IOTA’s engineers have found where the speed limit sits. Internal load tests on November 5 pushed the network’s new P-COOL transaction flow until it topped out at roughly 1,100 transactions per second on a setup with 11 validators. The choke point was not consensus. It was the checking of transactions after consensus. Five days later, developers posted code changes aimed at breaking through it.
Why the test matters
P-COOL is the transaction flow the IOTA core team named as its next priority in its Q3 report on October 13. It removes the certification step that transactions used to need before consensus. Conflicts over owned objects are now settled deterministically after consensus instead. IOTA has promised higher performance from P-COOL at roughly half the resources, and the Foundation said it would stress-test the design on public networks in the fourth quarter. The November 5 run is the first public look at where the ceiling is.
Where the bottleneck was
Measurements from the test showed that 69% of the consensus handler’s CPU time went to reading and checking transaction inputs. About 21% of the total went to specific lookups in the RocksDB database. At 900 TPS, the time needed to process each consensus commit climbed from about 28 to 37 milliseconds within three minutes, a sign the validator was starting to fall behind.
Three fixes in the pipeline
Developers have opened several pull requests on the IOTA repository, each aimed at a different part of the problem:
- Stop overfilling the queue (PR #13548, Roman Overko). Today, P-COOL’s congestion check counts a transaction only after several pre-checks finish. Transactions arriving in the meantime see an outdated load figure, so a validator can admit more than its queue can hold. The fix makes each transaction reserve a spot in the consensus queue the moment it is accepted, and release it if the transaction is rejected or the submission fails. It changes validator software only, not protocol rules.
- Stop repeating database reads (PR #13553, Filip Dulić). Inside a single consensus commit, the same data is often fetched more than once. In a test workload of about 80 transactions, roughly 400 of 1,360 input reads were repeats. The proposal reuses results within a commit, and a prefix Bloom filter would let the database skip storage ranges that cannot contain the object being searched for. Dulić’s estimate is a 10 to 15% saving in consensus-handler CPU time, which he notes still has to be measured.
- Read inputs in parallel (PR #13559). This change loads the inputs of multiple transactions at the same time. In another load test, validators used less than 30% of their available CPU, which points to idle capacity that parallel reads could put to work.
Why 1,100 TPS is not the headline number
This figure should not be compared with IOTA’s biggest throughput claims. The Rebased upgrade was pitched at more than 50,000 TPS, and simulations in the Starfish consensus paper reached about 250,000 transactions per second with 100 parties. Those benchmarks test consensus on its own. The Starfish research prototype, for example, measures consensus performance without execution or ledger storage.
The November 5 test measured a different thing: the full path on a small 11-validator network, including the work validators do after consensus. The gap between the two kinds of numbers shows how much throughput can be lost after the ordering step is done. It is also a reminder that lab figures and mainnet figures are not interchangeable.
Also Read: IOTA Could Open Up Funding for Small Export Deals: Here’s How Trade Finance Works
What is still unknown
None of these changes is live. They are still in development and review, and several are assigned to the v1.37.0 milestone. There is no confirmed mainnet activation date. The 10 to 15% CPU saving is an estimate rather than a result, and the public-network load tests planned for Q4 will show whether the fixes hold up outside a controlled setup.
FAQ
What is IOTA’s P-COOL? It is a transaction flow that skips the certification step before consensus and settles conflicts over owned objects after consensus. IOTA says it should deliver higher performance with roughly half the resources.
Is 1,100 TPS IOTA’s maximum speed? No. It is the ceiling a specific internal test hit on 11 validators, with transaction checking after consensus as the bottleneck. Developers have proposed fixes, and IOTA’s consensus research cites far higher figures under different test conditions.
Disclaimer: The information in this article is for general purposes only and does not constitute financial advice. The author’s views are personal and may not reflect the views of Chain Affairs. Before making any investment decisions, you should always conduct your own research. Chain Affairs is not responsible for any financial losses.
