Hi Alex,
My replies in this email are marked with 'Danny >>2'.
Regards,
/Danny
-----Original Message-----
From: Alexandre Petrescu [mailto:[email protected]]
Sent: Wednesday, May 18, 2016 16:33
To: Moses, Danny; [email protected]
Subject: Re: [DMM] OnDemand draft: 3 address types discussion
Danny,
Thank you for the reply.
I agree with most comments.
I appreciate the effort you made to make a common understandable email.
But we may further need to better understand each other by email. To that end,
we would need to use a common way of citing each other. There are many ways
that each person believes is the best. But there is only one way that is
agreed by most.
Le 16/05/2016 à 11:58, Moses, Danny a écrit :
> Hi Alex,
>
>
> Thank you very much for the detailed review and comments. I have tried
> to answer them, but if I was not able to be clear enough, I will be
> happy to discuss this with you.
>
> I am removing parts of the previous exchange so that everyone can find
> the open questions and answers more easily.
>
> Regards, /Danny
>
>
> --------------- Text removed from previous exchange-----------------
>
> ----------------------------------------------------------------------
> ----------
>
>
>
>
>
>
>
>
>
> Alex > Yet, some questions arise:
>
> The apps which don't require a session-lasting IP address can
> obviously work with a session-lasting IP address too. So some of the
> session-lasting IP addresses can be non-persistent IP addresses?
>
> Danny > True, applications that do not require a session-lasting IP
> address can work correctly with a session-lasting IP address. But in
> that case, the network will invest resources in guaranteeing this
> without any real need and thus, these resources will be wasted.
Intuitively it can appear so: more state in routers maybe needed to maintain
routes to an IP address which moves, rather than simply changing the IP
address. And that state may look more expensive.
On another hand, it may be that that 'cost' is artificial in the first place.
Modem connections were first charged on a per-time basis - it was artificial;
now there are flat rates. Initial 4G subscriptions plans allowed only for 1Gb
'fair use' - it was artificial; now there is 50Gb 'faire use' for same price as
1Gb only 2 years ago.
Currently, some cellular operators only offer session-lasting IP addresses (do
not offer non-persistent IP addresses).
Danny >>2
I am not sure I understood your comment here. When I referred to 'cost' I did
not mean actual dollars. I was referring to cost of infrastructure, cost in
terms of non-optimal routes (which may translate to latency in packet arrival),
cost of overhead (encapsulation overhead - for example) etc.
Clearly, all these are saved when there is no need for session-lasting IP
guarantee. So if the application does not need this guarantee and the network
supports the ability not to provide this guarantee, both sides gain.
Do you have a concern here? Am I relating to it?
---------------------------------
> From the application side, its traffic may suffer from the treatment
> associated with providing session-lasting characteristics (non-optimal
> routing, encapsulation/decapsulation overhead which influences MTU,
> etc).
Again, this may look so, intuitively speaking. But I beg to differ.
If one looks at a cellular network from a latency perspective, one has to
imagine the radio access link as very large, and the core network as very
small. That means that sub-optimal routing and encap/decap overheads are very
small compared to the radio access.
On 4G, latency between the UE and the first IP hop is in the order of 50
milli-seconds, whereas the Ethernet latency (supposedly used in the wired
segment to the first IP hop) is in the order of 1 micro-second.
So, if instead of the optimal routing 1 micro-second the non-optimal routing is
3 micro-seconds that is very small compared to the radio access latency. That
can not be considered as additional cost.
Danny >>2
As I mentioned, there are various costs, not just latency.
Regarding latency, I am not sure your data will be valid in DMM deployments
with multiple mobility anchors. One of the ideas which relates to multiple
mobility anchors is to place the anchors close to the base-station (or even
co-located in the base-station). This has various advantages (not relevant to
OnDemand), but also a disadvantage in terms of routing traffic after a hand-off
(since the traffic needs to flow through the original mobility anchor.
But the OnDemand draft does not mandate all future networks to support it. We
offer backwards compatibility of working with networks that provide IP address
continuity regardless of the application's request. Operators who believe that
the cost of providing this guarantee is negligible, may decide not to implement
this feature.
-----------------------------------------
> Still, I believe that there will be cases when this is performed.
> One example is in cellular networks that automatically provide
> tunneling and do not support the ability to receive IP address type
> requests from the mobile node.
Right.
> But, if an application requests a session-lasting IP address, and the
> network provides one, it should treat it as such - e.g. provide IP
> continuity guarantee throughout the lifetime of the session. The
> network cannot tell if the address is 'really' using the guarantee or
> not.
I can not agree that an application will require some kind of IP address.
Danny >>2
I do not understand what you cannot agree to. This whole draft is about
applications requesting specific types of IP address. So are you saying that
you cannot agree to the OnDemand concept?
--------------------------------
> I did not understand what you mean by 'So some of the session-lasting
> IP addresses can be non-persistent IP addresses?'.
