Re: [Qirg] Survey paper about the Quantum Internet Protocol Stack

[Alexandre Petrescu <[e-mail]>]

Le 08/06/2022 à 18:36, Joseph D Touch a écrit :
> Hi, Marcello, On 6/8/22, 3:45 AM, "Marcello Caleffi" 
> <[e-mail]> wrote:
> 
> Hi Joseph,
> 
> I understand that the sentence copied by Jessica, taken as per-se
> and without reading the long discussion about "physical quantum 
> connectivity" vs “entanglement-enabled virtual connectivity” 
> conducted within the paper, could raise some misunderstanding.
> 
> FWIW, I replied to the post, but based the reply on the paper, not 
> just the post.
> 
> Indeed, we believe that there is a lack of common terminology within 
> the community — with different authors using different names for 
> similar functionalities and vice-versa — and this might have impact 
> as well.
> 
> Agreed. If you want to talk about classical network terminology, 
> you’ve got 40+ yrs of expertise and its terminology to catch up 
> with.
> 
> I'll try to summarize some concepts at my best, tough some reading
> of the paper is the better option.
> 
> 
> ------------------------------------------------------------------------------------------------
>
>
> 
Coming to your observations about classical Internet:
> 
> - I agree with you about ND “discovering” L2 connectivity
> 
> - but L2 connectivity means that sender could reach destination by 
> simply transmitting the packet through the proper interface.. this 
> requires that there exists a physical channel (or a sequence of 
> physical channels with some optional hardware — the switch — in 
> between with the only purpose of reducing the collision domain) 
> interconnecting source and destination
> 
> The same can be said of L3 connectivity, L4, L5, etc. They all
> assume that sending a packet out *that* layer implies an ultimate
> physical path. There is nothing magic about the L2 seen by an L3.
> E.g., when your IP sends out Ethernet, it could be going over many
> layers - MPLS, SONET, WDM, etc. - including being tunneled back over
> IP or even UDP.
> 
> - clearly, probing the existence of a (sequence) of physical 
> channel(s) doesn’t imply that we can acquire full knowledge about
> the physical topology (and I substantiate your example about complex
>  ethernet networks by mentioning that ND in general can’t discover 
> redundant links closing a loop within switches but simply 
> "Spanning-Tree”-enabled links),
> 
> You never probe "physical channels". You probe interfaces. They can 
> be logical, e.g., when you run a VM inside a machine, the Ethernet 
> you think you send out the interface you think you have don't exist 
> "on a wire" unless that packet leaves the machine; even then, it can 
> be encapsulated.
> 
> ND won't find lower layers that have loops that the lower layers 
> don't detect or prevent - but that's a useless tautology.
> 
> - so, I stand with the sentence "ND gather information about the 
> physical connectivity”, or if you prefer with the sentence "ND
> gather partial information about the physical connectivity”, as long
> as we don’t write something like "ND gather full knowledge about the 
> physical connectivity”
> 
> ND gives you none of this on its own. ND is used by IPv6 to discover 
> L2 addresses (and sometimes to redirect to different routers), i.e., 
> it is ARP, proxy-ARP, and ICMP redirects rolled into one. There are 
> ways in which those tools can be USED to gather L2 **REACHABILITY**, 
> but that's not topology. And they don't just spit that info out. You 
> have to write a tool do do this (there are some available).

I agree.

Me too I cant really understand when it is said that ND might gather
some information about the physical connectivity.

I have a problem in understanding 'physical' connectivity in a first
place, anyways, but that's another matter.

I do applaud the intention of the paper to approach the two worlds:
classical Internet and quantum Internet.  I find the ways to achieve it
valuable; the comparisons of different directions are advantageous.

But in my humble oppinion, there are very very many things that can be
improved, in the detail of the text.

SInce the paper is well structured, and very detailed, and very helpful
with these boxes with quantum concepts, it really invites reading and
commenting.

About ND: if I wanted to compare the lack of a cable between two
entities in a quantum network to the necessity of a cable in the
classical Internet, then I would use the protocol IP as example, and not
ND.  I would then explain that, currently, IP is IPv6, but the version
number might be irrelevant for the discussion.

