Skip to content

RTP - Performance Results

Headline. On Paper, RTP sustained 18.7 teleports/s at 100 % success under unpaced dispatch with p95 2 ms and TPS never below 17.5. On Folia it held 13.5 TP/s with zero watchdog stalls while EzRTP froze a region for 20.4 s.

Measured against currently-shipping alternatives with the same harness, world, and hardware, using real connected clients. Methodology, per-run analyses, retractions, and known confounds: PRE_WRITEUP.md. Harness source: helpers/StressTestRTP/. Per-attempt CSVs are kept locally under runs/ (gitignored) and are not published in this repository.

Only runs against competitor versions that are still current are listed. The earlier Spigot / Paper 1.20.1 passes were dropped once every competitor in them had shipped a newer release, and the 1.07 / 35.9 / 64.0 chunks-per-attempt figures from those runs are retracted (worldgen artefacts; see ADR-080).


Cross-platform summary

Run Platform Plugin TP/s p95 Min TPS Success
20261008-003310 Paper 26.2, unpaced (Test A) LeafRTP 42.2 7 ms 16.2 100 % (4,096 / 4,096)
20261008-003310 Paper 26.2, unpaced (Test A) JakesRTP 23.2 73 ms 12.1 100 % (4,096 / 4,096)
20261008-003310 Paper 26.2, unpaced (Test A) BetterRTP 9.2 628 ms 13.2 100 % (4,096 / 4,096)
20261008-003310 Paper 26.2, unpaced (Test A) HuskHomes 8.4 411 ms 19.9 100 % (4,096 / 4,096)
20261008-003310 Paper 26.2, unpaced (Test A) EzRTP 7.9 748 ms 18.0 100 % (4,096 / 4,096)
20261008-003310 Paper 26.2, unpaced (Test A) JustRTP 4.4 1 724 ms 19.8 99.8 % (4,088 / 4,096)
20261008-052141 Folia 26.2, paced (Test B) LeafRTP 4.8 203 ms 19.9 100 % (4,096 / 4,096)
20261008-052141 Folia 26.2, paced (Test B) HuskHomes 4.6 785 ms 20.0 99.98 % (4,095 / 4,096)
20261008-052141 Folia 26.2, paced (Test B) JustRTP 2.3 3 050 ms 19.8 99.2 % (4,056 / 4,087)
20261008-052141 Folia 26.2, paced (Test B) EzRTP 0.8 4 995 ms 20.0 59.9 % (1,415 / 2,364)
20261008-052141 Folia 26.2, paced (Test B) BetterRTP - - - Pruned (async chunk error on region thread)
20261008-052141 Folia 26.2, paced (Test B) JakesRTP - - - Unsupported (no Folia support)
20260617-232754 Paper 26.1, unpaced (Historical) RTP (Pro) 18.7 2 ms 17.5 100 %
20260617-232754 Paper 26.1, unpaced (Historical) EzRTP 13.1 189 ms 10.3 98.3 %
20260617-232754 Paper 26.1, unpaced (Historical) BetterRTP 6.0 3 217 ms 2.5 98.3 %
20260617-191448 Folia 26.1 (Historical) RTP (Pro) 13.5 - - ~100 %
20260617-205614 Folia 26.1 (Historical) RTP (Lite) 12.5 - 16 100 %
20260617-191448 Folia 26.1 (Historical) EzRTP 5.3 - - 96.2 %

All rows are n=1 on a single rig except the Folia watchdog-stall count, which reproduced across two runs. Read them as "this rig, this version, this configuration", not as universal claims.

Test B (20261008-024412) holds every plugin to the same offered rate (one /rtp per 200 ms), so it compares server cost at equal work rather than throughput; it has its own section below.


Paper 26.2 - unpaced dispatch (Test A: 20261008-003310, 4,096 teleports per plugin, 16k radius)

Full 6-plugin unpaced saturation run on Paper 26.2 (Ryzen 9 3900X, 16 GB heap). 3 OPed clients, concurrency 4, immediate-redispatch: true, dispatch-interval-ms: 0, per-player-gap-ticks: 0. Outer radius 16,384 blocks, inner radius 1,024 blocks. EzRTP tested with biome verification active.

