Skip to Content
teigeodata
Data

Measured

Sizes and speed as the gates measured them, each with its date, machine and commit.

Generated from the gates

This page is written by cargo xtask data-pages from the records in site/data/measured, each written by the gate that measured it (--record). Every figure says when, where and at which commit it was measured: compare a timing only with timings from the same machine, and read one from a busy machine (a high load average) as a ceiling, not a typical value.

The bytes of each component of the released pack are on the coverage page; how good the data is, and how it is gated, is on accuracy.

WebAssembly packages

Each npm package's module after wasm-opt -Oz, raw and gzipped at level 9. cargo xtask package-wasm fails a build whose gzipped module is over its budget.

Measured 2026-11-15 on Darwin arm64, 10 cores (load average 10.36), commit 25ad7e5, by cargo xtask package-wasm --record.

PackageHoldsModulegzip -9BudgetHeadroom
@teispace/teigeodata-wasmevery component605118 B253910 B262144 B3.1%
@teispace/teigeodata-wasm-lookupevery query at a point or by ID380167 B145724 B153600 B5.1%

Library bytes per feature set

What each feature set of the runtime crates adds to a linked library (tools/size-probe, a baseline that any host library links subtracted), natively in the release profile and for wasm32 in the size profile.

Measured 2026-11-15 on Darwin arm64, 10 cores (load average 7.04), commit 25ad7e5, by cargo xtask sizes --record.

Feature setnative Bnative gzipwasm32 Bwasm32 gzip
open: head checks, section lookup16.4 KiB12.3 KiB18.4 KiB8.4 KiB
open + BLAKE3 of every section16.5 KiB13.5 KiB20.2 KiB9.2 KiB
open + Ed25519 signature check32.7 KiB27.8 KiB47.6 KiB20.0 KiB
open + every section view (places, hierarchy, ID map, names, components)64.9 KiB36.7 KiB103.5 KiB25.8 KiB
open + BLOBS (ruzstd)65.0 KiB43.3 KiB124.0 KiB34.5 KiB
the whole reader: all of the above129.7 KiB74.1 KiB171.8 KiB53.6 KiB
the writer alone (forge side)16.5 KiB16.3 KiB26.1 KiB11.2 KiB
the engine, points only: open, get, reverse113.5 KiB67.7 KiB143.4 KiB43.8 KiB
the engine, every query452.4 KiB248.1 KiB459.8 KiB184.1 KiB

Baseline (subtracted): native 278.5 KiB, wasm32 16.1 KiB.

Scaling across threads

Throughput of one shared engine on np-places-lite.tgd: each thread takes the current snapshot for every query, as a server's handlers do, and runs get, reverse (10 nearest), admin_at, tz_at and search (10 hits) in turn. The best of three rounds per thread count. The gate fails when four threads reach less than 1.5× one thread, the mark of a lock on the query path; the machine is shared and mixes performance and efficiency cores, so the figures are a floor.

Measured 2026-11-15 on Darwin arm64, 10 cores (load average 6.54), commit 668ec58, by cargo xtask scaling np-places-lite.tgd --record.

ThreadsQueries/sSpeed-upEfficiency
12077391.00×100%
24161202.00×100%
48111433.90×98%
812676336.10×76%

Server load

tgd serve on loopback, each route driven by oha 1.17.0 for 5 s at each point, twice; round two is reported, and the footprint's growth across it is the leak check. tools/load/budget.toml sets each point's floor and ceiling.

Measured 2026-11-15 on Darwin arm64, 10 cores (load average 22.79), commit 3688d14, by cargo xtask load <pack> --check tools/load/budget.toml --record.

np-places-lite

Routecreq/sp50 msp99 msrequestsfailed
search147930.1341.843239790
search16642150.1911.0543211810
search64611820.9133.3903060840
reverse16710360.1741.0073553210
get16842910.1540.8004217320
near16633040.1941.1103166280
hierarchy16901830.1570.5304510850
boundary1688751.5518.369443740
nominatim16534440.2561.0112673800
mcp search16393150.3701.2161966740

footprint: 5616 KB idle, 16384 KB peak, -3072 KB across round two

world-baseline-lite

Routecreq/sp50 msp99 msrequestsfailed
search141290.1031.951206490
search16734810.1620.9283675810
search64810550.6762.8754055020
reverse16581120.1871.3252906560
get16786540.1490.9303934020
near16535360.1901.4952677870
hierarchy16826380.1420.9504134130
boundary16319800.3013.0291599440
nominatim16698880.1721.0293496650
mcp search16427500.2921.4982137640

footprint: 5072 KB idle, 10240 KB peak, +0 KB across round two

Phone-class estimates

The phone-class bench (P1, G-phone) on world-np-lite.tgd: open, the first query of each kind, search a keystroke at a time and the point queries, each the minimum of several rounds here and multiplied by 6.5 (this machine's core over a low-end phone's), and the heap counted by allocator; the gate fails when an estimate is over its target. A point query is reverse_in_country (10 nearest), admin_at, tz_at and elevation_at; the page cache is warm, as on a relaunch.

Measured 2026-11-15 on Darwin arm64, 10 cores (load average 7.63), commit 41ba69a, by cargo xtask phone world-np-lite.tgd --record.

MeasureHerePhone estimateTarget
open133.75µs869.38µs50ms
first queries6.06ms39.40ms150ms
first get809.79µs5.26ms–
first reverse2.21ms14.39ms–
first admin_at458.00ns2.98µs–
first tz_at31.62µs205.56µs–
first elevation_at286.42µs1.86ms–
first search2.57ms16.68ms–
keystroke p509.25µs60.12µs–
keystroke p99203.46µs1.32ms16ms
keystroke max235.00µs1.53ms50ms
point query p9976.00µs494.00µs2ms
get p99250.00ns1.62µs500µs
derive SUBSTRING389.03ms2.53s3s
heap after open and the first queries0.05 MiB4 MiB
heap peak0.05 MiB16 MiB
heap peak while deriving SUBSTRING55.00 MiB64 MiB

On this page