The Ledger
Cardano Finds 32x Storage-Latency Gap in Leios Tests
Cardano’s latest Leios work found raw transaction-read latency varied about 32-fold across participating stake pool operator machines, narrowing to roughly 7–8-fold after CPU adjustment. Intersect also added nested-transaction and reference-script profiles to its Dijkstra benchmarking tools, but did not publish performance results for those profiles.

Raw transaction-read latency varied about 32-fold across the stake pool operator machines in Cardano’s latest Leios benchmark, according to an October 29 update from Intersect’s Performance and Tracing team. Adjusting for differences in CPU speed narrowed the spread to roughly seven to eight times, with disk write-cache behavior accounting for much of the remaining gap.
The results come from a transaction-validation benchmark distributed to stake pool operators (SPOs) so they could run it on their own hardware. The update also describes preliminary Leios cluster tests and new Dijkstra-era benchmark profiles, but it does not publish performance figures for either set of work.
CPU adjustment cuts the gap, but not the hardware effect
Intersect says the team gathered submissions from a broad sample of real SPO machines and measured raw read latency per transaction. CPU speed explained much of the initial variation; after accounting for it, the results still differed by roughly seven to eight times. The update identifies write-back versus write-through disk caching as the main factor behind that adjusted spread.
“Write-through disks were close to 6x slower than write-back ones before adjusting for CPU, and still more than 3x slower after”
Michael Karg, Performance and Tracing Team Lead
A battery-backed RAID controller largely closed the remaining gap for write-through disks, the update says. Intersect did not give a count of participating machines in the post, which links the full breakdown (“Benchmark results”), so the figures describe the submitted hardware sample rather than a quantified profile of every Cardano stake pool.
Leios cluster findings remain internal
The team also ran cluster benchmarks on a Leios prototype and a subsequent Week42 build. Its analysis examined the share of certified Endorser Blocks under different voting traffic and timing conditions, compared Ranking Block production, diffusion and adoption against a Praos-only baseline, and measured mempool fragmentation alongside possible gains from TxCache.
Those findings have been shared with the Leios team but are not public results, Intersect says. The tests do not describe an official release candidate, and the team is still working on known bottlenecks while refining its measurement coverage. The update therefore offers no public throughput figure from the cluster work.
Dijkstra adds two benchmark profiles ahead of Node 11.2
For the planned Node 11.2 release, the performance workbench gained profiles for nested transactions—transactions that carry sub-transactions—and Plutus calls based on reference scripts. The reference-script path is being measured alongside the existing witness-script path. Intersect describes this as an extension to its benchmarking framework, not as a report of comparative results.
To make the nested-transaction tests load the whole cluster quickly enough for representative measurements, the team reduced the wait before a node requests transactions from a newly connected peer from 60 seconds to one second. That setup change and the still-private Leios cluster analysis leave the public hardware benchmark as the update’s clearest quantitative finding: write-through disks were more than three times slower even after CPU adjustment, except where a battery-backed RAID controller narrowed the difference.
Sources
- Original source Cardano Development Updates updates.cardano.intersectmbo.org