@taiyo
It so unpredicted as the load from having followers on many different instances combined with forwarding of replies can result in long delays with outgoing events. It needs somehow for small instances to be able to pass this load to a paid distribution service so instances don't need to be sized for peak distribution load.
Notices by IanStockport (ianstockport@mstdn.social)
-
Embed this notice
IanStockport (ianstockport@mstdn.social)'s status on Saturday, 26-Nov-2022 06:26:40 JST IanStockport -
Embed this notice
IanStockport (ianstockport@mstdn.social)'s status on Saturday, 26-Nov-2022 06:26:39 JST IanStockport @taiyo These large distribution servers could keep connections open to most mastodon instances so advoid the TCP connection create overheads etc. Would need to be written in a language that is good at async IP with a runtime that has full mulple threading. I would choose C# as I know it, but Erlang/GP would also work.
-
Embed this notice
IanStockport (ianstockport@mstdn.social)'s status on Saturday, 19-Nov-2022 22:22:56 JST IanStockport @taiyo Why can't translation be done by jscript running in the user browser redirecting to something like Google translate?
-
Embed this notice
IanStockport (ianstockport@mstdn.social)'s status on Saturday, 19-Nov-2022 22:20:18 JST IanStockport @taiyo The issue is when a 101 small instances are all running in the same amazon data center and all paying to store the same file.
Maybe the file should get copied to an instance after a set number of views by people on that instance.
-
Embed this notice
IanStockport (ianstockport@mstdn.social)'s status on Saturday, 19-Nov-2022 21:28:55 JST IanStockport @taiyo Every instance paying do their own datastore that is mostly filled with the same readonly files as other instances seems pointless, when it should be possible to just play videos etc from their original source.