You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Speedtests that measure the line, and a dashboard search engines leave alone
An iperf3 speedtest ran on a single TCP stream by default, which
measures that stream's congestion window rather than your connection.
On a real ~350 Mbps path, 1 stream reported 116 Mbps, 4 reported 202
and 8 reported 268 - the same link, more than twice the answer. The
default is now 8, suited to a 500 Mbps - 1 Gbps line, and the ceiling
moves from 8 to 32 because that same path was still climbing at 12
with the server 93% idle. Past about twelve the differences are mostly
run-to-run noise, which the tooltip and the docs now say. The number
stays yours to pick: an instrument that retunes itself turns its own
choices into apparent ISP changes. The tester enforces the same
ceiling the settings do, so a raised count reaches the wire instead of
being accepted, stored, displayed and then quietly dropped.
The Legacy RSA padding toggle no longer promises something your own
machine cannot do. Whether that option can be sent is decided by the
iperf3 build on THIS host - it arrived in 3.17 and 3.20 reclassified
it as server-only - so on a current build, ticking it made the run
fail with a message that read like a server or credential problem. It
is now locked outside that window, with a note saying which side of it
this host falls on, and a build that cannot be identified locks
nothing. Locked against being turned on, never against being turned
off: a server saved with it already on - ticked under an older build,
or from before this host's iperf3 moved past 3.20 - keeps sending the
flag on every run, so that one stays clickable and says to turn it
off, rather than leaving the only cure as deleting the server and its
stored password. Unticking it locks the toggle again on the spot, so
the escape hatch cannot be used to put the flag back.
The warning about re-entering an iperf3 server's password stops
outliving the edit that raised it. Nothing used to clear it - not
opening another server, not closing and reopening the drawer - so it
sat there accusing a server that was saved and authenticating fine.
The message belongs to an open edit and now dies with
it - on save, on Done, on Load defaults, on opening another server -
and typing the original address back takes it away with the edit,
instead of demanding a retype for a password the box still had. What
survives is the durable signal: the password field itself says whether
this address has one saved.
A greyed-out setting no longer greys out the help tooltip inside it.
Opacity applies to everything an element contains, so a dimmed label
dimmed its own explanation to about 3.2:1 - the one thing an operator
opens precisely because the control is unavailable. The rest of the
settings list has carried an escape for this all along; the iperf3
server editor's fields now use the same one.
A dashboard reachable from the internet is no longer something search
engines will index. It is one person's view of one connection - ISP,
exit city, speed history, outage log - and because every install
renders identical titles, indexing also scattered duplicate branded
pages. The server now sends X-Robots-Tag: noindex, nofollow, which
covers /api and /metrics as well; the public demo is a separate build
on its own host and is unaffected.
Separately: the review-round labels are gone from the source. Forty-two
comments cited a finding id - C-57, audit F-11 - from a review document
no reader outside it can obtain, and three of them shipped inside the
dashboard, where every install and the public demo serve them in View
Source. Each one is now the reason the assertion exists, which is what
the label was standing in for. Nine test files named after the round
that produced them are named for what they test instead, and four tests
that were only grouped by that round moved to the file their subject
already lives in. No test was added, removed or changed: 966 before,
966 after.
Co-Authored-By: Claude <noreply@anthropic.com>
0 commit comments