100 TPS for linear leios (#463)

* Experimental design
* Analysis of results
* Analyzed disposition of blocks
* Updated logbook

Author: bwbush

Logbook.md

@@ -1,5 +1,33 @@
 # Leios logbook
 
+## 2025-07-19
+
+### 100 TPS experiment for Stracciatella and Linear Leios
+
+Experiments using the Rust simulation demonstrated the viability of Stracciatella and Linear Leios at 100 TPS if appropriate stage lengths and maximum EB sizes are chosen.
+
+- Variants
+    - Linear Leios, linear
+    - Linear Leios with tx references, linear-with-tx-references
+    - Stracciatella, full-without-ibs
+- 100 tx/s for first 900 seconds
+- 1400 B/tx
+
+Findings:
+
+1. 5 slot/stage is too short for Linear Leios at 100 tx/s.
+2. Including transactions in EBs (instead of references) results in congestion and delays at 100 tx/s.
+3. 10 MB/EB is sufficient for 100 tx/s but 5 MB/EB is not.
+4. EB-sortition unluckiness in Stracciatella can lengthen the transaction lifecycle, but this could be remedied by increasing the EB production rate.
+5. CPU and network usage peak high when transactions are included in EBs.
+6. Caveat: this conclusion may change when better transaction validation times are used in the simulator configuration.
+
+Artifacts:
+
+- [Slides](analysis/sims/2025w29b/ReadMe.pdf)
+- [Jupyter notebook](analysis/sims/2025w29b/analysis)
+- [Configurations](analysis/sims/2025w29b/)
+
 ## 2025-07-18
 
 ### Rust simulation

analysis/sims/2025w29b/ReadMe.pdf

@@ -0,0 +1,41 @@
+#!/usr/bin/env nix-shell
+#!nix-shell -i bash -p gnused gzip pigz "rWrapper.override { packages = with rPackages; [ data_table R_utils bit64 ggplot2 magrittr stringr ]; }"
+
+set -e
+
+mkdir -p results/$d
+for f in lifecycle resources receipts cpus
+do
+  DIR=$(find linear linear-with-tx-references full-without-ibs -type f -name $f.csv.gz \( -not -empty \) -printf %h\\n -quit)
+  HL=$(sed -n -e '1p' "$DIR/case.csv")
+  HR=$(zcat "$DIR/$f.csv.gz" | sed -n -e '1p')
+  if [[ "$f" == "lifecycle" || "$f" == "resources" ]]
+  then
+    FRACT=1.00
+  else
+    FRACT=0.20
+  fi
+  (
+    echo "$HL,$HR"
+    for g in $(find linear linear-with-tx-references full-without-ibs -type f -name $f.csv.gz \( -not -empty \) -printf %h\\n)
+    do
+      if [ ! -e "$g/stderr" ]
+      then
+        echo "Skipping $g because it has no stderr." >> /dev/stderr
+      elif [ -s "$g/stderr" ]
+      then
+        echo "Skipping $g because its stderr is not empty." >> /dev/stderr
+      else
+        BL=$(sed -n -e '2p' "$g/case.csv")
+        zcat "$g/$f.csv.gz" | gawk 'FNR > 1 && rand() <= '"$FRACT"' { print "'"$BL"'" "," $0}'
+      fi
+    done
+  ) | pigz -p 3 -9c > results/$f.csv.gz
+R --vanilla << EOI > /dev/null
+require(data.table)
+sampleSize <- $FRACT
+print(sampleSize)
+$f <- fread("results/$f.csv.gz", stringsAsFactors=TRUE)
+save($f, sampleSize, file="results/$f.Rdata", compression_level=9)
+EOI
+done

analysis/sims/2025w29b/analysis.ipynb

@@ -0,0 +1,6 @@
+full-without-ibs/8,20000000/run.sh
+full-without-ibs/5,20000000/run.sh
+full-without-ibs/8,10000000/run.sh
+full-without-ibs/5,10000000/run.sh
+full-without-ibs/8,5000000/run.sh
+full-without-ibs/5,5000000/run.sh

analysis/sims/2025w29b/combine-results.sh

@@ -0,0 +1,2 @@
+Simulator,Variant,Stage length,Max EB size
+Rust,full-without-ibs,5 slot/stage,10.0 MB/EB