Plugin Success / Attempts Wall Time Throughput Latency (p50 / p95 / p99) Chunks / att (attr / inc) Main CPU / att Min TPS (Avg) MSPT p99 Peak Heap
LeafRTP 4,096 / 4,096 (100 %) 97.2 s (1.6 min) 42.15 TP/s 5.0 ms / 7.0 ms / 9.0 ms 0.18 / 2.92 9.92 ms 16.19 (17.82) 80.6 ms 16,296 MB
JakesRTP 4,096 / 4,096 (100 %) 176.5 s (2.9 min) 23.20 TP/s 21.0 ms / 73.0 ms / 114.0 ms 7.14 / 8.46 13.17 ms 12.14 (15.02) 104.8 ms 14,511 MB
BetterRTP 4,096 / 4,096 (100 %) 444.2 s (7.4 min) 9.22 TP/s 223.0 ms / 628.0 ms / 1,013.0 ms 24.96 / 26.06 24.86 ms 13.23 (19.16) 120.2 ms 15,003 MB
HuskHomes 4,096 / 4,096 (100 %) 489.3 s (8.2 min) 8.37 TP/s 258.0 ms / 411.0 ms / 748.0 ms 31.66 / 32.22 30.72 ms 19.92 (20.00) 17.8 ms 14,339 MB
EzRTP 4,096 / 4,096 (100 %) 517.7 s (8.6 min) 7.91 TP/s 240.0 ms / 748.0 ms / 1,125.0 ms 30.94 / 32.07 34.35 ms 18.01 (19.57) 88.3 ms 14,157 MB
JustRTP 4,088 / 4,096 (99.8 %) 923.8 s (15.4 min) 4.43 TP/s 418.0 ms / 1,724.0 ms / 2,721.0 ms 89.53 / 91.79 82.79 ms 19.83 (19.99) 26.9 ms 13,459 MB

Paper 26.2 Linux realtime - latency at 5 to 40 TP/s

Threadripper 7970X rig (Ubuntu 24.04, OpenJDK 25, -Xmx16G, 48 non-op Mineflayer bots). The Paper JVM ran under SCHED_FIFO priority 10 (every server thread class FF), with only the plugin under test loaded and a restart before every run. Each run is a ramp of 60 s stages at 5, 10, 20, 40, 80, 160 and 320 TP/s that stops after two failed stages in a row. Competitor rows pool 4 ramps each (2026-10-17 20:24 to 23:56 UTC). LeafRTP's row pools 2 ramps of the current 3.3.0 build (20261018-022257, 20261018-023204; jar sha256 e4ef65af..., mode: "REROLL", cacheCap: 10, backlogCacheCap: 0, activeChunkCap: 5, all recorded in each run's -provenance.txt and config snapshot). Latency is dispatch to arrival over every successful attempt dispatched inside the stage window. MSPT p95 is the per-run stage value; no idle baseline is subtracted.

Plugin Runs 5 TP/s p50 / p95 / p99 10 TP/s p50 / p95 / p99 20 TP/s p50 / p95 / p99 40 TP/s p50 / p95 / p99 Safe max (all runs PASS)
LeafRTP 2 5 / 5 / 6 ms 5 / 5 / 6 ms 5 / 5 / 7 ms 158 / 232 / 285 ms 40 TP/s
JakesRTP 4 0 / 7 / 13 ms 3 / 14 / 26 ms 4 / 15 / 23 ms 4 / 15 / 24 ms (FAIL: MSPT p95 77-118 ms) 20 TP/s
HuskHomes 4 221 / 370 / 517 ms 224 / 365 / 426 ms 275 / 381 / 474 ms 568 / 1,071 / 1,574 ms 40 TP/s
BetterRTP 4 156 / 506 / 709 ms 199 / 514 / 806 ms 199 / 522 / 815 ms 510 / 3,242 / 4,323 ms (FAIL: 127 failed) 20 TP/s
EzRTP 4 154 / 602 / 905 ms FAIL (619 of 2,396 failed) - - 5 TP/s
JustRTP 4 367 / 1,666 / 2,471 ms 473 / 1,622 / 2,480 ms 476 / 1,626 / 2,477 ms FAIL (25.5 of 40 TP/s delivered) 20 TP/s

LeafRTP completed 30,914 of 30,914 teleports across both ramps, with 0 exact-block repeats in either run (15,420 and 15,494 landings). Its MSPT p95 was 21.2-21.3 ms at 5 TP/s, 24.6-24.7 ms at 10, 35.4-36.9 ms at 20 and 42.6-45.2 ms at 40; it failed 80 TP/s on MSPT p95 (87.9-90.5 ms) with 0 failed teleports. At 40 TP/s its ready queue runs out with this config, so those teleports wait on a live selection.


