Statistics

triple.isoimage.sbs · IPv4 147.45.39.32 · HTTP / HTTPS / rsync · AS209207, Frankfurt am Main · triple.isoimage.sbs

Counters are taken from the web server and rsync daemon logs, aggregated hourly and written to a text file per month. Egress figures are payload bytes delivered to clients; they exclude the traffic we pull from upstream during synchronisation, which is a further 1 to 3 TB per month. Units are decimal: 1 TB = 1,000 GB.

Egress by month

September 2026 covers 1–23 September and is incomplete. One # is about 2 TB.
MonthEgressRelative RequestsUnique clients Peak 5-min rate
2025-1031.4 TB################176.2 M288,4102.61 Gbit/s
2025-1133.8 TB#################189.7 M301,7722.84 Gbit/s
2025-1230.2 TB###############170.4 M279,3352.42 Gbit/s
2026-0135.9 TB##################201.3 M318,9043.05 Gbit/s
2026-0234.1 TB#################191.8 M309,5172.88 Gbit/s
2026-0339.7 TB####################223.1 M342,2603.24 Gbit/s
2026-0441.2 TB#####################231.5 M356,8813.47 Gbit/s
2026-0540.5 TB####################227.6 M351,0443.12 Gbit/s
2026-0643.8 TB######################246.1 M372,6193.55 Gbit/s
2026-0745.1 TB#######################253.4 M389,2053.60 Gbit/s
2026-0848.3 TB########################271.4 M412,5843.81 Gbit/s
2026-09 *37.6 TB###################211.2 M358,1273.44 Gbit/s
Total461.6 TB2,593.7 Mnot additive3.81 Gbit/s max

Client counts are distinct source addresses seen within the month and must not be added across rows: a machine that downloads updates every week is one address in every month it appears in. Roughly 23 % of the August addresses were IPv6 only.

Rates, August 2026

Total egress48.3 TB
Mean rate over the month144 Mbit/s
95th percentile rate521 Mbit/s
Peak 5-minute average3.81 Gbit/s
Time of peak2026-08-14 19:25 UTC (Ubuntu 24.04.6 point release)
Busiest day2026-08-14, 4.12 TB
Quietest day2026-08-03, 0.94 TB
Uplink capacity2 × 10 GbE LACP, policed at 10 Gbit/s
Peak concurrent HTTP connections6,412
Peak rsync sessions12 of 12
Requests271,400,000
Mean object size178 KB
Unique client addresses412,584

Traffic by repository, August 2026

Shares are rounded to one decimal and need not total exactly 100.0.
RepositoryEgressShare RequestsMean objectTree size
Ubuntu14.8 TB30.6 %89.4 M166 KB2.80 TB
Debian11.2 TB23.2 %71.8 M156 KB2.10 TB
Rocky Linux6.4 TB13.3 %28.3 M226 KB940 GB
Arch Linux5.1 TB10.6 %19.6 M260 KB84 GB
Alpine Linux3.9 TB8.1 %41.2 M95 KB78 GB
OpenWrt2.6 TB5.4 %12.4 M210 KB1.10 TB
Debian CD/ISO2.2 TB4.6 %0.6 M3.67 MB940 GB
Ubuntu releases (ISO)1.5 TB3.1 %0.4 M3.75 MB620 GB
CPAN0.4 TB0.8 %7.4 M54 KB17 GB
Other paths0.2 TB0.4 %0.3 M667 KB—
Total48.3 TB100.0 %271.4 M178 KB8.68 TB

Note the asymmetry between tree size and traffic: Alpine is 78 GB on disk and still moves 3.9 TB a month, because container builds pull the same handful of small packages over and over. OpenWrt is fourteen times larger on disk and moves less than a tenth of the requests, because a router is flashed once and then left alone. Arch produces an unusually large mean object because it ships few but large packages and its users upgrade the whole system at once.

Protocol and address family, August 2026

BreakdownEgressShareRequests
HTTPS (443)39.1 TB80.9 %241.8 M
HTTP (80)6.8 TB14.1 %29.1 M
rsync (873)2.4 TB5.0 %0.5 M
IPv429.8 TB61.7 %169.4 M
IPv618.5 TB38.3 %102.0 M
Total egress48.3 TB100.0 %271.4 M

Protocol rows and address-family rows each total 48.3 TB; they are two views of the same bytes, not five separate figures. rsync accounts for a large share of bytes per session and almost no requests, since a downstream mirror opens one session and copies a whole tree through it.

Growth

Egress has grown by about 38 % over the last twelve months, from 31.4 TB in October 2025 to 48.3 TB in August 2026, and September 2026 is on course to pass 49 TB. The growth is not evenly spread: point releases and the Alpine and Arch package churn move the line, while the ISO trees are flat outside release weeks. At the current rate the 10 Gbit/s policer is reached only during release peaks, so the next constraint is disk capacity rather than bandwidth; the array is at 54 % with no new trees planned.

Figures for closed months are not edited afterwards. If a month is corrected because of a counter reset or a log rotation fault, the correction is noted here. The last such correction was in March 2025, before the period shown above.

What is not measured