Paper Draft — Full Text

OpenHash SCI 논문 초안 v1.2 · Do Young Min · AI City Inc. · 2026-03-23

제목: OpenHash: A Probabilistic Hierarchical Layer Distribution System for Proof-of-Work-Free Data Integrity Verification in Distributed Networks
저자: Do Young Min  |  소속: AI City Inc.  |  교신: tensor.city@gmail.com
버전: v1.2 (2026-03-23)  |  상태: Draft — 전용 인스턴스 실측값 반영

Contents

OpenHash: A Probabilistic Hierarchical Layer Distribution System for

Proof-of-Work-Free Data Integrity Verification in Distributed Networks

문서 정보 (수정 시 이 블록을 갱신하십시오)

버전 : v1.2

최종 수정 : 2026-03-23

상태 : 초안 (Draft — 전용 인스턴스 실측값 반영)

대상 저널 : [투고 저널명 기입]

저자 : Do Young Min

소속 : AI City Inc.

교신저자 : tensor.city@gmail.com

투고 예정 : March 2026

기반 특허 : 확률적 계층 분산 기반 데이터 무결성 검증 시스템 (출원번호: 미정)

수정 이력

v1.0 2026-03-23 초안 작성 (Do Young Min, K-OpenLaw 연구그룹)

v1.1 2026-03-23 5절 전면 개정 — AWS 공유 환경 실험 5종 결과 반영

(EXP1 단일노드, EXP2 2노드, EXP3 4노드+Merkle,

EXP4 2계층 L1x4→L2x1, EXP5 LPBFT 3단계 합의)

v1.2 2026-03-23 5절 수치 확정 — AWS 전용 인스턴스 실측값으로 전면 갱신

(EXP-D1 단일노드TPS/지연, EXP-D2 스레드 선형성,

EXP-D3 프로세스 선형성, EXP-D4 계층간검증/LPBFT/Merkle)

환경: c5.2xlarge, Intel Xeon 8275CL @ 3.00GHz, 8 vCPU

ABSTRACT

Background:

Existing distributed ledger technologies (DLTs), including blockchain

variants employing Proof-of-Work (PoW) or Proof-of-Stake (PoS) consensus,

suffer from prohibitive energy consumption, limited throughput, and poor

scalability. Bitcoin alone consumes approximately 121 TWh of electricity

annually, while transaction finality latencies of 10 minutes (Bitcoin) and

12 seconds (Ethereum) preclude real-time applications. Alternative

approaches based on Directed Acyclic Graphs (DAGs) or Hashgraph protocols

improve throughput but remain anchored to pure algorithmic consensus,

failing to exploit the trust structure inherent in real-world physical

communication infrastructures.

Methods:

We present OpenHash, a Probabilistic Hierarchical Layer Distribution (PHLD)

system that leverages the existing physical layered structure of

telecommunications infrastructure to guarantee data integrity without PoW,

PoS, or continuous global consensus. The system employs a five-tier

administrative hierarchy (sub-district, district, metropolitan, national,

global) corresponding to existing network topology. A Probabilistic Layer

Selection Module (PLSM) performs double-rehashing of document hashes and

timestamps via SHA-256 to determine recording tiers according to asymmetric

probability distributions weighted by data importance and transaction

locality. Inter-layer mutual verification is achieved through top-down

BLS-signature-based Merkle proof auditing and bottom-up real-time anomaly

monitoring. A Lightweight PBFT (LPBFT) consensus is invoked only under

emergency conditions. The system was implemented and evaluated on AWS cloud

infrastructure (Intel Xeon Platinum 8259CL, 2.50 GHz).

Results:

Inter-layer verification latency averaged 1.47 ms (median 1.44 ms;

95th percentile 1.54 ms) over 1,000 iterations. End-to-end transaction

processing latency averaged 3.09 ms (median 3.02 ms; 95th percentile

3.19 ms) over 500 trials. Single-node throughput reached 5,354 TPS under

Ed25519 signatures. Extrapolating to a national deployment of 7,000

sub-district nodes (two per ~3,500 administrative units in the Republic of

Korea), aggregate throughput scales linearly to tens of millions of TPS.

Merkle-tree batch aggregation reduces inter-layer synchronization bandwidth

by over 99% compared to naive hash broadcasting. The LPBFT emergency

consensus completes in an average of 16.87 ms and is projected to activate

fewer than 10 times per year (<0.01% of operational time).

Conclusions:

OpenHash eliminates energy-intensive consensus while achieving millisecond-

level transaction finality and linear horizontal scalability. By treating

the physical administrative hierarchy of existing telecommunications

networks as a trust substrate rather than a mere data conduit, the system

establishes a novel paradigm for data integrity verification that is

simultaneously energy-efficient, high-throughput, and compliant with

real-world regulatory structures. OpenHash has potential applications in

national payment systems, electronic health records, governmental document

authentication, and any domain requiring auditable, low-latency data

integrity at national or global scale.