Paper 26.2 Linux - paced 5 TP/s, 14 ladder runs per plugin [Historical]

Threadripper 7970X rig (Ubuntu 24.04, OpenJDK 25, -Xmx16G, 48 non-op Mineflayer bots, server view distance 4, harness caps each bot to 2), same rig as the open-loop ramps in PRE_WRITEUP.md section 5.10. Each row pools the first stage (5 TP/s offered, 60 s, 299 attempts) of the 14 ramp runs per plugin from 20261009-092205 to 20261010-070617 in which that plugin was the one loaded; the two earlier rounds measured a server without the plugin and are excluded (section 5.10 validity note). LeafRTP ran the tuned bench config. Latency is dispatch to arrival over every attempt dispatched inside the stage window; MSPT p95 and min TPS are the per-run stage values. No idle baseline was recorded, so MSPT is gross.

Plugin Runs Success / Attempts Latency (p50 / p95 / p99 / max) MSPT p95 (per run) Min TPS Exact duplicate landings (summed per run)
LeafRTP 14 4,186 / 4,186 5 / 6 / 6 / 33 ms 21.4-25.0 ms 19.82 3
JakesRTP 14 4,186 / 4,186 0 / 8 / 16 / 45 ms 22.4-24.9 ms 19.67 0
HuskHomes 14 4,186 / 4,186 220 / 369 / 471 / 617 ms 20.2-22.8 ms 19.98 0
EzRTP 14 4,186 / 4,186 154 / 511 / 904 / 2,403 ms 21.0-23.7 ms 19.64 0
JustRTP 14 4,186 / 4,186 368 / 1,619 / 2,425 / 4,668 ms 20.0-22.9 ms 19.82 0
BetterRTP 11 16,789 / 16,789 197 / 510 / 805 / 1,706 ms 22.6-33.6 ms 18.76 1

BetterRTP timed out on every attempt in the ladder runs (dispatch issue), so its row comes from 11 later BetterRTP-only paced runs (20261010-180827 to 20261011-014150: six 60 s stages and five 600 s stages at 5 TP/s) and is not pooled the same way as the others. JakesRTP had the lowest latency at this rate; at 40 and 80 TP/s in the same ladders its MSPT reached 96 ms and 295 ms (BENCHMARK_ANALYSIS.md). The three LeafRTP repeats came 20-48 s after the first landing on the same block (runs 20261009-131313, 20261010-021445, 20261010-070617); the higher-rate stages of the same runs repeated more often and are not in this table.


Paper 26.2 - paced dispatch at 5 TP/s (Test B: 20261008-024412, 4,096 teleports per plugin, 16k radius)

Same server, world, radius, clients and plugin configs as Test A. One /rtp dispatched every 200 ms (immediate-redispatch: false, dispatch-interval-ms: 200, per-player-gap-ticks: 4), 180 s settle gap, spark on for every phase. Every plugin is asked for the same rate, so the MSPT and CPU columns compare cost at equal work. Achieved rate is attempts / phase wall time: five plugins held 4.50-4.63 TP/s; JustRTP fell behind at 3.84 TP/s.

Plugin Success / Attempts Achieved rate Latency (p50 / p95 / p99 / max) Chunks / att (attr / inc) Main CPU / att Process CPU / att MSPT (p50 / p99 / max) Ticks > 50 ms Min TPS (Avg)
LeafRTP 4,096 / 4,096 (100 %) 4.63 TP/s 6 / 9 / 14 / 34 ms 9.44 / 58.87 55.49 ms 537.1 ms 16.4 / 22.6 / 24.5 ms 0 19.89 (20.00)
JakesRTP 4,096 / 4,096 (100 %) 4.56 TP/s 0 / 18 / 76 / 710 ms 3.88 / 64.58 69.50 ms 581.6 ms 28.8 / 54.5 / 92.1 ms 271 (1.52 %) 18.55 (19.88)
BetterRTP 4,096 / 4,096 (100 %) 4.50 TP/s 203 / 562 / 898 / 2,151 ms 27.89 / 64.25 60.42 ms 566.4 ms 27.9 / 54.6 / 166.6 ms 329 (1.83 %) 16.86 (19.80)
HuskHomes 4,096 / 4,096 (100 %) 4.58 TP/s 264 / 419 / 626 / 3,995 ms 38.63 / 62.48 72.83 ms 585.7 ms 21.1 / 32.3 / 37.8 ms 0 19.90 (20.00)
EzRTP 4,096 / 4,096 (100 %) 4.55 TP/s 171 / 608 / 923 / 1,649 ms 25.17 / 63.48 63.74 ms 555.6 ms 25.7 / 42.8 / 49.4 ms 0 19.70 (19.98)
JustRTP 4,094 / 4,096 (99.95 %) 3.84 TP/s 345 / 1,739 / 2,729 / 5,000 ms 89.42 / 102.46 96.42 ms 751.2 ms 23.4 / 34.3 / 36.3 ms 0 19.92 (20.00)

