You are on page 1of 9

Is Filecoin a $257 million Ponzi scheme?

Marc Juchli Johan A. Pouwelse


Faculty of Electrical Engineering, Parallel and Distributed Systems Group
Mathematics and Computer Science Department of Software and Computer Technology
Delft University of Technology Delft University of Technology
m.b.juchli@student.tudelft.nl j.a.pouwelse@tudelft.nl

AbstractCentralized storage providers are dominating the sums by reference to heavily discussed topics in social media
storage market and host a major part of internet data. In the channels. A number of sections from the Simple Agreement
recent blockchain era, the idea of decentralized storage was for Future Tokens (SAFT) are then discussed; they include
connected with the incentive that leads to a decentralized storage terms which might be unattractive to the unaware investor
market. This paper analyzes Filecoins ICO and its technical and benefit the founder team of Filecoin. Regarding Filecoins
capabilities, as described in detail in its Whitepaper. Moreover,
the InterPlanetary File System (IPFS) project will be considered
technical potential, we mainly focus on what is proposed in
in detail; this is a peer-to-peer hypermedia protocol which the Whitepaper [6]. Thereby, we aim to uncover potential
Filecoin uses as its the underlying data store. Further, the question weaknesses but also highlight strengths. In addition, the novel
of whether or not Filecoin is an economically feasible project proof-of-spacetime consensus algorithm is highlighted and
is discussed by means of extending the work of a hypothetical compared with consensus proposals from projects such as StorJ
risk analysis, considering the possibility of a Ponzi scheme and and Sia. Moreover, we explore the details of the InterPlanetary
finally, sketching a back-of-the-envelope calculation. The paper File System (IPFS) project, a peer-to-peer hypermedia protocol
uncovers unattractiveness during the ICO which leaves investors which is used as the underlying data store of Filecoin. Not only
without any control after their investment. Filecoins Whitepaper do we set out the technical foundation, but we also explain a
introduces novel concepts and predecessor projects from the possible way for Filecoin to use IPFS as its storage adapter
team members proof technical capabilities. We highlight that the
integration of IPFS into Filecoin can be achieved by exploiting
by means of Bitswap [7]. The feasibility in terms of economic
the capabilities provided by Bitswap. The feasibility in economical design is briefly discussed by reference to a hypothetical risk
terms is hard to predict and the biggest hurdles is probably going analysis, a consideration of whether or not the ICO launch
to be the acceptance from the users. In addition, Filecoin shows might be a Ponzi scheme, as well as with a back-of-the-
certain characteristics of a Ponzi scheme but the trust built by envelope calculation.
the team members in the past leads to believe otherwise.
The structure of this paper is as follows: Section II lays
Keywordsfilecoin, crypto-currency, ipfs, libp2p, decentralized out a history of decentralized storage projects with a similar
storage, ico, proof-of-spacetime, ponzi scheme. ambitions to Filecoin, but that have failed over time. Section
III mentions recent competitor projects that make use of a
I. I NTRODUCTION blockchain. Subsequently, Section IV introduces the Filecoin
project by analyzing the ICO including reasons for the sale
A MASSIVE AMOUNT OF STORAGE SITS UN- and allocation of tokens. Section V considers whether the
USED IN DATA CENTERS AND HARD DRIVES promises of the Filecoin project are technically feasible or not.
AROUND THE WORLD. [1] Accordingly, we are then able to analyze the InterPlanetary File
System (IPFS) [8] in Section VI and highlight possible ways
With this slogan, Protocol Labs is about to disrupt the storage means of adoption. In Section VII, we question the economic
market by using proof-of-spacetime as their driving source. feasibility in a rudimentary manner and lastly draw a number
The Filecoin project describes a decentralized storage market of conclusions about the entire Filecoin project in Section VIII.
in which anyone, worldwide, is able to participate as a storage
provider. This concept is so promising that it has convinced II. 15 YEARS OF DOCUMENTED FAILURE
investors to provide a total of $257 million so far the
biggest initial coin offering (ICO) as of today (September In July 2000, long before the blockchain era, the Mojo
2017). However, the concept of a decentralized storage market Nation software [3] was released, with the aim of providing
is not novel. Others [2] [3] have tried in the past, yet were an emergent file store to its users. The project successfully
not able to reach Filecoins advertised scaling capability, and deployed a decentralized storage network in an environment
eventually failed. More recent projects [4] [5] are currently consisting of unmanaged nodes. It used consistent hashing [9]
working towards building a similar system with conceptual to locate nodes and data blocks. However, the project was shut
differences which are explained briefly in this paper. down in February 2002 due to a number of problems:
This paper aims to analyze the Filecoin project primar- Data Availability: the main issue was the inconsis-
ily with regard to its technical aspects, and secondarily, its tency of data available to its users as it depended
economic aspects. The most relevant financial figures relat- upon which server nodes were connected at any given
ing to the Filecoin ICO are highlighted and then followed time. According to Maymounkov et al. [10] this could
by suggested reasons for the exceptionally large investment have been avoided by heuristically favoring long-lived
nodes and by discriminating against newly joined Once the host disappears or goes offline, payments
nodes which show frequent join-and-leave behavior. are halted.