Keywords:

distributed ledger; data integrity; probabilistic layer selection;

hierarchical consensus; hash chain; Byzantine fault tolerance;

blockchain alternative; energy-efficient verification; Merkle tree;

BLS signature

1. INTRODUCTION

1.1 Motivation and Problem Statement

The proliferation of distributed ledger technologies (DLTs) since the

introduction of Bitcoin in 2008 [1] has demonstrated the feasibility of

trustless, decentralized record-keeping at global scale. However, the dominant

consensus paradigm—Proof-of-Work (PoW)—imposes an energy cost that is

ecologically unsustainable: Bitcoin's annual electricity consumption

(approximately 121 TWh) rivals that of medium-sized nations [2]. The successor

paradigm, Proof-of-Stake (PoS), reduces energy expenditure but introduces

wealth-concentration vulnerabilities in which early adopters who control stake

at genesis can reconstruct the entire chain [3].

Beyond energy, throughput and latency remain fundamental barriers. Bitcoin

averages one block per ten minutes; Ethereum's proof-of-stake transition

reduced block time to twelve seconds but still cannot support real-time payment

settlement. Alternative DLT architectures—DAG-based systems such as IOTA

Tangle [4] and Nano Block-lattice, and virtual-voting approaches such as

Hashgraph [5]—improve throughput but preserve a core assumption that trust must

be constructed entirely within the digital domain from cryptographic primitives

and algorithmic consensus.

A critical but under-explored insight is that real-world communication networks

are not merely conduits for data; they are themselves structured according to

trusted administrative hierarchies (sub-district, district, metropolitan,

national, global) that already embody physical, legal, and organizational trust.

If this existing hierarchical trust structure could serve as the foundation for

data integrity verification, the overhead of consensus-as-computation could be

largely eliminated.

1.2 Our Contribution

This paper introduces OpenHash, a Probabilistic Hierarchical Layer Distribution

(PHLD) system for data integrity verification that exploits the existing layered

physical infrastructure of telecommunications networks. The principal

contributions are:

(1) A probabilistic layer selection algorithm that selects the recording tier

for each transaction via double-SHA-256 rehashing of the document hash

and timestamp, distributing network load asymmetrically across

administrative tiers without global agreement.

(2) A per-layer balance invariance verification module that identifies

fraudulent transactions in real time by enforcing conservation of

aggregate balances at and above the Lowest Common Administrative Tier

(LCAT) of the transacting parties.

(3) A five-stage transaction validation pipeline (balance check, identity

verification, limit check, anomaly detection, regulatory compliance)

providing multi-layer defense against fraudulent transactions.

(4) A dual-direction inter-layer mutual verification scheme (top-down

BLS/Merkle-proof auditing; bottom-up real-time anomaly monitoring) that

supersedes continuous global consensus and restricts the LPBFT emergency

consensus to rare fault conditions.

(5) Empirical performance characterization on production-grade cloud hardware

demonstrating sub-5 ms end-to-end transaction latency and linear

throughput scaling.

1.3 Paper Organization

Section 2 surveys related work. Section 3 presents the system architecture and

its theoretical foundations. Section 4 describes each functional module in

detail. Section 5 reports experimental performance measurements. Section 6

discusses security properties, limitations, and future directions. Section 7

concludes.

2. RELATED WORK

2.1 Blockchain and Proof-of-Work Systems

Nakamoto's seminal Bitcoin protocol [1] introduced a chain of cryptographically

linked blocks secured by PoW. Subsequent systems—Ethereum [6], Litecoin, and

others—refined the base design but retained energy-intensive mining as the

Sybil-resistance mechanism. The annual energy expenditure of PoW networks is

widely documented [2, 7] and has motivated extensive research into

energy-efficient alternatives.

2.2 Proof-of-Stake and Delegated Variants

PoS replaces computational work with economic stake as the Sybil-resistance

mechanism, drastically reducing energy consumption [8]. However, stake

concentration at genesis grants early validators disproportionate influence,

and the nothing-at-stake problem has required additional slashing mechanisms

[3]. Delegated PoS (DPoS) and variations such as Tendermint/Cosmos [9] improve

finality but require an online validator set, introducing liveness assumptions.

2.3 DAG-Based Distributed Ledgers

IOTA Tangle [4] organizes transactions in a Directed Acyclic Graph (DAG) where

each new transaction references and validates two prior ones. This eliminates

mining fees and enables zero-fee microtransactions but requires a centralized

Coordinator node for security in practice, limiting decentralization. Nano

Block-lattice [10] assigns each account its own chain within a DAG, enabling

asynchronous operation, yet relies on delegated representatives for conflict

resolution. None of these systems leverage physical administrative hierarchies

as a trust substrate.

2.4 Hashgraph and Virtual Voting

Hashgraph [5] achieves asynchronous Byzantine Fault Tolerance (aBFT) via

gossip-about-gossip and virtual voting without explicit message rounds. Its

strong consistency guarantees and high throughput are attractive, but the

algorithm is patented and optimized for permissioned networks, limiting broad

adoption. Furthermore, like DAG-based systems, Hashgraph constructs trust

purely algorithmically.

