Repository navigation
v4.5.1
Fleet uses the network only while its window is on screen. It used to poll
every machine every 3 s for as long as the app ran, and each poll opened a new
TLS connection: about 11 KB a poll between two Macs, around 9 GB a month per
machine, most of it handshake rather than data. Now:
- Window closed or minimised: nothing. No machine is polled and every
connection is closed. Finding machines on your LAN keeps running. - Window open: one connection per machine, kept open between polls, and the
reply is gzipped — 7.7 KB → 2.4 KB from a Mac, 1.5 KB → 0.6 KB from a Linux
box. - Window back: polling resumes at once, and the charts show a break where
nothing was measured instead of joining two readings hours apart.
Measured against a cloud agent, one minute each: 106 KB with the window open in
4.5.0, 29 KB now before the agent was updated, nothing with the window closed.
Update your agents to 1.4.0 for the compression, and for a Mac agent to keep
the connection open. (#71,
diagnosed and measured by @fatcutegg)
- A Mac's Fleet tile said "ANE 0% · BW 271 GB/s", which read as the Neural
Engine's own traffic. It is the whole chip's memory bandwidth, and now says
"Mem BW".