Firewalls and NAT: networking hurdles such as fire- MaidSAFE goes beyond decentralized storage with a
walls and network address translation (NAT) prevented less sophisticated focus on efficiency compared to its
a substantial amount of nodes from acting as servers. rivals. MaidSAFE uses a novel scheme for achieving
consensus by not relying on a blockchain but instead
Mutual distrust: in order to enable a network of
on close group consensus and hence is not proof-of-
nodes to behave as designed, a motivation to coop-
work related.
erate needed to be established within the network,
combined with a sophisticated attack resistance mech-
anism which would have prevented nodes from using IV. ICO ANALYSIS
the resources of other nodes without offering an equal
The Filecoin ICO started on August 10, 2017 and com-
level of service in return.
pleted its offering on September 7, 2017. It was the first
ICO ever to comply with SEC securities regulations and
In addition, academia has been struggling to build self- hence only accredited investors were allowed to contribute
organizing systems. Tribler [2] is based on a robust reputation (Reg. D, 506(c), see [15]). Further, the ICO was conducted
and the sharing of expertise thereby, facilitating an environ- using CoinList [16], a platform for token sales, also built
ment where a decentralized marketplace can be thought of by Protocol Labs. CoinList partners with AngelList [17] who
as something which treats storage as an asset. In its 12- are responsible for legal compliance. In total, approximately
years development history, there have been many challenges. $257000000 was raised, composed of $52000000 from
For example, security issues including collusion attacks [11] presale and $205800000 from Reg D investments. In respect
have slowed down the progress of development enormously, of the latter category, $135 million was raised within the first
in particular, preventing the project from scaling up. hour.

III. R ECENT C OMPETITORS A. Simple Agreement for Future Tokens


At the time of writing, a sudden glut of ICOs has appeared, The issue of the tokens during the ICO was regulated
some of which are projects that are destined to have the same by a Simple Agreement for Future Tokens (SAFT). This
fate as Filecoin. Figure 1 shows that the funds invested in instrument provides a framework for Coinlist to grant to
ICOs have exceeded the investments made in the venture investors the right to redeem Filecoin tokens (FIL) in the
capital sector. The following are examples of some of the future. The definition of SAFT, however, also includes the
following statement:

...a significant portion of the amount raised


under the SAFTs will be used to fund the Compa-
nys development of a decentralized storage network
that enables entities to earn Filecoin (the Filecoin
Network) (sic). [18]
Therefore, the extent to which the development process is
being funded is not explicitly mentioned, and thus is left to be
decided solely by Protocol Labs Inc. Additionally, the SAFT
gives to the token intermediary, CoinList, a wide discretion
Fig. 1. Initial Coin Offerings (red), Blockchain Venture Capital (blue) [12] as the agreement not only enables the verification of investors
as accredited but also define certain events to be transparently
most capitalized projects to have resulted in decentralized, defined. Transparency is certainly appreciated when funds are
incentivized, and byzantine fault-tolerant storage networks being transferred, but needless to say, the terms have to be fully
which will eventually compete with centralized alternatives understood by both parties. One event, the impact of which
such as Amazon S3 [13] are: StorJ [4], Sia [5] and MaidSafe might be underestimated by the investor, is the Dissolution
[14]. Event and is defined as follows:

Sia: supports on-blockchain smart contracts which Dissolution Event means (i) a voluntary termi-
define payments for hosts while providing storage. nation of operations of the Company, (ii) a general
Payment is guaranteed once the contract has been assignment for the benefit of the Companys creditors
created and ensures that the host is still paid even if or (iii) any other liquidation, dissolution or winding
the uploader never accesses the file. Additionally, the up of the Company, whether voluntary or involun-
contract enforces penalties for hosts which go offline tary. [18]
or lose data.
Knowing that, at any given time, it is possible for Protocol
Storj: is similar to Sia but does not feature on- Labs to terminate their business, the entire commercial objec-
blockchain storage contracts; instead, it offers a pay- tive of the ICO becomes fragile, as is clear from this extract
as-you-go model for nodes which provide storage. from the execution plan:
If immediately prior to the consummation of
the Dissolution Event, the assets of the Company
that remain legally available for distribution to the
Purchaser and all holders of all other SAFTs (the
Dissolving Purchasers), as determined in good faith
by the Companys board of directors, are insufficient
to permit the payment to the Dissolving Purchasers
of their respective Returned Purchase Amounts, then
the remaining assets of the Company legally avail-
able for distribution will be distributed with equal
priority and pro rata among the Dissolving Pur-
chasers... [18]
Fig. 2. Filecoin sale price function
In other words, only the remaining assets which are legally
available will be distributed in the event of a Dissolution
Event, including voluntary termination as seen in definition As a result, the closing price for investments beyond the value
(i). Unfortunately, if one considers the worst-case scenario in of $257000000 can be estimated at approximately $6.425.
which Protocol Labs decide to terminate the operation after Hence, advisors and Protocol Labs Inc. itself were able to
having spent all the funds available, the investors would likely purchase at a price 8.57 times lower than paid by the later
be excluded from the distribution of any assets. Similarly, if investors.
the project does not announce a Network Launch [18] event by
July 18, 2022 (including a 60-day extension), then according
to definitions (i) or (iii), investors would have to expect a V. I S THE DESIGN TECHNICALLY FEASIBLE ?
Dissolution Event to be announced as a subsequent step. In this section, certain technical details will be discussed
and the issue of whether the decisions stated in the Whitepaper
B. Token allocation [6] will lead to a satisfactory outcome or whether any weak-
As presented in [19], Filecoin tokens are distributed to four nesses in these decisions will emerge and threaten the success
groups of participants: of the project. Further, we aim to make comparisons with more
recent decentralized storage projects involving blockchains,
70% to Filecoin Miners as mining block rewards once such as StorJ [4] and Sia [5].