I would then say that IP needs electromagnetic waves (be them through
copper, fiber or through nothing, like WiFi) - in order to transmit
datagrams, otherwise IP would not work.  I suspect the qubits dont need
electromagnetic waves in order to appear elsewhere (aka 'to be
transmitted'), hence there is a difference.

I would search which field in an IP header better expresses that need of
electromagnetic waves.  There is no field which tells that it might need
a certain power of these waves, or more Teslas than Volts.

But there is an enormous number of kinds of ways to transmit these
eletromagnetic waves.  In this set, there are even more kinds of
'links'.  For example, for transmitting electromagnetic waves only
through copper there are many Ethernet, USB and ATM kinds of links; for
'wireless' too (transmit electromagnetic waves through nothing) there 
are many kinds of 802.11 and 802.15.4 kinds of links.

There is a relationship that can be made between IP and the
heterogeneity of the links carrying these electromagnetic waves.  IP
exists only because there is heterogeneity of links.  If all links were
alike there would be no IP.

The field 'Hop Limit' tells how far a packet could go, in terms of
number of IP hops.  It is the number of intermediary hops - the
different links - which is decremented at each hop; if one sets that Hop
Limit to 0, and tries to transmit it, it will not go anywhere, because
it would be 'dropped', like when grounding a signal.  In a quantum 
network, maybe one has to reset it like that (0) in order to make it 
appear elsewhere (aka 'transmit' it).

There might be other fields in IP headers that might be used as help to
illustrate that concept of IP needing electromagnetic waves in classical 
Internet.  But the relationship of ND to an assumed physical layer is 
probably less (hmm, maybe as much as) appropriate.

Of course, there would be a fundamental question of what the IP address
proper might mean when even a single bit of it might be a qubit,
actually.  When I manually assign an address to an interface, that IP
address is immuable: it is 1::1 forever.  But with quantum addressing,
this IP address is 1::1 'give or take'.

There are more things that I would like to tell about this paper, but
let me finish reading it.