analysis/sims/2025w29b/full-without-ibs.list

@@ -0,0 +1,225 @@
+# yaml-language-server: $schema=./config.schema.json
+
+################################################################################
+# Simulation Configuration File
+################################################################################
+#
+# This file contains the default configuration for running Leios simulations in
+# the Haskell simulation (`simulation/`) and the Rust simulation (`sim-rs/`).
+#
+################################################################################
+# Simulation Configuration
+################################################################################
+
+relay-strategy: "request-from-first"
+tcp-congestion-control: true
+multiplex-mini-protocols: true
+simulate-transactions: true
+treat-blocks-as-full: false
+cleanup-policies: ["cleanup-expired-vote"]
+timestamp-resolution-ms: 0.1
+
+################################################################################
+# Leios Protocol Configuration
+################################################################################
+
+leios-variant: full-without-ibs
+linear-vote-stage-length-slots: 5
+linear-diffuse-stage-length-slots: 5
+leios-stage-length-slots: 5
+leios-stage-active-voting-slots: 1
+leios-vote-send-recv-stages: false
+leios-late-ib-inclusion: true
+leios-header-diffusion-time-ms: 1000.0
+leios-mempool-sampling-strategy: ordered-by-id
+leios-mempool-aggressive-pruning: true
+# TODO: revise default
+praos-chain-quality: 100
+praos-fallback-enabled: true
+
+################################################################################
+# Transaction Configuration
+################################################################################
+
+tx-generation-distribution:
+  distribution: exp
+  lambda: 0.85
+  scale: 8.95
+tx-size-bytes-distribution:
+  distribution: constant
+  value: 1400
+tx-conflict-fraction: 0
+tx-overcollateralization-factor-distribution:
+  distribution: constant
+  value: 0
+tx-validation-cpu-time-ms: 0.065
+tx-max-size-bytes: 16384
+tx-start-time: 0
+tx-stop-time: 900
+
+################################################################################
+# Ranking Block Configuration
+################################################################################
+
+# 1/leios-stage-length-slots, targeting one RB per pipeline.
+# Also 20s is current rate of praos blocks.
+rb-generation-probability: 5.0e-2
+# Eng. team targets 1kB as worst case upper bound.
+# Actual size fairly close.
+rb-head-size-bytes: 1024
+rb-body-max-size-bytes: 90112
+# Note: certificate generation/validation is not included in the
+# timings here, see cert-* fields.
+rb-generation-cpu-time-ms: 1.0
+rb-head-validation-cpu-time-ms: 1.0
+
+# On average, no Txs directly embedded in blocks.
+rb-body-legacy-praos-payload-avg-size-bytes: 0
+rb-body-legacy-praos-payload-validation-cpu-time-ms-constant: 50.0
+# the -per-byte component is meant to be using size as a (bad)
+# proxy for the complexity of the Txs included.
+rb-body-legacy-praos-payload-validation-cpu-time-ms-per-byte: 0.0005
+
+################################################################################
+# Input Block Configuration
+################################################################################
+
+ib-generation-probability: 2.0
+ib-shards: 1
+ib-shard-period-length-slots: 1
+ib-shard-group-count: 1
+
+# ProducerId  32
+# SlotNo      64
+# VRF proof   80
+# Body hash   32
+# RB Ref      32
+# Signature   64
+# Total       304
+#
+# NOTE: using a KES Signature (like for Praos headers)
+#       would instead more than double the total to 668.
+#       And even 828 including Op Cert.
+ib-head-size-bytes: 304
+# 98KB to optimize for 3 TCP round trips
+ib-body-avg-size-bytes: 98304
+ib-body-max-size-bytes: 262144
+# Here we also use praos blocks as ballpark estimate.
+# Sec 2.3 Forging, of the benchmark cluster report, lists
+#   * Slot start to announced: 0.12975s
+ib-generation-cpu-time-ms: 130.0
+ib-head-validation-cpu-time-ms: 1.0
+ib-body-validation-cpu-time-ms-constant: 50.0
+ib-body-validation-cpu-time-ms-per-byte: 0.0005
+ib-diffusion-strategy: "freshest-first"
+
+# Haskell prototype relay mini-protocol parameters.
+ib-diffusion-max-bodies-to-request: 1