the Network Launch has taken place
15% to Protocol Labs as its genesis allocation with A. Decentralized Storage Network
6-year linear vesting Conceptually, a Decentralized Storage Network (DSN) is a
10% to investors as its genesis allocation with linear scheme with associated protocols that enables the aggregation
vesting of 6 months to 3 years and provision of data to be decentralized in a coordinated man-
ner. The protocols verify that the parties involved execute their
5% to the Filecoin Foundation as its genesis allocation operations securely and so facilitate coordination without the
with 6-year linear vesting aid of a trusted party. The Filecoin DSN introduces three types
of participants: (1) Clients who pay FIL tokens to the store and
The total of US$257 million raised during the advisor presale
retrieve data, (2) Storage Miners who pledge FIL tokens as col-
and investor sale (see IV-C) therefore only accounts for a
lateral while providing amounts of storage which correspond to
maximum of 10% of the total coins expected to be in cir-
their value in FIL, thereby making them eligible to mine new
culation several years after the Network Launch. The half-
tokens, and (3) Retrieval Miners who provide data in response
life of Filecoin block rewards is set to 6 years and this will
to client requests, and oftentimes also act as Storage Miners.
likely not change significantly for several years given that the
Further, Filecoin personifies all the participating parties which
Network Launch requires a solid implementation of the Filcoin
run a full nodes as The Network (N ). Hence, the protocol
ecosystem.
scheme is being constellated as a tuple in a very suitable and
abstract definition: (Put, Get, Manage). The process of
C. Token sale storing and retrieving data is undertaken by clients using the
There was a two-phase process for the distribution of Put or Get Protocol. The Network is coordinated using the
SAFTs: the first phase enabled Protocol Labs as well as Manage protocol, which serves to control the available storage,
Filecoins advisors to purchase SAFTs prior to the broader audits offered services by Storage Providers and repair possible
group of later investors. The latter were able to purchase in faults. Little is known about Manage.AssignOrders or
the second phase. In the first phase, the price was fixed at Manage.RepairOrders, such as whether the Manage pro-
0.75 USD/FIL. For the second phase, a pricing function was tocol follows incentive and rewards nodes in order to actively
introduced: maintain the network. However, on the basis of our research
so far, we consider that the Filecoin DSN, is conceptually
amountRaised
price = max($1, ) excellent, compared to its competitors identified in Section III.
$400 0000 000 The simplicity of (Put, Get, Manage) implicitly covers
The price determined by this function increased linearly ac- additional client operations such as PING and FIND_NODE,
cording to the amount being raised, as indicated in Figure 2. as defined in StorJ [4].
The process of storing data, initiated by a client, involves confirm a match of bid- and ask orders in the form of a Deal
splitting the data into pieces in order to store it within order. Lastly, a Pledge engraved in the ledger represents
sectors, as part of the disk space provided by storage the collateral of the storage miner deposited in order to accept
miners. Conceptually, the data handling is very similar to StorJ orders from the clients, as described in V-A. In contrast, Sia
[4] where data, after it undergoes AES256-CTR encryption only stores storage contracts which define the terms of the
executed by the client, is split into shards. While StorJ agreement between the parties and therefore it relies on a
dynamically adjusts the size of the shards depending on the file variant of the Bitcoin protocol [21]. As Filecoin is clearly
to be stored, it is not yet known how Filecoin will approach more diverse in terms of data structures, additional complexity
the splitting into pieces. Filecoin, like StorJ, delegates the in the client implementation is to be expected. From another
encryption task to the client. However, whereas StorJ enforces perspective, this decision might enable the protocol to be
encryption in its reference client implementation, Filecoin extended more easily as the ledger is already laid out in such a
currently has no plan for offering built-in encryption and leaves way as to handle various types of data structures. In addition,
this responsibility solely to the client before they access the the decision to have a native blockchain was pre-condition
Filecoin network. In order to introduce redundancy for the data for the use of Proof-of-Spacetime, see V-C. It was announced
to be stored, Filecoin neatly introduces a replication factor to at DEVCON2 [22] that the Ethereum Network [23] would
be chosen during the Put protocol, and which can be used be relied upon, however this decision was later abandoned.
to increase the tolerance of storage faults. In contrast, StorJ This meant that there was no ERC20 token, [24] which in
therefore uses a distinctive mirror method. turn prevents direct compatibility with Ethereum and therefore
requires a bridge of some sort.
The guarantees and requirements laid out by the Filecoin
DSN can be summarized as follows:
C. Proof-of-Spacetime
Integrity: In order to ensure that clients do not accept The Whitepaper [6] describes a novel consensus protocol:
altered or falsified data, cryptographic hashes are Proof-of-Spacetime (PoSt). Proof-of-Spacetime makes it pos-
used as a naming convention and serve as identifiers sible to check if a prover is storing data for a range of time,
for data retrieval (Get protocol) and verification of or more formally:
content. Unlike its competitor StorJ [4], Filecoin does
not rely on any metadata but relies on hashes only. A PoSt scheme enables an efficient prover P
Retrievability: after the data is successfully stored, to convince a verifier V that P is storing data D for
clients are assured that the very same data can eventu- some time t. [6]
ally be retrieved. The (f, m) tolerant system spec- This scheme takes advantage of the capabilities of Proof-
ifies that, assuming m storage miners, a maximum of of-Storage [25], which can confirm if a storage provider is
f faults are tolerated. Increasing the replication factor storing data at the time of the challenge, by sequentially
hence implicitly increases the chances of recovery. generating proofs and recursively composing the executions
Public Verifiability and Auditability: as storage miners thereof. The precise implementation of Proof-of-Storage is
are obligated to submit proofs of storage (see V-C) described as Proof-of-Replication (PoRep), a novel concept
to the blockchain, any user can verify their validity that enables a storage miner to convince a client that its
without having access to the data. Since proof-of- data has been replicated to a uniquely dedicated physical
spacetime: implicitly guarantees the continuous ex- storage. Other Proof-of-Storage schemes such as Provable
istence of the data on the storage miner side, no Data Possession (PDP) [26] and Proof-of-Retrievability
challenge-response communication is required. Com- (PoR) [27] essentially guarantee possession of some data
pared to StorJ [4], the communication overhead is at the time of the challenge/response. Proof-of-Replication,
therefore reduced while providing the same guaran- however, improves those schemes by preventing Sybil Attacks,
tees. Outsourcing Attacks, and Generation Attacks. As Proof-of-
Spacetime is based on Proof-of-Replication, these properties
Incentive compatibility: like any of the projects men- are inherited. Thus, PoSt prevents claims that there are copies
tioned in III, Filecoin enforces incentive by rewarding which are not physically stored (Sybil Attack). Further, it
parties which store data and punishing those who lose denies storage miners the possibility of offering more storage
data. than is physically available (Outsourcing Attack) and also
Confidentiality: as mentioned earlier in this section, protects against on-demand data generation thereby ensuring
Filecoin is weaker in terms of confidentiality as it fully that the data is stored on a physical disk (Generation Attack).
