" is not good as it clashes with IPv6 Local addresses.

The best term in my mind would be: "An IP address without any NW guarantee to continue to be valid after a potential host movement to a new LAN with a different IP prefix". So far, we could not come with a good name for such an address. May be "Guarantee-less"?

Let's not forget: this is going to be used by application developers who do not necessarily fully understand how networks allocate IP addresses and maintain their validity.

Regarding when On-Demand resolution occurs (Point 6):
How about if we say:

If applications want to influence the type of IP address their generated traffic will use, they must do so after creating a Socket and prior to generating the first transmitted packet.

We do not want to be too specific since it is not a user's manual.

Does this sound reasonable to you?

Thanks,
        /Danny

-----Original Message-----
From: Dave Dolson [mailto:[e-mail]]
Sent: Thursday, February 18, 2016 22:46
To: Moses, Danny; [e-mail]
Subject: RE: [DMM] WGLC #1 for draft-ietf-dmm-ondemand-mobility-01

Danny and Alper,
Thanks. I hope your helpful explanations below make it into the draft as well.


On the "nomadic address" question, this name really feels wrong to me. The address doesn't move and hence isn't nomadic.
You don't like "ephemeral" because it doesn't convey movement.
But a question, does the host have to move for the principles of this document to apply?
Couldn't a host get a new address (perhaps from a different wireless provider) without moving?
Maybe "non-nomadic address" or "provider-local address" ?


On point 6, yes I think you have to specify when the resolution occurs, since it affects which functions return error codes.
I don't think the local address can be resolved before connect(), since the choice of local address may depend on the remote address to connect. E.g., connections to a peer on a local LAN subnet need to use an address on that interface.


-Dave




-----Original Message-----
From: Moses, Danny [mailto:[e-mail]]
Sent: Thursday, February 18, 2016 9:13 AM
To: Dave Dolson; [e-mail]
Subject: RE: [DMM] WGLC #1 for draft-ietf-dmm-ondemand-mobility-01

Hi Dave,

Sorry for the very late response. Somehow we missed the original email and were reminded by Jouni (Thanks Jouni).

Anyway, thank you for the thorough and helpful comments. We have produced a new version and will publish it shortly.

Please see further details to your comments:

Dave>
1. Was the term "Nomadic" discussed? To me, a nomadic thing moves around, but in this draft, Nomadic IP addresses do not move around; they are replaced. "Ephemeral IP Address" might convey the idea more clearly.

Reply>
"Nomadic" was discussed, it is not the best name but we could not find a better one.
What 'moves around' is the mobile host and as a result of that movement, its source IP address might become obsolete (when it moves from its original LAN to another with a different prefix). An application on a mobile host requires a 'Nomadic' IP address when it does not care for the IP continuity guarantee services provided by the network. This is either because it knows that the mobile host is not really mobile, or because it has other means of maintaining session continuity and does not want the overhead associated with network-provided IP continuity services.

The term 'Ephemeral' is not associated with movement - we think it is better to have a 'movement'-related name.

Dave>
2. I know this is really picky, but it wouldn't hurt to spell out "REQUIRE" in IPV6_REQ_FIXED_IP etc.
 - IPV6_REQUIRE_FIXED_IP in parallel with IPV6_PREFER_SRC_PUBLIC

Reply>
Make sense. Will be changed in the next version.

Dave>
3. There is a "TDB" in section 3.4. "TBD: Disallow this case?"  This needs to be resolved. I suggest the most restrictive flag applies.

Reply>
This is about whether to enable the application to set more than one flag or not per a given socket. We left that for discussion in the draft but never resolved it.

We prefer not to enable setting more than one flag since it add complexity. The application can check the returned code and act upon it. For example, if it uses setsockopt() with IPV6_REQUIRE_FIXED_IP and the call returns with an error code, it can re-attempt with IPV6_REQUEST_SUSTAINED_IP or IPV6_REQUEST_NOMADIC_IP (or without the flags - in case they are not supported by the network).

So we are replacing the original text:
More than one of these flags may be set on the same socket. In that case, an IP address compliant with any one of them shall be selected.
TBD: Disallow this case?

With:
Only one flag of these flags may be set on the same socket. If an application attempts to set more than one flag, the most recent setting will be the one in effect.

Dave>
4. Must resolve "Application of this solution to IPv4 is TBD."
 - clearly the socket option is IPV6_something, but...
 - there is uncharted territory surrounding IPv4-mapped-IPv6 addresses (https://tools.ietf.org/html/rfc4291#section-2.5.5.2 ), when socket option IPV6_V6ONLY is false.
 - I think the goal should be that the API *does* apply to IPv4 via IPv4-mapped-IPv6 addresses, while acknowledging that the operating system or network may not be able to fulfill the request.
    - I.e., permit the application to request constraints on IPv4, but also expect the application to try for a relaxed address if the initial request fails.
 - I see no need to support the API with AF_INET sockets, since AF_INET6 can be used with IPV6_ONLY=false and IPv4-mapped-IPv6 addresses.

Reply>
Actually, the charter of DMM addresses IPv6 only. So we are removing the text that refers to IPv4 altogether. If we see in the future a need to address IPv4, we will do so in a separate draft.

Dave>
5. The error codes are not clearly defined. What is the errno for failure of setsockopt() and others?

Reply>
True.

We added the following text at the end of section 3.4:

The following new error codes are also defined in the document and will be used in Socket API in compliance with the [RFC5014]:
EAI_REQUIREIPNOTSUPPORTED /*The network does not support the ability to request that IP address type */ EAI_REQUIREIPFAILED /* The network could not assign the specific IP address type */

Dave>
6. I'm unclear on when the on-demand resolution occurs? I don't think it can be done at the time of setsockopt(); as I understand it, the addresses are resolved later, at connect() or listen(). I'm not sure about this, but it should be explained. If the resolution is done at connect() time, is a new errno required?

