Blockchain TPS Benchmarks: Why They Often Mislead

Blockchain networks frequently tout staggering transactions per second (TPS) figures—sometimes measured in millions—to highlight their scalability. Yet, as Preston Evans explains in a recent analysis for Celestia, these benchmarks often fail to reflect real-world performance. The gap between marketing claims and production realities can mislead developers, investors, and users alike.

Evans notes that most TPS benchmarks are conducted under short bursts of traffic in controlled environments, often ignoring critical constraints like sustained load, storage scalability, and syncing times in live networks. For instance, a blockchain that claims to handle one million TPS in a lab might struggle to sustain even a fraction of that under decentralized, production-grade conditions.

Storage and Indexing Bottlenecks

One major limitation is storage. A system processing a million TPS with 100-byte transactions would generate roughly 8.5 terabytes of data per day. Benchmarks typically sidestep this issue by running only for hours, avoiding the storage and indexing challenges that arise with prolonged activity. In production, indexers—off-chain systems enabling transaction queries—require up to ten times the raw storage of the chain itself, further amplifying costs and complexity.

Developers across high-TPS chains, including Ethereum, Solana, and Layer 2 networks like Optimism and Arbitrum, have acknowledged these constraints. Ethereum, for example, intentionally caps its average throughput far below peak capacity to ensure long-term viability, a limitation reflected in upgrades like blob expansion for Layer 2 scalability.

IOPS and Database Performance

Another overlooked factor is Input/Output Operations Per Second (IOPS). Blockchain states are stored in databases like RocksDB, which use an append-only design for write efficiency. However, compaction processes—critical for preventing storage bloat—can significantly degrade throughput over time. This means TPS benchmarks based on freshly loaded databases often understate the real-world IOPS demands of long-running systems.

Sync Times Under Production Load

Node syncing is another critical issue rarely addressed in benchmarks. Syncing a node involves downloading a snapshot of the blockchain’s state and catching up on recent transactions. If a network operates near its capacity, syncing times can balloon. For example, syncing a node from a six-hour-old snapshot could take over 60 hours on a network running at 90% capacity, creating operational risks.

The Data Availability Problem

Data Availability (DA)—the ability for all nodes to retrieve and verify transaction data—is another scalability bottleneck. Most blockchains rely on peer-to-peer broadcast for DA, which becomes bandwidth-intensive at scale. Solana reportedly requires up to 20Mbps for its current throughput, capping its practical TPS ceiling at around 12,500 with current infrastructure. Advanced techniques like Celestia Fibre’s ZODA aim to address these limits, scaling DA to handle up to 1Tbps or 625 million TPS in the future.

Implications for Developers and Traders

For developers, understanding the limitations of TPS benchmarks is crucial when designing scalable applications. Marketing-level TPS figures often exclude factors like failed transactions, validator decentralization, or realistic transaction mixes. As recent independent analyses suggest, production-grade throughput requires standardized benchmarking methodologies and transparency around hardware specs, validator counts, and network conditions.

For traders, inflated TPS claims can mislead investment decisions. As seen with Polygon’s May 2026 upgrade, which boosted throughput by 14%, real-world improvements often fall short of theoretical claims. Investors should focus on verifiable upgrades and independent audits rather than headline-grabbing metrics.

Ultimately, the blockchain industry must prioritize accuracy and transparency in performance evaluations. While next-generation designs like those from Celestia show promise, real-world scalability will remain a critical benchmark for adoption and long-term success.