delegates the encryption task to the client. The cryptographic building blocks of PoRep and PoSt rely
on zero-knowledge Succinct Non-interactive Arguments of
Knowledge (zk-SNARKs). Zero knowledge proof, including
B. Ledger
zk-SNARKs, have been shown to have great potential, for
The Ledger L described in the Filecoin Whitepaper is example, to aid the building of large scale projects such as
represented by a native blockchain, as announced in [20], and ZeroCash [28].
supports various types of data structures. The state of the
DSN is stored within an AllocTable, which keeps track Apart from the extensive capabilities in terms of storing
of pieces and their assigned sectors. The Orderbook data, Proof-of-Spacetime attempts to reduce resource waste
is responsible for storing Orders, which either state a request by treating the storage of users data as a form of work.
to store data (Bid order), offer a service (Ask order) or It is therefore a consensus protocol based on useful work.
The usefulness means that computational power is not blockchain acting as a trusted party introduces the need for
wasted, as is the case for Proof-of-Work [21] and other other ways of forming trust between clients and retrieval
consensus algorithms. In Filecoin, the voting power (the miners. Filecoin essentially relies on token exchange as an
probability of a miner being elected to create a new block) an instrument of trust. The data to be delivered is split into
increases proportionally with the amount of storage offered multiple pieces and, for every successful exchange of a piece,
to the network in relation to the storage resources of the the client pays the miner. If one of the parties involved does
entire network. Due to the fact that Proof-of-Spacetime not fulfill their obligations, the other party is free to stop.
composes Proof-of-Replication executions sequentially, it is In order to process payments, the first requirement is a
further considered to be non-parallelizable and thus greater network of payment channels. Filecoin has not stated more
computational resources will not have any noticeable effect. detailed plans of how to integrate payment channels into the
[6] retrieval market. Although this is a risk, much research has
been done and promising projects have been developed [30].
However, given that storage miners may offer varying
The order-matching phase of the verifiable market
amounts of storage results, their influence on the network
protocol differs significantly from the storage market. Since
is distributed continuously and unequally. Hence, a naive
the orderbook cannot be recorded in a blockchain, clients and
Byzantine Fault Tolerance approach that uses the number of
retrieval miners have to gossip about their orders. Filecoin
faulty nodes is, in the case of Filecoin, not a measure sufficient
assumes that, at this point, there is always at least one honest
to determine the outcome of a protocol. Instead, Filecoin
retrieval miner. [6]
proposes Power Fault Tolerance (PFT) [29], an abstraction
that defines byzantine faults in terms of the influence of a
participant. VI. IPFS
As described in [6], Filecoin serves as an incentivized
D. Decentralized Markets seeding layer on top of the InterPlanetary File System (IPFS)
Filecoin introduces two types of markets, Storage Market [8]. The Filecoin project is therefore beneficial to IPFS as it
and Retrieval Market, with two accompanying independent, intends to add more storage to the network. At the same time, it
decentralized exchanges. In the Whitepaper, the notion of critically depends on the technical capabilities and robustness
Verifiable Markets is presented; it refers to a market where of IPFS. This section aims to highlight the capabilities of IPFS
participants are able to verify exchanges between buyers and explains the conceptual decisions made while building a
and sellers. Specifically, the protocol describes a two-phase peer-to-peer hypermedia protocol. The latter includes a brief
process: Order matching enables participants to add buy- overview of the libp2p project [31], a modular peer-to-peer
and sell orders to the orderbook and eventually to create networking stack. Lastly, we discuss possible approaches to
deal orders once the two opposite (buy and sell) orders have the integration of IFPS into Filecoin.
been matched. The second phase, described as Settlement,
involves ensuring that the network correctly executes the A. The flaws of HTTP
transfer of goods (data) and eventually initializes a payment. The Hypertext Transfer Protocol (HTTP) can be seen as
The Verifiable Market Protocol applies to both of the the global information protocol that has standardized how
aforementioned markets. [6] people distribute and present information. Publishing content
with HTTP is nowadays almost free and has led to great
The main function of the Storage Market is as a platform outpouring of innovation, shaping economics and culture since
for clients to request storage and for storage miners to offer its inception. However, the way this valuable content is dis-
their resources. Bid- and ask orders from those parties are tributed has a significant flaw: HTTP allows content to erode
placed in an orderbook and are therefore publicly available [32]. As represented by error code 404 [33], HTTP fails to
to any network participant. Filecoin states that every honest maintain links between websites and allows valuable content
user has the same view of the orderbook, that is, that the to vanish completely. The main reason for the occurrence of
orderbook is a data-structure incorporated into the blockchain. this problem is centrally managed servers, free to shut down
Hence, orders are added to the blockchain if they are valid. at any given time. While short-term availability of content
As the Storage Market is a verifiable market, orders of type tends to be sufficient, the ongoing erosion of data impedes the
bid, ask and deal can be validated by every participant. effective maintenance of long-term availability and results in a
The only security-related parameter required in the order- vast amount of broken links [34]. HTTP naturally encouraged
matching phase is a field (ts) which describes the duration large organizations to centralize their services which has given
of the validity of a deal order. This prevents a client from rise to an ever-increasing loss of data. As a result, what used
holding back data once the storage miner has committed its to be a decentralized web structure is now becoming more and
resources. The settlement phase involves the sealing of the more centralized.