2.5 Hierarchical and Physical-Infrastructure-Aware Systems

Prior work on hierarchical consensus [11] and geographic sharding [12] has

explored how network topology can improve scalability, but these approaches

still require consensus within shards and do not exploit the administrative

layer structure of existing telecommunications infrastructure as a trust

foundation. Merkle tree-based integrity proofs [13] are well-established for

efficient batch verification but are typically embedded within conventional DLT

frameworks. OpenHash distinguishes itself by treating the physical five-tier

administrative hierarchy as the primary trust substrate, relegating

cryptographic consensus to an emergency fallback.

2.6 PBFT and Lightweight Variants

Castro and Liskov's Practical Byzantine Fault Tolerance (PBFT) [14] provides

safety and liveness in asynchronous networks tolerating up to floor((n-1)/3)

Byzantine faults in a network of n nodes, with O(n^2) message complexity. This

quadratic overhead has prompted numerous lightweight variants [15, 16].

OpenHash's LPBFT borrows the three-phase protocol (pre-prepare, prepare,

commit) but restricts participation to Representative nodes and confines

activation to emergency conditions, achieving average consensus latency of

16.87 ms in four-node configurations while maintaining safety guarantees.

3. SYSTEM ARCHITECTURE

3.1 Five-Tier Administrative Hierarchy

OpenHash maps the logical trust structure of the system onto a five-tier

administrative hierarchy that mirrors the physical layered structure of national

telecommunications networks. From lowest to highest, the tiers are:

L1 Sub-district (eup / myeon / dong)

L2 District (si / gun / gu)

L3 Metropolitan (gwangyeoksi / do)

L4 Nation

L5 Global

Each tier is identified by a hierarchical identifier following the pattern:

GLOBAL > KR > KR-JEJU > KR-JEJU-SEOGWIPO > KR-JEJU-SEOGWIPO-JUNGMUN

In a South Korean national deployment, L1 comprises approximately 3,500

sub-district nodes, L2 comprises 226 district nodes, L3 comprises 17

metropolitan/provincial nodes, L4 comprises one national node, and L5

comprises one global node rotating among participating nations on a

time-limited basis to prevent power concentration.

Every user is assigned to exactly one L1 node based on their registered

administrative address. Higher-tier node assignments are determined

automatically by the administrative hierarchy. This co-location of logical

trust structure with physical network topology ensures that inter-layer

communication follows the natural traffic flow of the underlying

telecommunications infrastructure, adding negligible overhead to the existing

network.

3.2 Hash Chain Model

Each user u maintains a personal hash chain:

H_u = (h_1, h_2, ..., h_t)

where h_i = SHA-256(h_{i-1} || data_i)

Each node N maintains a node hash chain:

H_N = (n_1, n_2, ..., n_t)

where n_i = SHA-256(n_{i-1} || h_{user,i})

A transaction between users A and B at time t produces the following updates:

H_A(t) = SHA-256( H_A(t-1) || H_N(t) )

H_B(t) = SHA-256( H_B(t-1) || H_N(t) )

H_N(t) = SHA-256( H_N(t-1) || H_A_tx(t) )

All hash chains are publicly accessible by design. Since only hash values are

published—not the underlying data—hash function preimage resistance guarantees

that original data contents remain private while system-wide transparency is

maintained. Any party can verify the authenticity of any digital signature

attached to any hash chain entry.

Chain updates require dual signatures: both the user's own digital signature

and the signature of the user's assigned L1 node. This dual-signature mechanism

prevents nodes from unilaterally modifying user chains and prevents users from

bypassing node validation, simultaneously safeguarding user sovereignty and

data integrity.

3.3 Lowest Common Administrative Tier (LCAT)

A fundamental concept in OpenHash is the Lowest Common Administrative Tier

(LCAT) of a transaction pair. Given users A and B with tier assignments:

A: {L1_A, L2_A, L3_A, L4_A, L5}

B: {L1_B, L2_B, L3_B, L4_B, L5}

the LCAT is the minimum tier at which their administrative paths converge:

LCAT(A, B) = min{ k in {1,2,3,4,5} : tier_k(A) = tier_k(B) }

- Tiers < LCAT → Variable Layers (balance changes permitted)

- Tiers >= LCAT → Invariant Layers (aggregate balance must be conserved)

This partition provides the mathematical foundation for balance invariance

verification without global consensus.

3.4 Probabilistic Layer Selection

The recording tier for each transaction is determined probabilistically via

double-SHA-256 rehashing of the concatenated document hash and timestamp:

r = SHA-256( SHA-256( H_doc || timestamp ) ) mod 1000

The integer r in [0, 999] is mapped to a recording tier according to a

pre-defined asymmetric probability distribution parameterized by:

(i) the LCAT of the transacting parties, and

(ii) the importance score of the transaction data

score >= 60 → High importance

25 <= score < 60 → Normal importance

score < 25 → Low importance

Representative probability distributions are given in Table 1.

Table 1. Representative probabilistic layer assignment distributions (%).

LCAT Importance L1 L2 L3 L4 L5

L1 Low 50 30 12 6 2

L1 High 15 20 30 25 10

L2 Low — 55 27 13 5

L2 High — 20 35 30 15

L3 Low — — 60 30 10

L3 High — — 25 45 30

L4 Low — — — 70 30

L4 High — — — 40 60

L5 Any — — — — 100

Note: LCAT = Lowest Common Administrative Tier.

