# Last edited on 2016-04-22 23:35:57 by stolfilocal # NOT POSTED **The centralized solution** Discussions of the Lightning Network sometimes assume a centralized topology, with a few large hubs that route most payments, while most clients (non-hub players) have only one or two payment channels to hubs. Say, at some point there could be 1'000'000 clients and 100 hubs, each hub serving on average 10'000 clients. With such a topology, the network could be technically viable. The hubs would be permanently on-line and well-connected, and would know each other and their clients. So, when alice wants to send a payment to Zoe through the LN, she just tells that to her hub, and the hubs take care of the rest. They can easily and efficiently find a path between Alice and Zoe that has only hubs as intermediate nodes (if it exists). That path will probably have enough balances in the hub-hub channels; its capacity will be limited only by the Alice-hub and hub-Zoe channels. However, that is essentially a recreation of the traditional bank network, with most of the constraints that bitcoin was supposed to remove. For instance, Alice may identify Zoe by an opaque ID, but Zoe's hub will know her real identity since it has to comply with AML/KYC, economic sanctions, etc.. Being unavoidable intermediaries, the hubs will be able to charge high fees. And both Alice and Zoe would have to be online, at least temporarily, while the payment is set up and executed. **The decentralized solution** Can the Lightning Network work without depending on big central hubs? Imagine again that there are 1'000'000 clients and 10 hubs, but each clients also has payment channels to 3-4 other clients, basically chose at random. Let's assue tha those channels connect all clients in a single unstructured network. How could Alice find a path to pay Zoe through in that network, avoiding the hubs? One can develop algorithms that would solve that problem in a distributed way; but they would be rather inefficient. Alice, or other clinets on her behalf, would have to propagate the query through half of the network, on average -- that is, to 500'000 nodes -- before it reaches Zoe. As a result of that query, Alice would receive one or more paths, possibly overlapping (technically, a subgraph of he network), that could be used for that purpose. Alice then has to choose the actual path(s), and negotiate the chain of payments. The information returned by the query must include only clients that are currently online, and must tell what payment channels they have, and what is the current balance of those channels. That information changes all the time, so it will have to be updated for every payment. Indeed, there is a chance that the path selected by Alice will become invalid before it can be committed, so the information will have to get updated more than once per payment. The query received by each client would be like "is there a way to send X bitcoins to Zoe through you?". The answer that each client should return is a set of paths (tecnically, a directed subgraph) from himself to Zoe, that can be used to send that payment. Now imagine 1'000'000 nodes, each trying to make 1 payment per day, and therefore each having to query the status of 500'000 other nodes at least once a day... I cannot see how a decentralized LN could work beyond a toy example of a dozen nodes, all of them online all the time. AFAIK, its proponent's can't, either. If the LN works, it would have to be centralized in a few huge hubs. That is, the LN would be a reinvention of the credit card network, including AML/KYC and censorship -- but without credit, security, insurance, charge-backs, etc... One can think of many networks that combine large hubs and p2p channels. But it is the proponents who should specify at least one large network model on which the LN could work. (It does not have to be a likely scenario, or even a scenario that can be reached in practice; the purpose is only to check whether the LN could be viable, at least in theory.) Anyway, in a mixed network one could have (say) . In a mixed network model, Alice could opt to find a path to Zoe in the p2p part of the network, avoiding the hubs. Then the above problem remains: in order to find such a path, Alice (or other non-hub nodes on her behalf) must explore half of the 1'000'000 nodes, on average, before finding a capable path to Zoe. If each plain user has 4 channels to other random plain users, the path will probably have 5 hops or more. Then, if the average plain user makes 1 hub-free payment per day, it would have to act as an intermediary 4 times a day or more. In fact, there may be no single path from Alice to Zoe that has enough balances to carry the full amount. Then Alice may have to split the payment into two or more paths, possibly overlapping but distinct (technically, a subgraph rather than a path). Or the payment may not be possible at all... permalinksavecontextfull comments (74)editdisable inbox repliesdelete Maaku: The scaling issue is settled. Bitcoin transactions will scale off-chain with payment channel solutions like lightning. by awemany in btc [–]jstolfi 11 points 12 hours ago Does the Lightning Networ