storage miners sectors and submitting the proofs of storage
generated to the blockchain. [6]
B. Distributed file system
The retrieval markets sole purpose is the retrieval of According to the readme [35]: IPFS is a distributed file
miners to provide data to clients upon request. One very system that seeks to connect all computing devices with the
challenging requirement for this market is the fast retrieval of same system of files. In practice, IPFS is a peer-to-peer
data. Therefore, and unlike the storage market, the retrieval distributed system that connects all networks using the same
market maintains an off-chain orderbook that allows clients system of files. The files are version controlled and represented
and retrieval miners to find each other. The absence of a with a unique hash; as opposed to HTTP, where content is
searched by location, IPFS searches by content. Hence, IPFS Content Addressing (content is uniquely identified by its
is a content-addressable peer-to-peer hypermedia distribution multihash checksum), Tamper resistance (content is verified
protocol. with its checksum) and Deduplication (objects with the exact
same content are equal and are stored only once). [8] The
Nodes, which are incentivized to remain the same or oth- traversal of the Merkle DAG is enabled with UNIX-like paths
erwise lose network benefits, are identified by a cryptographic describing the multi-hashes of the object:
hash of a public-key. IPFS does not rely on a particular hash /ipfs/<hash-of-object>/<name-path-to-object>
function format but instead uses the multihash [36] format such IPFS makes an attempt to reuse long-established file system
that any well-established cryptographic hash function can be properties while handling data according a fundamentally
used in order to create a node. different concept than what was known previously.
The Network then allows nodes to communicate while
Finally, and in order to make IPFS a fully useful file
providing a framework which ensures that any client, no matter
system, IPFS has introduced IPNS [43], the InterPlanetary
what the network stack looks like, can participate. In other
Naming System. The goals are to ensure that the Merkle DAG
words, the client can use any transport protocol and IPFS
contains immutable content-addressed objects, and to apply
provides reliability functions if the underlying network does
Naming by using mutable pointers to the Merkle DAG, thus
not do so. Connectivity is enhanced using ICE NAT traversal
enabling user-friendly naming. [8]
techniques [37] which, for example, also include relaying
in such a way that nodes can find other nodes to provide
demanded content on their behalf. Furthermore, IPFS does not C. Libp2p
rely on IP only but instead uses the multiaddr [38] format to The development of IPFS was a demanding challenge in the
express addresses and their protocols. understanding of the internet network stack. [31] As a result,
Routing is required for nodes to find other peers which the libp2p library was created in order to bundle protocols
can serve objects. A distributed sloppy hash table (DSHT) [39] used for building large scale peer-to-peer applications. Thus,
serves this purpose and is based on S/Kademlia [40] and Coral libp2p can be regarded as an entire network stack, represented
[41]. Again, IPFS remains true to its flexible and modular by a protocol suite. Conceptually, libp2p is structured accord-
structure and thus enables a range of possible implementations ing to Figure 3. 2.
of the IPFSRouting interface, for example, the use of a
regular hash table. [8] in a local environment.
The block exchange procedure for distributing data within
IPFS is implemented with BitSwap [7], a protocol inspired
by BitTorrent [42]. BitSwap is incentivized as it uses a credit
system where peers track their and others balances and send
blocks to debtor peers; this is implemented by means of a
probabilistic approach. In this way, an additional timeout is
applied in the event that a sender does not fulfill their duties,
and this prevents gaming the probabilities. Whereas BitTorrent
uses tit-for-tat strategy, BitSwamp relies on a debt ratio factor:
bytes sent Fig. 3. Libp2p network stack
r=
bytes received + 1
Further, a probabilistic function describes the likelihood of a The interfaces can be described as follows:
peer sending data as follows:
Peer Routing: this identifies the peers to which a mes-
1 sage should be routed. The implementations include
P (send|r) = 1
1 + exp(6 3r) kad-routing, a kademlia routing table where each peer
holds a set of k-buckets; and mDNS-routing, which
One effect of the function is that, once the debt ratio of a identifies local area network peers.
peer surpasses twice the value of their credit, the likelihood of
sending - and hence their trustworthiness - drops radically. As Swarm: this handles everything related to the open-
a result, BitSwap builds a measure of trust using the debt ratio ing of a stream. A stream muxer: this multiplexes
factor and therefore provides resistance against Sybil attacks connections per peer and streams per connection. The
and values successful relationships among peers, yet with the protocol muxer: enables the multiplexing of transport
tolerance of temporary unavailability. [8] protocols on application levels. This allows several
protocols to be muxed in the same socket, in which
In order to effectively store and distribute blocks in quick case only one NAT traversal is required, if any.
and a robust manner, IPFS has introduced the Merkle DAG. Furthermore, Relay provides an end-to-end encrypted
The Directed Acyclic Graph is built up with cryptographic connection where a node asks another node to connect
links to the underlying objects. The object is a data on its behalf, enabling it to overcome NAT traversal.
structure consisting of a name, multihash and size,
thereby permitting any type of data to be represented with. Distributed Record Store: The aim is to store and
The Merkle DAG, described as a generalization of the distribute records; this activity is performed in a sim-
Git data structure [8], hence provides features including: ilar way to DNS as it is used for signaling links, and
TABLE I. E STIMATION OF ECONOMIC FEASIBILITY
announcing peers and content. The record store was
further modularized under the name, InterPlanetary Aspect Rating
Image left after ICO -2
Record Store (IPRS) [44]. The interface provides Overestimating the teams abilities
record stores for abstract records, kademila routing -2
and/or underestimating the cost
records as well as mDNS routing records. Such a Possible hurdles for the use of Filecoin
-3
once ready
record keeping system allows content to be verified Ability to lure miners and customers
by any record store user. away from Storj and Siacoin due to +4
technical excellence
Positioning for marketing opportunities
+3
Discovery: libp2p provides several ways for peers to and business partnerships
find and identify each other in the network. LAN Response of the development team to user
+1
demands
discovery is implemented with mDNS, Random Walk Huge pool of free resources
+2
using a DHT (distributed hash table) discovery, which made available by potential miners
executes random queries, which in turn enable the dis-
covery of peers outside the LAN. Finally, a Bootstrap-
list enables peers to store highly stable (and possibly Our estimation shows that the biggest advantage of the
trusted) peers locally. project is its technical excellence in conjunction with its so-
phisticated marketing and its connectivity with other projects.