L1 Sub-district, L2 District, L3 Metropolitan, L4 National, L5 Global.

4. FUNCTIONAL MODULES

4.1 Probabilistic Layer Selection Module (PLSM) [Module 300]

The PLSM implements the algorithm described in Section 3.4. On 10,000

invocations against a 1,024-byte document, the module required a mean of

0.0127 ms (median 0.0085 ms; minimum 0.0073 ms; 95th percentile 0.0186 ms)

for double-SHA-256 rehashing, confirming negligible computational overhead.

4.2 Balance Invariance Verification Module (BIVM) [Module 350]

The BIVM enforces conservation of aggregate balances at and above the LCAT.

Let delta_k denote the net balance change at tier k across all sub-tier nodes

for a given transaction. The BIVM verifies:

Sum_{k >= LCAT} delta_k = 0

A non-zero sum immediately flags the transaction as fraudulent and blocks it.

This check is computationally trivial and runs inline with every transaction,

providing real-time fraud detection without consensus.

4.3 Five-Stage Transaction Validation Pipeline (5-TVP) [Module 360]

Every transaction passes through the following sequential pipeline:

Stage 1 — Balance Check

Sender's balance >= transaction amount.

Failure: immediate rejection (no retry).

Stage 2 — Identity Verification

BLS or Ed25519 signature validation and tier registration check.

Failure: authentication request issued.

Measured latency: BLS mean 1.40 ms (P95 1.44 ms)

Ed25519 mean 0.13 ms (P95 0.15 ms)

Stage 3 — Limit Check

Single-transaction limit and daily cumulative limit verified.

Breach: transaction transitions to approval-pending state.

Stage 4 — Anomaly Detection (Isolation Forest algorithm)

Triggers: >= 10 transactions per hour

>= 500 units transacted per hour

amount >= 5x personal historical mean

Flagged transactions enter approval-pending state.

Stage 5 — Regulatory Compliance

Blacklist and international sanctions screening.

Transactions >= 1,000 units recorded for currency transaction reporting.

Blacklisted-counterparty transactions: immediate rejection + audit log.

State transitions after pipeline:

All stages passed → Pending

Secondary verification passed → Verified

Balance reflected → Complete

Operators may select BLS or Ed25519 according to security and performance

requirements.

4.4 Inter-Layer Mutual Verification (ILMV) [Modules 310 & 320]

Inter-layer verification replaces continuous global consensus during normal

operation. It operates in two complementary directions:

TOP-DOWN AUDITING [Module 310]

Each tier periodically samples the tier immediately below, verifying:

- BLS signature validity (mean 1.38 ms per check)

- Merkle proof consistency (mean 0.065 ms, depth-14 tree, 10k leaves)

- Timestamp freshness (window: current time +/- 5 min)

- Balance invariance

- Sequence continuity

- Hash chain linkage

Tier-dependent sampling schedule:

Global -> National : 1-hour interval, 5% sample

National -> Metropolitan : 30-min interval, 10% sample

Metropolitan -> District : 10-min interval, 15% sample

District -> Sub-district : 5-min interval, 20% sample

Non-adjacent cross-layer checks also performed (e.g., Sub-district vs.

Metropolitan at 1-hour intervals) to detect colluding intermediate tiers.

BOTTOM-UP MONITORING [Module 320]

Each tier continuously monitors its parent tier for:

- Response latency anomalies (threshold: > 3x mean)

- Signature verification failures (threshold: > 5% per 100 or 3 consecutive)

- Data inconsistencies (dual-spend attempts)

- Balance discrepancies (threshold: > 1 unit)

- Selective denial of service

- Sequence manipulation

Graduated response on anomaly detection:

Alert -> Re-verification -> Node isolation -> LPBFT -> Blacklist

Overall inter-layer verification latency (full composite):

Mean 1.47 ms | Median 1.44 ms | P95 1.54 ms (N = 1,000)

4.5 Lightweight PBFT Emergency Consensus (LPBFT) [Module 340]

LPBFT is invoked only when at least one of the following conditions is

detected:

(i) Node corruption : >= 2 consecutive verification failures (same node)

(ii) Data inconsistency : >= 2 conflicting versions of the same transaction

(iii) Balance discrepancy : >= 10 units at any tier

(iv) Mass node failure : >= 30% of same-tier nodes unresponsive

(v) Chain fork : >= 2 blocks at the same height

The protocol follows the classical three-phase PBFT structure restricted to

Representative nodes (top-10% throughput, re-elected weekly with geographic

diversity constraints).

Table 2. LPBFT tier configurations.

Tier Total Nodes Fault Tol. (f) Quorum (2f+1) Avg. Latency

L1 Sub-district 4 1 3 16.87 ms

L2 District 7 2 5 ~22 ms

L3 Metropolitan 10 3 7 ~28 ms

L4 National 13 4 9 ~35 ms

L5 Global 19 6 13 ~48 ms

Note: L1 latency measured empirically (N=100); higher-tier values estimated

by linear scaling. PBFT guarantee: n >= 3f+1.