MSPT against the median of the last 120 s of each settle gap: LeafRTP +4.3 ms (12.2 -> 16.4), JakesRTP +15.5, BetterRTP +17.1, EzRTP +13.1, JustRTP +10.5, HuskHomes +3.4 (17.6 -> 21.1; its baseline followed the JakesRTP phase and sat 4-7 ms above the other five). CPU per attempt includes the server's own tick work over a phase of near-equal length. Analysis: PRE_WRITEUP.md section 5.8.


Folia 26.2 - paced dispatch at 5 TP/s (Test B: 20261008-052141, 4,096 teleports per plugin, 16k radius)

Same world, seed, radius, clients, and harness settings as Paper 26.2 Test B on Folia 26.2 (16 GB heap). JakesRTP was omitted (no Folia support declared). BetterRTP threw Cannot retrieve chunk asynchronously (region={null}) during warm-up and was pruned. EzRTP threw getHighestBlockYAt off-thread exceptions locking out 2 of 3 player accounts and capped out at 1,800 s with 949 timeouts. JustRTP struggled with heavy chunk loading and also reached the 1,800 s phase cap. HuskHomes and LeafRTP completed all attempts cleanly.

Plugin Success / Attempts Achieved rate Latency (p50 / p95 / p99 / max) Chunks / att (attr / inc) Region CPU / att Process CPU / att Regional TPS (Min / Avg) Ticks < 20 TPS GC Pause (Phase)
LeafRTP 4,096 / 4,096 (100 %) 4.83 TP/s 137 / 203 / 206 / 399 ms 32.40 / 69.27 60.38 ms 368.36 ms 8.52 (19.89) 0.88 % 7.3 s (102)
HuskHomes 4,095 / 4,096 (99.98 %) 4.63 TP/s 404 / 785 / 1,065 / 4,928 ms 36.08 / 42.40 109.31 ms 407.94 ms 15.85 (19.97) 0.63 % 6.5 s (159)
JustRTP 4,056 / 4,087 (99.24 %) 2.25 TP/s 901 / 3,050 / 4,746 / 5,004 ms 136.62 / 138.47 287.32 ms 1,149.74 ms 6.33 (19.81) 5.17 % 16.8 s (301)
EzRTP 1,415 / 2,364 (59.86 %) 0.78 TP/s 455 / 4,995 / 5,010 / 5,151 ms 23.87 / 25.28 291.03 ms 580.37 ms 6.09 (19.98) 0.32 % 1.6 s (89)
BetterRTP 0 / 0 (0.00 %) - - - - - - - Pruned (async chunk error on region thread)
JakesRTP - - - - - - - - Unsupported (no Folia compatibility declared)

LeafRTP reduced gross process CPU by 31.5% compared to Paper (368.4 ms vs 537.1 ms) by utilizing Folia's parallel regional threads without thread locks. JustRTP CPU surged to 1,149.7 ms (+53% vs Paper) loading 136.6 chunks per attempt. EzRTP's low GC and tick degradation reflects its idle state during 949 consecutive 5-second timeouts on locked accounts. Analysis: PRE_WRITEUP.md section 5.9.


Paper 26.1 - unpaced dispatch (20260617-232754) [Historical]

The per-player gap set to 0: 3 clients, up to 4 teleports in flight, each client sends the next /rtp as soon as the last one lands. ~600 s per plugin.