On the negative side, we see two major weaknesses: first of
all, the ICO left many investors and upcoming users with
mixed feelings about the project. We support the argument that
the complexity of the implementation is likely to have been
D. Filecoin integration underestimated, and this has given rise to uncertainty; equally
uncertain is the likelihood of users using the project at all.
As at the time of writing, it is not known how Filecoin
plans to adapt IPFS as its underlying data store. Section A. Is it a Ponzi scheme?
VI-B above introduced the components of IPFS briefly and
provided a foundation for Filecoin to take advantage of the Given that Filecoin raised a substantial amount of money
IFPS ecosystem. We consider that the most obvious component with the ICO instrument, it is legitimate to raise the question
to hook in is Bitswap (see Section VI-B). Bitswap manages of whether Filecoin itself is a Ponzi scheme.
requests from peers in the network and therefore is considered By definition, a Ponzi scheme is a fraudulent investment
as the data trading module of IPFS [7]. Essentially, Bitswap operation that pays returns to its investors either from the
acquires blocks requested by the client and initiates a send to investors own money or from money paid by subsequent
the peers who demand these blocks. We believe that, in the case investors, rather than from any actual profit earned by the
of Filecoin, the native measure of trust of Bitswap, as provided company. [46]
by the debt ratio factor, has to change. Instead of relying on the
number of bytes sent and received, Filecoin provides a measure Indeed, in view of the information in Section IV, there are
of trust by relying on FIL token exchange and the guarantees multiple factors which support this argument:
provided by proof-of-spacetime (see Section V-C). Therefore,
in order to distribute the blocks among peers, Bitswap would Full control over investments by Protocol Labs (IV-A)
react to DEAL orders (see Section V-B). Depending on whether Exponentially higher return for early investors (IV-C)
the order emerged from the storage or retrieval market, blocks
would be sent to the node corresponding to either the ask or Reward for vesting (IV)
the bid side. As a result, Bitswap serves as the API used by
Lack of implementation
the decentralized markets (see Section V-D) and handles data
exchange according to Filecoins incentive. On the other hand, there are numerous logical reasons
why an investor would trust the people behind the Filecoin
project: The substantial technical capabilities of Protocol Labs
have already been proven in the building of IPFS and libp2p;
this could lead one to believe that their ability to overcome
VII. I S THE DESIGN ECONOMICALLY FEASIBLE ? the upcoming hurdles in the building of Filecoin will be
handled equally competently. Furthermore, the investments
raised are controlled by an intermediary (see IV), which holds
While analyzing the Filecoin ICO, people have also tried connections to the SEC.
to measure the risk that comes with the investment. Since there
has not yet been any implementation, the investor has to rely on The authors of this paper conclude that the proven technical
what is written in the Whitepaper and consider hypothetically capabilities of Protocol Labs should naturally be reason enough
the pros and cons of the project and the merits of its team to believe that Filecoin is a legitimate project. However, the
members. Therefore, the author of [45] has introduced a simple terms and the way the ICO has proceeded are far from
measure by adding (or subtracting) on a scale of 1 to 5 for ideal. Essentially, Protocol Labs allowed themselves to remain
every hypothetical pro (or con). Table I summarizes this work as the single entity that holds full control over the assets
and extends it with the knowledge accumulated while writing raised, without giving investors the possibility of intervening
this paper. effectively. Therefore, if goodwill fades, the entire project
could turn into a Ponzi scheme. Note: this is a hypothetical economically feasible project firstly by extending the work of
scenario and we are not implying that this scenario will occur. a hypothetical risk analysis, then considering the possibility
We are merely discussing possibilities. of a Ponzi scheme, and finally with a simplified, hypothetical
comparison with Airbnb.
B. A hypothetical comparison
Filecoins economic feasibility is hard to predict and, given
In view of the above, we would like to challenge the reader a simple summary of risk related points, the biggest hurdles are
with speculation drawn from weakly supported statements and probably going to be acceptance from the users. In addition,
some hypothetical numbers. The speculation is meant to be Filecoin shows certain characteristics of a Ponzi scheme but
taken with a pinch of salt, however, it may also serve as an the trust built by the team members in the past leads us to
alert to the excessively bullish investor. believe that it is not one. Filecoins Whitepaper introduces
novel concepts and predecessor projects by the team members
In a recent interview [47], Juan Benet compares Filecoin which prove their technical capabilities. Hence, it appears that
with Airbnb [48], a website where people can rent out their Filecoin is poised to be an outstanding project although it
homes (instead of disk storage). Therefore, the following remains to be seen if it will be adopted by the average cloud
calculation compares the valuation of both companies against storage user.
each other according to their resources. Apartment space for
Airbnb and disk space for Filecoin: Airbnb is valued today
(September 2017) at approximately $31 billion while holding R EFERENCES
around 3 million listings in total [49]. The average apartment [1] Filecoin website, [Online; accessed September 20, 2017]. [Online].
in the United States was 934 square feet in 2016 [50]. In a Available: https://filecoin.io/
hypothetical scenario, Airbnb is therefore valued at $11.06 [2] Tribler project, [Online; accessed September 22, 2017]. [Online].
per square feet. If one compares this number to the median Available: https://www.tribler.org/
price per square foot in the United States, which is $123 [3] B. Wilcox-OHearn, Experiences deploying a large-scale emergent
[51], the Airbnb ecosystem diminishes the median price by network, Peer-to-Peer Systems, pp. 104110, 2002.
a factor of 11.12. The average price for hard drives in 2017 [4] S. Wilkinson, T. Boshevski, J. Brandoff, and V. Buterin, Storj a peer-
is $0.03 per gigabyte [52]. Dividing by the same factor that to-peer cloud storage network, 2014.