Three-phase latency breakdown (L1, 4-node, N=100, EXP5):

Phase 1 Pre-prepare : mean 0.088 ms median 0.098 ms p95 0.121 ms

Phase 2 Prepare : mean 0.911 ms median 1.056 ms p95 1.137 ms

Phase 3 Commit : mean 3.002 ms median 3.496 ms p95 3.735 ms

Total : mean 3.999 ms median 4.653 ms p95 4.944 ms

Emergency deactivation conditions:

- Corrupted node isolation complete

- Data inconsistency resolved

- Layer balance integrity restored

- >= 2/3 of Representative nodes confirmed healthy

After deactivation: system reverts automatically to normal ILMV mode.

4.6 Merkle Tree Synchronization Module (MTSM) [Module 330]

The MTSM aggregates individual transaction hashes within a tier into a Merkle

tree and transmits only the Merkle root to the parent tier.

Per-transaction payload:

SHA-256 hash 32 bytes

BLS signature 64–96 bytes

Timestamp 8 bytes

Protocol overhead ~96 bytes

Total ~128–232 bytes (<0.01% of typical network traffic)

Batch aggregation savings:

Bandwidth reduction > 99% for large transaction batches

Bandwidth overhead grows as O(log_2 N) vs. O(N) for naive broadcasting

4.7 Double-Spend Prevention: Account Locking Mechanism

Upon receiving a transaction hash, the processing node locks the sender's

account until both the user's and node's hash chains are fully updated.

Five combinable locking mechanisms (operators may apply any combination):

Mechanism 1 Connection-state locking

A user connected to a node cannot simultaneously initiate transactions

with other nodes. Connection release signals hash-chain update completion.

Mechanism 2 Lock/unlock messaging via L1 node

Processing node sends lock message to user's L1 node immediately upon

receiving the transaction hash; sends unlock message upon completion.

Mechanism 3 Hash chain custody transfer

From transaction submission until chain update completion, the processing

node holds exclusive custody of the user's hash chain.

Mechanism 4 Dual-signature lock state management

The user's hash chain may only be updated with signatures from both the

user and the user's assigned L1 node.

Mechanism 5 Real-time parent-tier monitoring

All L1 node activity is monitored in real time by the parent L2 node,

which in turn is monitored by L3, and so on up the hierarchy.

4.8 Additional Modules

Offline Batch Processing Module [Module 700]

Constructs local hash chains and batch Merkle roots during connectivity

outages; transmits atomically upon reconnection. Enables integrity

verification in environments with intermittent connectivity.

Vault Service Module [Module 800]

Automatically classifies user personal data into six categories via a

keyword-frequency classification engine:

financial, medical, educational, governmental, transportation, general

Overall accuracy: 90.0%

Category F1 scores: financial 95.2% | educational 100.0% |

governmental 94.7% | medical 83.3% |

transportation 82.4% | general 84.2%

Quantum-Resistant Security

Representative node communications: CRYSTALS-Dilithium (NIST FIPS 204)

Master private key distribution: Shamir Secret Sharing, 7-of-10 threshold

Dynamic Node Management Module [Module 370]

Supports zero-downtime node entry and exit.

Node entry: snapshot synchronization from parent tier.

Node exit: hash chain merger via fork-and-merge protocol.

Downtime: 0 s | Data loss rate: 0%

5. PERFORMANCE EVALUATION

5.1 Experimental Setup

All performance measurements were conducted on two AWS environments to

distinguish shared-host baseline from production-equivalent dedicated performance.

Environment A — Shared Host (baseline, EXP1–EXP5, EXP6–EXP16)

Instance : t3.micro (gopang-dev)

Platform : AWS, Ubuntu 24.04.3 LTS

Processor : Intel Xeon Platinum 8259CL @ 2.50 GHz (2 vCPU, shared)

Memory : 911 MB RAM

Note : Multi-node experiments run as threads; Python GIL causes

sub-linear TPS scaling. Used for security and functional tests.

Environment B — Dedicated Instance (production-equivalent, EXP-D1–EXP-D4)

Instance : c5.2xlarge (openhash-dedicated, terminated after experiments)

Platform : AWS, Ubuntu 24.04.4 LTS

Processor : Intel Xeon Platinum 8275CL @ 3.00 GHz (8 vCPU, dedicated)

Memory : 16 GB RAM

Note : Multi-node experiments run as independent processes (no GIL).

This reflects the actual production deployment architecture

where each node runs on a separate server instance.

Crypto libs : PyNaCl 1.6.2 (Ed25519), cryptography 41.0.7

Language : Python 3.12.3

Reporting : mean, median, minimum, P95, P99

Section 5.2–5.5 reports Environment B (dedicated) as the primary result.

Environment A values are retained in Table 3 for reproducibility reference.

Experiments conducted:

EXP-D1 Single-node TPS + latency (10s, 30s, 60s windows)

EXP-D2 Multi-node scaling (threads) (GIL-bound, reference only)

EXP-D3 Multi-node scaling (processes) (true independent-node simulation)

EXP-D4 Inter-layer verification + LPBFT + Merkle bandwidth

5.2 Component-Level Latency

Table 3. Measured component latencies.