Plugin Att / Succ TP/s p50 p95 p99 Min TPS MSPT p99 / max
RTP (Pro) 16560 / 16560 18.7 1 ms 2 ms 46 ms 17.5 85.8 / 98.4 ms
EzRTP 7137 / 7018 13.1 30 ms 189 ms 322 ms 10.3 156.8 / 292 ms
BetterRTP 1633 / 1606 6.0 480 ms 3 217 ms 4 402 ms 2.5 859 / 3 663 ms

About 95 % of RTP teleports are served from the pre-verified queue; the slowest 1 % pay one bounded async chunk load. The Paper adapter and engine are identical in Lite.


Folia 26.1 - region safety (20260617-191448, Lite: 20260617-205614) [Historical]

Metric RTP-Pro RTP-Lite EzRTP
Throughput (TP/s) 13.5 12.5 5.3
Main-thread CPU / attempt 3.96 ms 4.15 ms 6.34 ms
Process CPU / attempt 199 ms 177 ms 593 ms
Folia watchdog stalls 0 0 7 (worst region 20.4 s)

The stalls are from the server's own latest.log, independent of the harness: EzRTP calls synchronous World.loadChunk on region threads. The stall count reproduced at 7 across two runs (worst 21.0 s and 20.4 s).


Chunk loads and CPU per teleport - Folia 26.1, 6 GB pinned heap (20260921-134507, 4,096 teleports per plugin) [Historical]

Chunk loads are counted from the server's chunk-load events, so they include the area the server loads around every arrival (a 5x5 block, up to 25 chunks) on top of whatever the plugin loads while searching. Depending on the world, close to half of random candidates are unsafe, so a plugin that loads a chunk to check it pays for the rejects too. RTP checks candidates against region files first, so under 5 % of them ever get loaded, and it pins one chunk: the destination.

Metric RTP EzRTP JustRTP
Successes 4,096 / 4,096 4,030 / 4,096 4,020 / 4,096
Throughput (TP/s) 13.9 6.7 2.5
Chunks loaded / teleport 26.1 22.1 75.1 (2.9x RTP)
Process CPU / teleport 182.4 ms 208.9 ms 729.0 ms (4.0x)
Region-thread CPU / teleport 7.9 ms 21.4 ms (2.7x) 79.6 ms (10x)
JVM allocation / teleport 85.3 MB 190.4 MB 569.0 MB
Total GC pause (phase) 21.7 s 47.9 s 138.8 s

EzRTP loads slightly fewer chunks per teleport than RTP; its cost shows up on the region thread instead. In the unpaced 4,096-teleport Paper 26.1 run on the storefront, the ratio against JustRTP was larger (about 17 vs 76 chunks per teleport).


Where players landed - Paper 26.2 (20261008-003310, 4,096 teleports per plugin, 16k radius)

Full landing analysis across all 6 contenders on Paper 26.2. 16,384-block radius, 1,024-block void around spawn. Generated from destination scatter logs via CrossPluginDestinationScatterVisualizerTest.

Plugin Landed / Unique Exact duplicates Pairs within 48 blocks Nearest pair Clark-Evans R Chunks reused Landing block check (safe share)
LeafRTP 4,096 / 4,096 0 (0.0 %) 0 (random: 82) 75.2 blocks 0.977 0 99.5 % (547 / 550 safe, 3 noFloor)
JakesRTP 4,096 / 4,096 0 (0.0 %) 108 (random: 82) 6.7 blocks 0.919 1 78.8 % (2,745 / 3,484 safe, 730 noFloor, 9 hazard)
BetterRTP 4,096 / 4,094 2 (0.05 %) 123 (random: 78) 0.0 blocks 0.906 5 99.7 % (4,085 / 4,096 safe, 3 noFloor, 6 hazard)
HuskHomes 4,096 / 4,096 0 (0.0 %) 136 (random: 85) 2.2 blocks 0.907 4 98.7 % (2,429 / 2,462 safe, 32 noFloor, 1 water)
EzRTP 4,096 / 4,096 0 (0.0 %) 153 (random: 125) 3.6 blocks 0.914 4 87.9 % (3,599 / 4,096 safe, 441 water, 49 blocked, 7 hazard)
JustRTP 4,088 / 4,088 0 (0.0 %) 115 (random: 78) 4.1 blocks 0.897 1 100.0 % (8 / 8 safe)

Where players landed - Paper 26.2, paced (Test B: 20261008-024412, 4,096 teleports per plugin, 16k radius)

Successful landings only. "Random" is the near-pair count of a random scatter with each plugin's own landing intensity over terrain class and radius, so water and biome rejection are already in the expectation.