Airbnb applies for square feet to the storage price, the average [5] D. Vorick and L. Champine, Sia: Simple decentralized storage, 2014.
gigabyte in the Filecoin system yields $0.0027. This means that [6] Filecoin, [Online; accessed September 20, 2017]. [Online]. Available:
https://filecoin.io/filecoin.pdf
the $2570 0000 000 launched at the Filecoin ICO would require
950 1850 1850 185 gigabytes (950 185 petabytes) of storage to be [7] Bitswap, [Online; accessed October 13, 2017]. [Online]. Available:
https://github.com/ipfs/go-ipfs/tree/master/exchange/bitswap
offered by storage miners. Considering that Dropbox [53]
[8] Ipfs - content addressed, versioned, p2p file system, [Online;
currently holds around 500 petabytes of user data [54], one accessed October 13, 2017]. [Online]. Available: https://ipfs.io/ipfs/
could argue that Filecoin is overvalued. QmR7GSQM93Cx5eAg6a6yRzNde1FQv7uL6X1o4k7zrJa3LX/ipfs.
draft3.pdf
VIII. C ONCLUSION [9] D. Karger, E. Lehman, T. Leighton, R. Panigrahy, M. Levine, and
D. Lewin, Consistent hashing and random trees: Distributed caching
Projects such as Sia and StorJ have attracted attention protocols for relieving hot spots on the world wide web, in Proceedings
by allowing people to provide and rent storage without any of the twenty-ninth annual ACM symposium on Theory of computing.
ACM, 1997, pp. 654663.
central party involved. Filecoins ICO began on August 10,
2017 and raised $257 million, driven by investors who are [10] P. Maymounkov and D. Mazieres, Kademlia: A peer-to-peer informa-
tion system based on the xor metric, in International Workshop on
also sympathetic to the idea of a decentralized storage market. Peer-to-Peer Systems. Springer, 2002, pp. 5365.
This paper now analyzes Filecoins ICO and highlights [11] Tribler issue: blockchain-regulated markets, [Online; accessed
some notable and disturbing facts. For example, early in- September 22, 2017]. [Online]. Available: https://github.com/Tribler/
tribler/issues/2559\#issuecomment-307353664
vestors, including Protocol Labs as the founder team, had to
[12] Icos vs venture capital, [Online; accessed September 28,
pay up to 8.57 times less than later investors. Next, the Simple 2017]. [Online]. Available: http://fingfx.thomsonreuters.com/gfx/rngs/
Agreement for Future Token (SAFT), signed by every investor, USA-VENTURECAPITAL-DIGITALCURRENCY/0100502J05M/
contains terms which strongly favor Protocol Labs. It states index.html
that the invested money, none of which remains in the control [13] Amazon s3, [Online; accessed September 28, 2017]. [Online].
of the investors, is refunded in cases which seem to be unlikely Available: https://aws.amazon.com/s3/
to occur. In addition, many of the terms have been defined in [14] Maidsafe.net announces project safe to the community, [Online;
such a way as to minimize the sums which investors might accessed September 22, 2017]. [Online]. Available: https://github.com/
maidsafe/Whitepapers/blob/master/Project-Safe.md
expect to be returned.
[15] Rule 506 of regulation d, [Online; accessed September
Filecoins technical capabilities, as described in the 22, 2017]. [Online]. Available: https://www.sec.gov/fast-answers/
Whitepaper, have been studied in great detail. We considered answers-rule506htm.html
the details of the InterPlanetary File System (IPFS) project, [16] Coinlist, [Online; accessed September 22, 2017]. [Online]. Available:
https://coinlist.co/
a peer-to-peer hypermedia protocol which is used as the
storage adapter for Filecoinm, with Filecoin serving as the [17] Angellist, [Online; accessed September 22, 2017]. [Online]. Available:
https://angel.co/
incentivized layer. It appears that IPFS can seamlessly serve as
[18] Simple agreement for future tokens, [Online;
a storage adapter for Filecoin by introducing a new strategy for accessed September 25, 2017]. [Online]. Available:
Bitswap whereas the orders define which peers shall serve data. https://coinlist.co/static/media/Protocol\%20Labs\%20-\%20SAFT\
Further, we have raised the question of whether Filecoin is an %20for\%20Filecoin\%20Token\%20Presale.6ddb6fb6.pdf
[19] Filecointokensaleeconomics, [Online; accessed September [44] Interplanetary record store, [Online; accessed October 17, 2017].
25, 2017]. [Online]. Available: https://coinlist.co/static/media/ [Online]. Available: https://github.com/libp2p/interface-record-store
Filecoin-Sale-Economics.0ae9a53f.pdf [45] Filecoin ico analysis, [Online; accessed September 20, 2017].
[20] Filecoin investor faq, [Online; accessed September [Online]. Available: https://hacked.com/ico-analysis-filecoin/
28, 2017]. [Online]. Available: https://ipfs.io/ipfs/ [46] Ponzi scheme, [Online; accessed November 05, 2017]. [Online].