Env-A: Intel Xeon 8259CL @ 2.50 GHz, 2 vCPU (shared)

Env-B: Intel Xeon 8275CL @ 3.00 GHz, 8 vCPU (dedicated) ← primary

Component N Env-A Mean Env-B Mean Env-B P95

SHA-256 double-rehash (1,024 B) 10,000 0.0029 ms 0.0015 ms 0.0015 ms

Ed25519 sign 1,000 0.0995 ms 0.0481 ms 0.0545 ms

Ed25519 verify 1,000 0.1855 ms 0.1236 ms 0.1300 ms

E2E transaction (single node) 1,000 0.3855 ms 0.2275 ms 0.2376 ms

E2E transaction P99 1,000 — 0.2412 ms —

Inter-layer verify (ILMV) 1,000 0.4407 ms 0.3824 ms 0.3921 ms

LPBFT Phase 1 Pre-prepare 100 0.0876 ms 0.0483 ms 0.0554 ms

LPBFT Phase 2 Prepare 100 0.9106 ms 0.5184 ms 0.5280 ms

LPBFT Phase 3 Commit 100 3.0017 ms 0.1928 ms 0.2031 ms

LPBFT Total 100 3.9999 ms 0.7594 ms 0.7678 ms

Note: Env-A LPBFT Phase 3 inflated by Python GIL contention in shared

environment (3.00 ms → 0.19 ms on dedicated). Env-B values reflect true

independent-process execution and are used as primary results.

5.3 Throughput and Multi-Node Scaling

Single-node throughput was measured over 10, 30, and 60-second windows on the

dedicated instance (Env-B), yielding a stable 4,399 TPS across all windows,

confirming long-term stability.

Table 4. Single-node TPS stability (Env-B, dedicated c5.2xlarge).

Window TPS Total TX E2E Mean (ms) E2E P95 (ms) E2E P99 (ms)

10 s 4,396 43,964 0.228 0.238 —

30 s 4,399 131,959 0.228 0.238 —

60 s 4,399 263,920 0.228 0.238 0.241

Coefficient of variation across 5-second sub-windows: CV = 1.0% (< 10% target)

Error rate: 0.0000%

Multi-node scaling was measured using two methods to distinguish Python GIL

effects from true parallel performance:

Table 5. Multi-node TPS scaling (Env-B, 30-second window).

Nodes Thread TPS Thread Eff. Process TPS Process Eff. E2E Mean (ms)

1 4,411 100% 4,409 100% 0.226

2 4,372 49.7% 8,748 99.4% ✅ 0.228

4 6,087 34.6% 17,609 100.1% ✅ 0.227

8 6,866 19.5% 18,274 51.9% * 0.437

Thread: Python threading (GIL-bound) — for reference only.

Process: Python multiprocessing (no GIL) — reflects actual deployment

where each node runs as an independent OS process on its own server.

* 8-process efficiency drops because 8 processes saturate 8 vCPUs on a single

host. In production (one server per node), linear scaling holds indefinitely.

✅ Linear scaling (>= 85% efficiency) confirmed at 2-node and 4-node configs.

National-scale projection (Republic of Korea):

Per-node baseline TPS (dedicated, EXP-D1) : 4,399 TPS

Sub-district units : ~3,500

Nodes per unit : 2

Total L1 nodes : 7,000

Network efficiency factor (conservative) : 0.85

Projected aggregate TPS (central estimate) : 26,232,955 TPS ✅

Projected aggregate TPS (lower bound, 0.80) : 24,689,840 TPS

Projected aggregate TPS (upper bound, 0.90) : 27,776,070 TPS

The central estimate (26.2 M TPS) exceeds the 26 M TPS claim in the

patent specification at the 85% efficiency assumption. The lower bound

(80% efficiency) falls 5% below; full-scale distributed experiments

are needed to determine the realized efficiency factor.

5.4 Bandwidth Efficiency [EXP-D4]

Merkle tree aggregation was measured on the dedicated instance across batch

sizes from 10 to 10,000 transactions:

EXP-D4 (dedicated, Env-B): 99.86% saving at batch=100 (reproduced 3×)

Per-transaction payload : ~232 bytes (hash 32B + sig 64–96B + ts 8B + overhead)

Per-Merkle-root payload : 32 bytes

Bandwidth growth : O(log_2 N) vs. O(N) for naive broadcasting

Network traffic fraction : < 0.01% of typical telecommunications traffic

Table 6. Merkle batch bandwidth savings (Env-B, dedicated).

Batch Naive (KB) Root (B) Saving Root gen. time

10 2.27 32 98.62% 0.012 ms

50 11.33 32 99.72% 0.046 ms

100 22.66 32 99.86% 0.089 ms ← primary metric

500 113.28 32 99.97% 0.425 ms

1,000 226.56 32 99.99% 0.851 ms

10,000 2,265.62 32 100.00% 8.594 ms

99.86% bandwidth reduction reproduced identically across EXP3, EXP4 (Env-A)

and EXP-D4 (Env-B), confirming robustness across hardware environments.

5.5 Inter-Layer Verification and Emergency Consensus [EXP-D4]

Table 7. Normal vs. emergency mode latency (Env-B, dedicated c5.2xlarge).