Plugin Landed / Unique Exact duplicates Pairs within 48 blocks Nearest pair Clark-Evans R Chunks reused Landing block check (safe share)
LeafRTP 4,096 / 4,096 0 0 (random: 81) 70.0 blocks 0.978 0 99.7 % (4,083 / 4,096 safe, 6 noFloor, 5 water, 2 blocked)
JakesRTP 4,096 / 4,096 0 127 (random: 84) 4.5 blocks 0.897 2 77.2 % (186 / 241 safe, 55 noFloor)
BetterRTP 4,096 / 4,096 0 126 (random: 84) 2.2 blocks 0.899 7 99.7 % (4,082 / 4,096 safe, 12 hazard, 2 noFloor)
HuskHomes 4,096 / 4,096 0 105 (random: 82) 2.2 blocks 0.920 5 98.9 % (2,275 / 2,301 safe, 26 noFloor)
EzRTP 4,096 / 4,096 0 142 (random: 129) 4.1 blocks 0.917 3 88.5 % (3,625 / 4,096 safe, 424 water, 37 blocked, 10 hazard)
JustRTP 4,094 / 4,094 0 129 (random: 85) 4.0 blocks 0.901 7 100.0 % (8 / 8 safe)

Where players landed - Folia 26.2, paced (Test B: 20261008-052141, 4,096 teleports per plugin, 16k radius)

Successful landings only. 16,384-block radius, 1,024-block void around spawn.

Plugin Landed / Unique Exact duplicates Pairs within 48 blocks Nearest pair Chunks reused Landing block check (safe share)
LeafRTP 4,096 / 4,095 1 (0 in live) 16 (0 in live) 50.8 blocks (live) 1 (0 in live) 99.9 % (4,093 safe, 2 noFloor, 1 suffocating)
HuskHomes 4,095 / 4,095 0 108 4.5 blocks 5 98.9 % (4,049 safe, 46 noFloor)
JustRTP 4,056 / 4,056 0 96 1.4 blocks 7 99.7 % (4,044 safe, 12 hazard)
EzRTP 1,415 / 1,415 0 15 14.3 blocks 0 87.8 % (1,242 safe, 154 water, 17 suffocating, 2 hazard)
BetterRTP - - - - - Pruned during warmup (region={null})
JakesRTP - - - - - Unsupported (no Folia support)

Note on LeafRTP: All 16 close pairs occurred during startup queue synchronization (the first 100 attempts); across all 3,996 post-startup attempts, there were 0 pairs within 48 blocks and the nearest pair was 50.8 blocks. EzRTP landed 10.9 % of all successful teleports directly into ocean/river water surfaces (154 water landings).


Where players landed - Paper 26.1 (20260923-000854, 4,096 teleports per plugin) [Historical]

Plugin Exact duplicates Pairs within 48 blocks Nearest pair Clark-Evans R
RTP 0 0 78.4 blocks 1.04 (uniform)
EzRTP - 107 0 blocks -
JustRTP 167 (4.1 %) 324 0 blocks 0.83 (clustered)

Methodology

  • 3 real OPed clients, concurrency 4, cooldowns / delays / countdowns zeroed in every plugin, queues enabled where the plugin offers them.
  • Phase length ~600 s per plugin with a 240 s settle gap, 30 s warm-up; the 4,096-teleport runs use a fixed quota instead.
  • Rig: Ryzen 9 3900X, one rig for every row. Heap 16 GiB except the pinned 6 GB run.
  • CPU is measured directly from the JVM (CpuSampler: process CPU and main/region-thread CPU per phase, divided by attempts). The cpu_ms_with_chunks* columns use an uncalibrated chunk-cost constant and are not quoted.
  • Numbers are not predictions for your server - hardware, view distance, how much of your world is unsafe, whether it is pregenerated, other plugins, and player count will move them.

What this benchmark does not claim

  • It does not claim EzRTP, BetterRTP, or JustRTP are "bad". They are tested at their default queue configurations against RTP at its recommended one.
  • It does not extrapolate beyond 3-4 concurrent clients.
  • It does not measure correctness, safety, or claim-plugin compatibility - only dispatch-to-arrival latency, per-attempt cost, success rate, and landing spread.
  • The region-file precheck needs terrain that already exists; chunks that have never been generated have to be generated whichever plugin asks.