Alex


> 
> 
> ------------------------------------------------------------------------------------------------
>
>
> 
Coming to the paper sentence (BTW, we didn't wrote that L2 classical
> topology is correlated with quantum link topology), the “physical” 
> connectivity should be placed in the context of “physical" vs 
> "entanglement-enabled virtual” (with the main meaning of “physical" 
> for distinguish from virtual) rather than L1 (physical layer) vs L2 
> (link layer).
> 
> I mentioned that L2 classical isn't correlated with quantum as a 
> caveat in general. It wasn't aimed at something you said. However, 
> you do imply that quantum has a level of logical connectivity that 
> classical does not, which is the converse; classical has many much 
> more rich variants of logical, including concurrent overlays, nested 
> overlays, etc.
> 
> Let me give some more contest about the sentence (but, please, these 
> are just short extracts from the paper with no ambition of providing 
> a complete self-consistent description)
> 
> Pag. 13, Sec. V.1
> 
> <<<
> 
> In classical networks, a single concept of connectivity arises, 
> referred to as \textit{physical} connectivity.
> 
> Reachability is a single concept. Connectivity is not  - again,
> there is no way (from inside a network) to discover physical
> connectivity, only reachability.
> 
> Whenever there exists a physical communication 
> link\footnote{Obviously, the definition of physical connectivity can 
> be easily extended to a multi-hop route constituted by several 
> communication links.}
> 
> That is known as reachability, as above.
> 
> between two nodes, these nodes are defined ``\textit{connected}''. 
> And the successful transmission of a classical message between these 
> two nodes requires at least one use of the physical communication 
> link. As a consequence, the successful transmission depends on the 
> instantaneous propagation conditions of the physical channel 
> underlying the communication link. Stemming from these 
> considerations, the classical connectivity is \textit{physical}
> since it strictly depends on the physical channel.
> 
> Quantum connectivity has the same dependence (see below). All 
> communication depends on a physical channel, in that a physical 
> channel is defined as the physical means by which two or more 
> endpoints establish communication (quantum or classical). Again,
> this is a tautology driven by definitions and not enlightening.
> 
> Conversely, quantum teleportation enables the transmission of one 
> qubit without any use of a quantum link. Specifically, as long as an 
> entangled state -- say an EPR pair for the sake of simplicity -- is 
> shared between two nodes, they can transmit a qubit regardless of the
> instantaneous conditions of the underlying physical quantum channel.
> 
> I would agree to the claim: "as long as a quantum connectivity was 
> established before, teleportation can transmit a qubit without that 
> link being available now". But you can't establish entanglement 
> without quantum communication. Note - as with classical, quantum 
> comms includes "sneaker-net" (physically moving fermions).
> 
> Remarkably, the qubit transmission is still possible even if the 
> nodes are not anymore interconnected by a quantum link\footnote{It
> is worthwhile to note that, thanks to the \textit{deferred
> measurement} principle \cite{NieChu-11,CuoCalKrs-21,IllCacMan-21},
> the transmissions of the two classical bits -- and the subsequent 
> post-processing at the destination needed for performing a 
> teleporting operation -- can be delayed at any convenient time. 
> Accordingly, in this section we focus on the peculiar connectivity 
> characteristics arising with quantum entanglement.}. In this sense, 
> we can say that entanglement enables a \textit{virtual} quantum
> link, and consequently the concept of \textit{virtual connectivity} 
> arises.
> 
> This is the key point and agreed.
> 
>>> 
> 
> By oversimplifying:
> 
> - physical connectivity means that the connectivity requires the 
> exchange of a message through a sequence of channels, whose 
> instantaneous conditions affect the exchange. During the
> distribution of entanglement or by directly transmitting a qubit,
> you’re facing with physical connectivity.
> 
> - Yet, once shared, entanglement provides a virtual connectivity
> that is independent from the instantaneous propagation conditions of
> the physical channel
> 
> - (see also the remaining part of Sec. V.1 for further details).
> 
> Agreed; my observation is that some of the rest of the discussion of 
> classical networking and the implied new capability of ND (vs ARP, 
> ICMP ping, etc.) is unnecessarily misleading.
> 
> Pag. 25, Sec., VII.5
> 
> <<<
> 
> Clearly, as for the classical Internet, the Quantum Internet should 
> rely on some networking functionalities such as path discovery, 
> forwarding and routing [110, 21]. But these functionalities must be 
> designed to account for the peculiarities of the entanglement as 
> communication resource.
> 
> Indeed, part of these functionalities can be carried out through 
> classical networks by existing protocols. As instance, neighbor 
> discovery – used by network nodes to gather information about the 
> physical connectivity – could be accomplished by resorting to 
> classical protocols [111, 112, 113].
> 
> Reachability, not connectivity. AODV and other ad-hoc protocols are 
> no more useful than would be RIP, OSPF, or IS-IS, or even BGP. And 
> IPv6 ND is no more useful than are IPv4 ARP and/or ICMP. AODV and 
> ad-hoc protocols discover connectivity over multipoint links, which 
> you haven’t discussed at all and are not uniquely relevant.