QmWdXyhqHJWJut5wt\\4gSCueTjSnFyDHBy3SRfmcqArtz1a/ Available: https://www.sec.gov/fast-answers/answersponzihtm.html
2017-08-08-Filecoin-Investor-FAQ.html
[47] Podcast - juan benet, [Online; accessed September
[21] S. Nakamoto, Bitcoin: A peer-to-peer electronic cash system, 2008. 28, 2017]. [Online]. Available: https://a16z.com/2017/09/14/
[22] Devcon2, [Online; accessed September 28, 2017]. [Online]. Available: networks-protocols-labs-tokens/
https://ethereumfoundation.org/devcon2/ [48] Airbnb, [Online; accessed September 20, 2017]. [Online]. Available:
[23] G. Wood, Ethereum: A secure decentralised generalised transaction https://airbnb.com
ledger, Ethereum Project Yellow Paper, vol. 151, 2014. [49] Airbnb statistics and facts, [Online; accessed September 28,
[24] Erc-20 token standard, [Online; accessed September 28, 2017]. [Online]. Available: https://expandedramblings.com/index.php/
2017]. [Online]. Available: https://github.com/ethereum/EIPs/blob/ airbnb-statistics/
master/EIPS/eip-20-token-standard.md [50] Airbnb just closed a $1 billion round and became
profitable in 2016, [Online; accessed September 28,
[25] S. Kamara, K. E. Lauter et al., Cryptographic cloud storage. in
2017]. [Online]. Available: https://www.cnbc.com/2017/03/09/
Financial Cryptography Workshops, vol. 6054. Springer, 2010, pp.
airbnb-closes-1-billion-round-31-billion-valuation-profitable.html
136149.
[51] U.s. home sellers in march 2016 realized highest home
[26] G. Ateniese, R. Burns, R. Curtmola, J. Herring, L. Kissner, Z. Peterson,
price gains since december 2007, [Online; accessed September
and D. Song, Provable data possession at untrusted stores, in Proceed-
28, 2017]. [Online]. Available: http://www.realtytrac.com/news/
ings of the 14th ACM conference on Computer and communications
home-prices-and-sales/march-q1-2016-home-sales-report/
security. Acm, 2007, pp. 598609.
[52] Hard drive cost per gigabyte, [Online; accessed September
[27] H. Shacham and B. Waters, Compact proofs of retrievability. in 28, 2017]. [Online]. Available: https://www.backblaze.com/blog/
Asiacrypt, vol. 5350. Springer, 2008, pp. 90107. hard-drive-cost-per-gigabyte/
[28] E. B. Sasson, A. Chiesa, C. Garman, M. Green, I. Miers, E. Tromer, and [53] Dropbox, [Online; accessed September 20, 2017]. [Online]. Available:
M. Virza, Zerocash: Decentralized anonymous payments from bitcoin, https://dropbox.com
in Security and Privacy (SP), 2014 IEEE Symposium on. IEEE, 2014,
pp. 459474. [54] Scaling to exabytes and beyond, [Online; accessed September
28, 2017]. [Online]. Available: https://blogs.dropbox.com/tech/2016/
[29] Power fault tolerance, [Online; accessed October 31, 2017]. [Online]. 03/magic-pocket-infrastructure/
Available: https://filecoin.io/power-fault-tolerance.pdf
[30] J. Poon and T. Dryja, The bitcoin lightning network: Scalable off-chain
instant payments, Technical Report (draft), 2015.
[31] libp2p: Modular peer-to-peer networking stack, [Online; accessed
October 31, 2017]. [Online]. Available: https://github.com/libp2p/specs
[32] Http is obsolete. its time for the distributed, permanent web, [Online;
accessed October 11, 2017]. [Online]. Available: https://ipfs.io/
ipfs/QmNhFJjGcMPqpuYfxL62VVB9528NXqDNMFXiqN5bgFYiZ1/
its-time-for-the-permanent-web.html
[33] N. W. Group et al., Rfc 2616: Hypertext transfer protocolhttp/1.1,
R. Fielding, J. Gettys, J. Mogul, H. Frystyk, L. Masinter, P. Leach, T.
Berners-Lee, 1999.
[34] J. Markwell and D. W. Brooks, Broken links: The ephemeral nature
of educational www hyperlinks, Journal of Science Education and
Technology, vol. 11, no. 2, pp. 105108, 2002.
[35] Ipfs - the permanent web, [Online; accessed October 11, 2017].
[Online]. Available: https://github.com/ipfs/ipfs/
[36] Self describing hashes - for future proofing, [Online; accessed
October 11, 2017]. [Online]. Available: https://github.com/multiformats/
multihash
[37] J. Rosenberg, Interactive connectivity establishment (ice): A method-
ology for network address translator (nat) traversal for offer/answer
protocols, 2010.
[38] The network address multiformat, [Online; accessed October 11,
2017]. [Online]. Available: https://github.com/multiformats/multiaddr
[39] M. J. Freedman and D. Mazieres, Sloppy hashing and self-organizing
clusters, in International Workshop on Peer-to-Peer Systems. Springer,
2003, pp. 4555.
[40] I. Baumgart and S. Mies, S/kademlia: A practicable approach towards
secure key-based routing, in Parallel and Distributed Systems, 2007
International Conference on. IEEE, 2007, pp. 18.
[41] M. J. Freedman, E. Freudenthal, and D. Mazieres, Democratizing
content publication with coral. in NSDI, vol. 4, 2004, pp. 1818.
[42] B. Cohen, Incentives build robustness in bittorrent, in Workshop on
Economics of Peer-to-Peer systems, vol. 6, 2003, pp. 6872.
[43] The inter-planetary naming system, [Online; accessed October
13, 2017]. [Online]. Available: https://github.com/ipfs/examples/tree/
master/examples/ipns