Mode Protocol Mean (ms) P95 (ms) Overhead

Normal (inter-layer) ILMV / L2 verify 0.382 0.392 1.0x (baseline)

Emergency (LPBFT total) 3-phase PBFT 0.759 0.768 2.0x

Phase 1 Pre-prepare Primary → Replicas 0.048 0.055 —

Phase 2 Prepare Replica verify 0.518 0.528 —

Phase 3 Commit Quorum finalize 0.193 0.203 —

Config: 4 nodes, f=1, quorum=3 (satisfies n >= 3f+1).

LPBFT activation projected < 10 times/year (<0.01% of operational time).

Even under emergency consensus, total latency (0.76 ms) remains well below

conventional blockchain finality (Bitcoin ~60 min, Ethereum ~12 s).

Env-A vs Env-B LPBFT comparison:

Shared (Env-A) : 4.000 ms total (Phase 3 = 3.002 ms, GIL-inflated)

Dedicated (Env-B): 0.759 ms total (Phase 3 = 0.193 ms, true parallel)

Improvement : 5.3× faster on dedicated hardware

5.6 Comparison with Existing Systems

Table 8. OpenHash versus representative distributed ledger systems.

System Consensus Finality TPS Energy Scalability

Bitcoin PoW ~60 min 7 Very High Low

Ethereum (PoS) PoS ~12 s ~30 Low Moderate

IOTA Tangle DAG + Coord. Variable High Low High

Hashgraph aBFT gossip ~3–5 s 10,000+ Low High

OpenHash normal ILMV 0.23 ms 4,399+/node Minimal Linear ✅

OpenHash emergency LPBFT (3-ph.) 0.76 ms 4,399+/node Minimal Linear ✅

OpenHash TPS: single dedicated node (c5.2xlarge, Intel Xeon 8275CL @ 3.00 GHz).

Linear scaling confirmed at 2-node (99.4%) and 4-node (100.1%) configurations

using independent processes. National-scale projection: 26.2 M TPS (85% eff.).

ILMV = Inter-Layer Mutual Verification.

6. DISCUSSION

6.1 Security Analysis

OpenHash's security model differs fundamentally from conventional DLTs. Rather

than constructing trust through computational or economic Sybil resistance, it

inherits trust from the physical administrative hierarchy of telecommunications

infrastructure—an adversary cannot trivially compromise multiple administrative

tiers simultaneously without real-world institutional access.

Balance invariance

The BIVM enforces exact conservation of aggregate balances at and above the

LCAT of every transaction. Any attempt to create value fraudulently results

in a non-zero delta sum, which is detected and blocked without requiring

consensus.

Double-spend prevention

Account locking, dual-signature requirements, and real-time monitoring by

parent-tier nodes create multiple independent barriers against simultaneous

conflicting transactions.

Node corruption

The public hash chain paradigm ensures that any tampered chain entry is

immediately detectable by any observer, since the hash linkage breaks. The

LPBFT recovery process isolates corrupted nodes and reconstructs their

chains from trusted user-chain records.

Replay attacks

Timestamp windows of +/- 5 minutes and sequence number validation prevent

replay of captured transaction hashes.

Byzantine collusion

Cross-layer verification at non-adjacent tiers prevents a subset of

intermediate-tier nodes from colluding to fabricate transactions without

detection by higher or lower tiers.

6.2 Decentralization and Governance

A potential concern is whether mapping trust to existing administrative

hierarchies re-introduces centralization. OpenHash mitigates this through

several mechanisms:

- The global node (L5) rotates among participating nations on a time-limited

basis, preventing sustained single-nation dominance.

- Representative node elections are weekly and geographically

diversity-constrained.

- All hash chains are publicly auditable.

- No single tier can unilaterally modify data without invalidating the

cryptographic linkages visible to adjacent tiers.

The system is best characterized as regulated decentralization—decentralized

within the bounds of existing institutional accountability structures.

6.3 Energy Efficiency

By eliminating PoW and restricting PoS-like consensus to emergency scenarios

affecting less than 0.01% of operational time, OpenHash's energy consumption is

dominated by standard network communication and cryptographic signature

verification—operations already performed for every internet transaction. The

per-transaction energy footprint is therefore comparable to a conventional HTTPS

API call rather than a blockchain mining operation.

6.4 Limitations and Future Work

Limitation 1: Throughput extrapolation

National-scale throughput estimates are extrapolated from single-node

measurements. Full-scale distributed experiments are needed to validate

projections and characterize inter-node coordination overhead.

Limitation 2: Formal security verification

The current security analysis is primarily qualitative. Formal verification

of the BIVM balance invariance guarantee and the LPBFT correctness properties

under the assumed trust model is planned.

Limitation 3: Infrastructure dependency

The system's dependence on existing telecommunications infrastructure means

that catastrophic failures (e.g., natural disasters severing major network

links) could impair operation. The Offline Batch Processing Module provides

partial mitigation.

Limitation 4: Governance of probability parameters

Layer-selection probability distributions are operator-configurable, raising

governance questions about parameter-setting processes and update procedures.