> 
> However, as widely described in Sec.5, the concept of virtual 
> connectivity – including its variations such as the augmented and 
> on-demand connectivity – arises with entanglement.
> 
> Classical nets have virtual connectivity too. What you’re referring 
> to is what I would call “temporally discontinuous reachability” (or 
> just “discontinuous reachability”), and exists in classical nets too 
> – e.g., email relay and DTN (delay tolerant networking – which is 
> basically re-invented email relay).
> 
> IMO, the unique feature of quantum nets is what I would call “post 
> facto reachability”, which classical networks cannot achieve.
> 
> (note: the signature I use on my personal email is not accidental: 
> “Temporal epistemologist”, a term that basically means “one who 
> studies communication protocols”, and was coined by my wife, Gail.
> It refers to the fact that communication protocols describe how 
> information flows over time)
> 
> Whether existing neighbor discovery algorithms can be employed for 
> virtual neighbor discovery – and how the physical neighbor discovery 
> should interacts with the virtual one – is yet to be determined. 
> Indeed, not only the virtual connectivity dynamics are intrinsically 
> different from the ones arising with physical connectivity – as 
> described in Section 5 – but when it comes to multipartite 
> entanglement the concept of neighborhood evolves from a binary 
> question – “is a certain node one of my neighbors?” – to a more 
> complex question, including at the very least the discovery of the 
> identities of all the entangled nodes.
> 
> They’re the same question and “neighbor discovery algorithm” isn’t 
> relevant here. That implies the difference between one-hop L2 
> reachable and L3 one-hop reachable, which isn’t relevant for either 
> classical or quantum.
> 
> You do need knowledge of reachability as a topology, but again, why 
> does that need to be physical, even for the quantum layer? What do 
> you care if you send a qubit or some lower layer teleports it? Isn’t 
> it the same from your layer’s viewpoint?
> 
> The only issue appears to be the difference between quantum 
> reachability (which can be discontinuous) and classical 
> reachability.
> 
> Furthermore, both physical and virtual neighbor discoveries play a 
> pivotal role for the deign of routing services such as path
> discovery and path forwarding.
> 
> Again, these are dependent on reachability only.
> 
> Here, the first step is to identify, within the quantum network 
> infrastructure responsible for the distribution of shared entangled 
> states, at least a physical quantum path between source and 
> destination. This quantum path must be augmented by a classical
> path, so that quantum nodes can exchange proper classical signaling
> as discussed in Section 7.2. In this regard, one should argue that 
> physical connectivity enables direct communications between neighbor 
> nodes, whereas quantum repeaters and entanglement swapping extend
> the spatial domain of the virtual connectivity, enabling direct 
> communications between nodes that may be topologically remote. 
> However, virtual connectivity should not be considered as the main 
> connectivity, as well as neither physical and virtual connectivity 
> should be considered as mutually exclusive strategies. On the 
> contrary, they are strictly correlated and path discovery should be 
> able to evaluate – case by case – whether entanglement-based 
> communications outperform direct dispatch, where quantum information 
> is directly transmitted through the physical quantum channel.
> 
>>>> 
> 
> So, in a nutshell, in the Quantum Internet:
> 
> - we need classical topology discovery AND quantum topology 
> discovery
> 
> Agreed. Where topology is defined by reachability, IMO.
> 
> - classical topology discovery might resort to classical protocols
> -> but we don’t believe this is the correct direction, see as
> instance [1,2] where we gave some insights on quantum solutions to
> classical network functionalities such as MAC
> 
> What you really need is to have classical reachability between each 
> pair of quantum nodes at the quantum layer you are designing. There 
> is no need for a complete classical net per se.
> 
> - quantum topology discovery can’t resort only to a quantum version 
> of “classical” protocols because the connectivity is virtual, 
> augmented, on-demand.. so we need to figure out new algorithms and 
> the worst decision would be starting to design a Quantum ND without 
> fully understanding the peculiar features with no classical 
> counterpart of entanglement
> 
> The only way to “discover” a quantum topology is to try to send 
> qubits and see where they reach. It’s the exact analog of classical 
> topology discovery. In both cases, as they say with investment
> advice “past behavior is not a guarantee of future success”. That’s
> even more true for quantum nets, because the use of entanglement for
> qubit transmission is destructive (to the entanglement).
> 
> [1] A.S. Cacciapuoti, J. Illiano, S. Koudia, K. Simonov, M. Caleffi, 
> "The Quantum Internet: Enhancing Classical Internet Services one 
> Qubit at a Time", arXiv, May 2022.
> 
> https://arxiv.org/abs/2205.09476
> 
> [2] J. Illiano, M. Viscardi, S. Koudia, M. Caleffi, A.S.
> Cacciapuoti, "Quantum Internet: from Medium Access Control to
> Entanglement Access Control", arXiv, May 2022
> 
> https://arxiv.org/abs/2205.11923
> 
> 
> ------------------------------------------------------------------------------------------------
>
>
> 
Btw, thanks for the interest and the email, and feel free to reach
> us out for any further comment or doubt.
> 
> Regards,
> 
> Marcello
>