Reply>
The addresses should be resolved before connect() or listen() in order to avoid unnecessary delay caused by the need to request the address from the network (with DHCP - for example). Furthermore, UDP should also be supported, hence connect() or listen() may not be used. There for, they must be resolved at setsockopt() prior to the above calls. Do you see any problem with that?

Do you think text should be added to clarify this?

Dave>
7. Some socket interfaces are not mentioned. What about sendto(), sendmsg(), which are not connection-oriented?

Reply>
Both assume that a source IP address was assigned to the host.
If the application invoked setsockopt() successfully prior to attempting to send a message, the appropriate source IP address type will be used for the transmitted packets. If not, whatever source IP address assigned to the host will be used.

Dave>
8. In section 4.1, support for legacy applications does not really say how to support legacy applications.
 - my opinion: the default (lacking new socket options) should be to require Fixed for listen() and Sustained for connect(), and Nomadic/Ephemeral for non-connected datagrams.

Reply>
Legacy application will not use these new flags, and as a result, will not be able to influence the source IP address type and related mobility support. As a result, they will behave exactly as applications behave today.

Clearly there will be some default behavior as to the assigned source IP address. Today, this depends on the access networks. Cellular networks usually provide Fixed or Sustained source IP addresses, and WiFi networks usually provide Nomadic source IP addresses.

We do not think we should specify a strict behavior in this case, but rather leave it to be implementation-specific.

Once again, thanks for the good comments.

Alper and Danny

-----Original Message-----
From: dmm [mailto:[e-mail]] On Behalf Of Dave Dolson
Sent: Wednesday, December 02, 2015 18:42
To: Jouni Korhonen; [e-mail]; Dapeng Liu
Subject: Re: [DMM] WGLC #1 for draft-ietf-dmm-ondemand-mobility-01

(I haven't paid close attention to the list, so apologies if I'm raising old issues.)

1. Was the term "Nomadic" discussed? To me, a nomadic thing moves around, but in this draft, Nomadic IP addresses do not move around; they are replaced. "Ephemeral IP Address" might convey the idea more clearly.

2. I know this is really picky, but it wouldn't hurt to spell out "REQUIRE" in IPV6_REQ_FIXED_IP etc.
 - IPV6_REQUIRE_FIXED_IP in parallel with IPV6_PREFER_SRC_PUBLIC

3. There is a "TDB" in section 3.4. "TBD: Disallow this case?"  This needs to be resolved. I suggest the most restrictive flag applies.

4. Must resolve "Application of this solution to IPv4 is TBD."
 - clearly the socket option is IPV6_something, but...
 - there is uncharted territory surrounding IPv4-mapped-IPv6 addresses (https://tools.ietf.org/html/rfc4291#section-2.5.5.2 ), when socket option IPV6_V6ONLY is false.
 - I think the goal should be that the API *does* apply to IPv4 via IPv4-mapped-IPv6 addresses, while acknowledging that the operating system or network may not be able to fulfill the request.
    - I.e., permit the application to request constraints on IPv4, but also expect the application to try for a relaxed address if the initial request fails.
 - I see no need to support the API with AF_INET sockets, since AF_INET6 can be used with IPV6_ONLY=false and IPv4-mapped-IPv6 addresses.

5. The error codes are not clearly defined. What is the errno for failure of setsockopt() and others?

6. I'm unclear on when the on-demand resolution occurs? I don't think it can be done at the time of setsockopt(); as I understand it, the addresses are resolved later, at connect() or listen(). I'm not sure about this, but it should be explained. If the resolution is done at connect() time, is a new errno required?

7. Some socket interfaces are not mentioned. What about sendto(), sendmsg(), which are not connection-oriented?

8. In section 4.1, support for legacy applications does not really say how to support legacy applications.
 - my opinion: the default (lacking new socket options) should be to require Fixed for listen() and Sustained for connect(), and Nomadic/Ephemeral for non-connected datagrams.



-Dave




-----Original Message-----
From: dmm [mailto:[e-mail]] On Behalf Of Jouni Korhonen
Sent: Tuesday, December 01, 2015 1:07 PM
To: [e-mail]; Jouni; Dapeng Liu
Subject: [DMM] WGLC #1 for draft-ietf-dmm-ondemand-mobility-01

Folks,

This mail starts two week WGLC for the I-D:
        https://tools.ietf.org/html/draft-ietf-dmm-ondemand-mobility-01

The WGLC ends 12/15/2015.

Provide your reviews and comments to the mailing list. For the better tracking of issues and proposed changed use the Issue Tracker to submit your issues/proposals.

- Jouni & Dapeng

_______________________________________________
dmm mailing list
[e-mail]
https://www.ietf.org/mailman/listinfo/dmm

_______________________________________________
dmm mailing list
[e-mail]
https://www.ietf.org/mailman/listinfo/dmm
---------------------------------------------------------------------
A member of the Intel Corporation group of companies

This e-mail and any attachments may contain confidential material for the sole use of the intended recipient(s). Any review or distribution by others is strictly prohibited. If you are not the intended recipient, please contact the sender and delete all copies.

---------------------------------------------------------------------
A member of the Intel Corporation group of companies

This e-mail and any attachments may contain confidential material for
the sole use of the intended recipient(s). Any review or distribution
by others is strictly prohibited. If you are not the intended
recipient, please contact the sender and delete all copies.