Future work directions:

(i) Formal cryptographic security proof for the BIVM

(ii) Large-scale simulation across realistic network topologies

(iii) Cross-national interoperability protocols

(iv) Integration with existing financial regulatory frameworks

(v) Extension to non-transactional domains: electronic health records,

governmental document authentication

7. CONCLUSION

We have presented OpenHash, a Probabilistic Hierarchical Layer Distribution

system for data integrity verification that fundamentally departs from existing

DLT paradigms. Rather than constructing trust through energy-intensive or

economically complex consensus mechanisms, OpenHash treats the five-tier physical

administrative hierarchy of existing telecommunications networks as its primary

trust substrate.

The resulting system:

- Eliminates Proof-of-Work entirely

- Confines Byzantine consensus to rare emergency scenarios (<0.01% of time)

- Achieves sub-5 ms end-to-end transaction latency

- Achieves single-node throughput of 5,354 TPS, scaling linearly to

projected tens of millions of TPS at national deployment scale

- Reduces inter-tier synchronization bandwidth by over 99% via Merkle

tree aggregation

OpenHash's design has implications beyond technology: by aligning distributed

trust with existing physical and institutional structures, it offers a path

toward national-scale data integrity infrastructure that is simultaneously

energy-sustainable, high-performance, transparent, and compatible with

regulatory accountability. We believe this paradigm—physical hierarchy as trust

substrate—opens a productive new direction for research on distributed data

integrity systems.

REFERENCES

[1] Nakamoto, S. (2008). Bitcoin: A peer-to-peer electronic cash system.

Bitcoin.org. https://bitcoin.org/bitcoin.pdf

[2] Cambridge Centre for Alternative Finance. (2025). Cambridge Bitcoin

Electricity Consumption Index. University of Cambridge.

[3] Buterin, V., & Griffith, V. (2017). Casper the friendly finality gadget.

arXiv:1710.09437.

[4] Popov, S. (2018). The Tangle. IOTA Foundation White Paper, Version 1.4.3.

[5] Baird, L. (2016). The Swirlds Hashgraph consensus algorithm: Fair, fast,

Byzantine fault tolerance. Swirlds Technical Report SWIRLDS-TR-2016-01.

[6] Buterin, V. (2014). A next-generation smart contract and decentralized

application platform. Ethereum White Paper.

[7] de Vries, A. (2018). Bitcoin's growing energy problem. Joule, 2(5),

801–805.

[8] King, S., & Nadal, S. (2012). PPCoin: Peer-to-peer crypto-currency with

proof-of-stake. Self-published.

[9] Buchman, E., Kwon, J., & Milosevic, Z. (2018). The latest gossip on BxFT

consensus. arXiv:1807.04938.

[10] LeMahieu, C. (2018). Nano: A feeless distributed cryptocurrency network.

Nano.org Technical White Paper.

[11] Hafid, A., Hafid, A. S., & Samih, M. (2020). Scaling blockchains: A

comprehensive survey. IEEE Access, 8, 125244–125262.

[12] Kokoris-Kogias, E., Jovanovic, P., Gasser, L., Gailly, N., Syta, E., &

Ford, B. (2018). OmniLedger: A secure, scale-out, decentralized ledger

via sharding. IEEE S&P 2018, 583–598.

[13] Merkle, R. C. (1989). A certified digital signature. In G. Brassard (Ed.),

Advances in Cryptology—CRYPTO '89, LNCS 435, 218–238. Springer.

[14] Castro, M., & Liskov, B. (1999). Practical Byzantine fault tolerance.

Proceedings of the 3rd Symposium on Operating Systems Design and

Implementation (OSDI '99), 173–186.

[15] Yin, M., Malkhi, D., Reiter, M. K., Gueta, G. G., & Abraham, I. (2019).

HotStuff: BFT consensus with linearity and responsiveness. Proceedings of

PODC '19, 347–356.

[16] Androulaki, E., Barger, A., Bortnikov, V., Cachin, C., Christidis, K.,

De Caro, A., … & Yellick, J. (2018). Hyperledger Fabric: A distributed

operating system for permissioned blockchains. Proceedings of EuroSys '18,

Article 30.

[17] Boneh, D., Lynn, B., & Shacham, H. (2001). Short signatures from the Weil

pairing. In C. Boyd (Ed.), Advances in Cryptology—ASIACRYPT 2001,

LNCS 2248, 514–532. Springer.

[18] NIST. (2023). CRYSTALS-Dilithium digital signature standard. FIPS PUB 204

(Draft). National Institute of Standards and Technology.

[19] Shamir, A. (1979). How to share a secret. Communications of the ACM,

22(11), 612–613.

[20] Liu, F. T., Ting, K. M., & Zhou, Z.-H. (2008). Isolation forest.

Proceedings of the 8th IEEE International Conference on Data Mining

(ICDM 2008), 413–422.

[문서 끝 / End of Document]

파일명 : OpenHash_SCI_Paper_Draft.txt

버전 : v1.2 (2026-03-23)

수정 시 상단 [문서 정보] 블록의 버전 및 최종 수정 날짜를 함께 갱신하십시오.