+ib-diffusion-max-headers-to-request: 100
+ib-diffusion-max-window-size: 100
+
+################################################################################
+# Endorsement Block Configuration
+################################################################################
+
+# We want one per pipeline, but not too many.
+eb-generation-probability: 1.0
+# ProducerId  32
+# SlotNo      64
+# VRF proof   80
+# Signature   64
+# Total       240
+#
+# See Note about signatures on ib-head-size-bytes.
+eb-size-bytes-constant: 240
+# IB hash
+eb-size-bytes-per-ib: 32
+# Collecting the IBs to reference and cryptography are the main tasks.
+# A comparable task is maybe mempool snapshotting.
+# Sec 2.3 Forging, of the benchmark cluster report, lists
+#   * Mempool snapshotting: 0.07252s
+# 75ms then seems a generous estimate for eb generation.
+eb-generation-cpu-time-ms: 75.0
+# Validating signature and vrf proof, as in other headers.
+eb-validation-cpu-time-ms: 1.0
+
+eb-diffusion-strategy: "peer-order"
+
+# Haskell prototype relay mini-protocol parameters.
+eb-diffusion-max-bodies-to-request: 1
+eb-diffusion-max-headers-to-request: 100
+eb-diffusion-max-window-size: 100
+
+# The maximum age of EBs included in RBs.
+# A an EB from slot `s` can only be included in RBs
+# up to slot `s+eb-max-age-slots`.
+# In short leios we expect votes to diffuse within 3 stages lengths of
+# EB generation, we allow for 2 more stage lengths to account for
+# variance in the interval within RBs.
+eb-max-age-slots: 240
+
+# The maximum age of EBs to be relayed.
+# An EB from slot `s` will only be relayed
+# up to slot `s+eb-max-age-for-relay-slots`.
+eb-max-age-for-relay-slots: 40
+
+# The maximum size of transactions (in bytes) which an EB can reference.
+# Only relevant when running with the "full-without-ibs" variant.
+eb-referenced-txs-max-size-bytes: 10000000
+eb-body-avg-size-bytes: 10000000
+
+################################################################################
+# Vote Configuration
+################################################################################
+
+# Cryptography related values taken from [vote-spec](crypto-benchmarks.rs/Specification.md)
+# using weighted averages of 80% persistent and 20% non-persistent.
+
+# vote-spec#Committe and quorum size
+#
+# Note: this is used as the expected amount of total weight of
+# generated votes in the sims.
+vote-generation-probability: 500.0
+# vote-spec#"Committe and quorum size"
+# 60% of `vote-generation-probability`
+vote-threshold: 300
+# vote-spec#"Generate vote" 0.8*135e-3 + 0.2*280e-3
+vote-generation-cpu-time-ms-constant: 164.0e-3
+# No benchmark yet.
+vote-generation-cpu-time-ms-per-ib: 0
+# vote-spec#"Verify vote" 0.8*670e-3 + 0.2*1.4
+vote-validation-cpu-time-ms: 816.0e-3
+# The `Vote` structure counted in the -per-eb already identifies slot
+# (in Eid) and voter. We can assume a vote bundle is all for the same
+# voter and slot, so for non-persistent voters we could factor their
+# PoolKeyHash (28bytes) here, but that is for 20% of cases.
+# More relevant if EB generation is very high.
+vote-bundle-size-bytes-constant: 0
+# vote-spec#Votes 0.8*90 + 0.2*164
+vote-bundle-size-bytes-per-eb: 105
+
+vote-diffusion-strategy: "peer-order"
+
+# Haskell prototype relay mini-protocol parameters.
+vote-diffusion-max-bodies-to-request: 1
+vote-diffusion-max-headers-to-request: 100
+vote-diffusion-max-window-size: 100
+
+################################################################################
+# Certificate Configuration
+################################################################################
+
+# vote-spec - certificate size plot.
+# Realistic stake distributions need about 7 kilobytes for the certificate.
+cert-size-bytes-constant: 7168
+cert-size-bytes-per-node: 0
+
+# For certificate timings we have bulk figures for realistic scenarios,
+# so we do not attempt to give -per-node (i.e. per-voter) timings.
+#
+# vote-spec#"Generate certificate"
+cert-generation-cpu-time-ms-constant: 90.0
+cert-generation-cpu-time-ms-per-node: 0
+# vote-spec#"Verify certificate"
+cert-validation-cpu-time-ms-constant: 130.0
+cert-validation-cpu-time-ms-per-node: 0

analysis/sims/2025w29b/full-without-ibs/5,10000000/case.csv

@@ -0,0 +1 @@
+../../linear/5,5000000/run.sh