> No, if the network provides a session-lasting IP address, it is
> committed to guarantee it (even if the application does not need the
> service).
>
> Alex > The apps which require a session-lasting IP address wish to
> introduce overhead in the network?
>
> Danny > Apps which require session-lasting IP addresses are apps that
> cannot recover from the event of source IP addresses becoming obsolete
> as a result of the mobile node moving to a LAN with a different IP
> prefix. This is why they request a session-lasting IP address. Not
> because they wish to introduce overhead - this is a by product...
>
> Alex > Mobile operators using GTP or PMIP do not provide
> non-persistent IP addresses?
>
> Danny > Mobile operators using GTP or PMIP provide a guarantee that
> the source IP address they allocated to a mobile node will continue to
> exist (and be valid) as long as the mobile node is connected to the
> network (or as long as the DHCP lease condition are meat - if DHCP is
> used). This even a better guarantee than what we defined by
> 'session-lasting' IP address, because the GTP/PMIP source IP address
> is guaranteed regardless of the initiation/end of IP session.
I can not imagine a Mobile operator which will give an address to a mobile that
is not accepted on another access point. That is the whole point of 'Mobile'
in mobile operator. Imagine - the end users will have to stay fixed (dont
move), if they started some application. And there are many applications out
there unaware. Or one would have to update all existing applications?
This will _not_ work whenever the user wants to get in a train or car either
(despite having the feeling of being fixed). Or do you consider the Mobile
operator will equip the trains and the cars as well?
Danny >>2
The draft does not say that the end user cannot be mobile. It indicates that
there are more and more applications that do not break when the source IP
address they are using is obsolete. They simply re-open the Socket and use the
new source IP address. So for these types of applications, the session-lasting
guarantee is redundant (even when the user is mobile).
--------------------------------------------
> In our understanding, this extra guarantee is not really needed and in
> the DMM environment where they may be multiple mobility anchors might
> even introduce inefficient routing (which could be avoided in the
> newly introduced scheme).
>
> To the best of my knowledge, mobile operators do not provide
> non-persistent IP addresses.
>
> Alex>
>> Basically, current implementations provide a guarantee for the
>> source IP address to be valid throughout the time the mobile host is
>> connected to the mobile network. We concluded that mobile hosts do
>> not really require such a guarantee. It is sufficient to require a
>> guarantee of the IP address availability while there is/are an IP
>> session(s) using this IP address and hence the more accurate
>> definition.
>
> I dont understand.
>
> Until here the app requirements where important. Now we change to
> make the mobile host to be important(?)
>
> Danny > In current implementation a source IP address is allocated to
> the mobile host and is valid throughout the connection of the mobile
> host to the network. This is why I use the term 'mobile host'
> in the description that you quoted.
>
> We are introducing an new concept - 'OnDemand' - where each time an IP
> session is created (by an application running on the host), an
> application can request a specific source IP address for that session
> (with specific type). From the network's perspective, this address was
> allocated to a mobile host. This means that at any given time, a
> mobile host might have more than one source IP address allocated to
> in, and different IP sessions initiated by that host (by different
> applications running on that host) may be used in different packets
> being sent/received by it.
>
> I can understand why this is confusing. If this explanation is not
> sufficient, I will be happy to discuss this point with you.
Yes, please clarify whether one has to update all applications in order to take
advantage of this 'ondemand' aspect?
Danny >>2
Yes. If an applications wish to request a specific type of source IP address,
they have to use the new flags we are specifying in the Socket interface
-----------------------------
> Alex >
>> Furthermore, some WG members have shown cases in DMM where it is more
>> efficient for applications to request a new Session-lasting IP
>> address when launched rather than using an existing one that was
>> allocated to the mobile host in the past.
>
> Well, I wonder about this.
>
> Danny > Well, they did not use the term 'OnDemand' specifically, but
> they did show that in a distributed mobility anchor scheme, there are
> cases where allocating a new IP address may result in a more optimal
> route of the IP flow compared with using the already-allocated IP
> address. This is because the original IP address is served by one
> mobility anchor (hence all traffic must be routed through it), and
> after the mobile host moves to a new location, traffic may possibility
> be routed via a different mobility anchor which is topologically
> better. Once again, we can further discuss this topic).
> Please refer to the following IDs:
> https://datatracker.ietf.org/doc/draft-bernardos-dmm-distributed-ancho
> ring/
>
>
>
>
>
>
>
https://tools.ietf.org/html/draft-seite-dmm-dma-07
>
> Alex > In the environments I work I never saw an application (e.g. a
> browser) to request an IP address. It is the connection manager which
> deals with address configuration. This connection manager is not in
> contact with other applications like web browsers.
>
> Danny > That is correct - application do not request IP addresses.
> This is a new concept we are introducing. Application, through the
> Socket interface, can indicate to the TCP/IP stack, the type of source
> IP address they require. The TCP/IP stack, will ether associate an
> existing IP address with that Socket or initiate a request to the
> network for a source IP address of the desired type.
It makes sense in a way: applications featuring 'ondemand' may put less
apparent strain in the network.
However, we need to make sure that all existing applications continue to work
as before even if they dont run 'ondemand'.
Danny >>2
That is correct. The draft has a chapter on backwards compatibility.
------------------------------------------
>
> Alex >
>> This is due to possible movement of the mobile host to a LAN which is
>> being served by a mobility anchor that is different from the one that
>> was used when the older Session-lasting IP address was assigned to
>> the mobile host. Fixed IP address (no renaming ...):
>> We believe that this is where our original text was the most unclear
>> leading to the confusion on the mailing list and the comments from
>> the flour.
>
>> A Fixed IP address is guaranteed by the network to Always be valid,
>> even if the mobile host is not utilizing any IP sessions, or has been
>> disconnected from the network for some time. This is a special
>> service that mobile network operators provide for a premium charge,
>> for servers, VPNs , secured content and other applications.
>> With this IP address type the network operator provide IP address
>> reachability in addition to IP session continuity, and mobile hosts
>> may register these addresses in DNS infrastructure for name
>> resolution.
>
> I can understand the intention of the fixed IP address definition.
>
> But I wonder whether there can be an improved definition of a fixed IP
> address. Because of the following:
>
> A 'fixed' IP address, as much as it can be guaranteed by an operator
> at a premium cost, will not be possible if moving to a more remote
> area: for example, when moving from US to Europe can not maintain that
> fixed IP address even if it is paid a very high price.
>
> Danny > With the appropriate roaming implementation, even this can be
> achieved.
Well, I think that can be a good goal, but I can't see it happening.
The roaming concepts in telecom operators kept evolving in recent years:
acquisitions of networking assets across multiple continents; regulatory bodies
fix budget limits while roaming; billing agreements between certain countries
exist to make look as if at home. Yet in none of these is it possible to keep
same IP address while moving across large areas.
Whatever amount of money one pays, one can't just make host-based routes across
the globe - it equates to the price of all Internet altogether.
> But, when moving to an area where no roaming partner exists, you are
> correct. But this sounds to me like a business issue, not technical.
I think it is technical.
I think even when roaming partners exist, some things may appear as if at home,
but never the IP address is the same.
Danny >>2
OK. Let's agree to disagree on whether or not Fixed IP addresses can be
supported in roaming scenarios.
Does that impact the whole concept? I do not think so...
--------------------------------------------
> Alex >
>> Clearly, most mobile hosts do not require Fixed IP
>
> Again: _mobile hosts_ require? Or apps require?
>
> Danny > You are correct. Its applications, not mobile hosts...
>
> Alex >
>
> A more coherent definition can take advantage of using only app
> requirements, or only mobile host reqs, or both but everywhere (i.e.
> each of the 3 types of addresses relates to both MH reqs and to app
> reqs).
>
> Danny > I agree
>
> Alex >
>> addresses and their owners will not pay the premium cost for this
>> service, but still, it is a service that mobile operators provide and
>> this is enough proof for us to acknowledge its need.
>
>
>> Please see some examples
>
> Thank you very much for these pointers. This makes it easier to
> understand the intention.
[...]
> Is this IPv4 only?
>
> I am asking the IP version question because address configuration is
> very different in IPv6 than IPv4.
>
> For example, in IPv6 the network does not assign an address to a host
> (as in IPv4 is done with context setup), but advertises a prefix to a
> link and the host forms an address. In such a context the potential
> mechanism to achieve static IP addresses is very different
> - not only the network is in charge but the terminal too.
>
> Moreover, whereas in IPv4 cellular networks the mechanism to achieve
> static IP address is standardised (NAI, PDP context setup, ppp), in
> IPv6 there is no such mechanism standardised nor deployed.
>
> Is there an example of deployed static IPv6 addresses in cellular
> networks? (as the IPv4 example of AT&T, Verizon, Sprint). That would
> be very relevant too.
>
> Danny > Well, to the best of my knowledge, IPv4 is still more popular
> than IPv6 in the States. But if this service is available for
> IPv4 today, I believe operators will provide it with IPv6 service once
> they believe there is a business justification. We do not want to
> prevent this right?
Right, we dont want to prevent that.
But we dont want to prevent these operators to migrate to IPv6 either, simply
because IPv4 had the above cited Static IP addresses, whereas
IPv6 didnt have.
If the referred operators dont talk IPv6 in the first place in their web pages,
then I think it is not worth talking about them here; moreover - it is not
worth designing solutions satisfying some need they may (or may not) have. We
are a huge distance apart.
Alex
---------------------------------------------------------------------
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.
_______________________________________________
dmm mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/dmm
