Request for next net_kernel release: new switch "net_setuptime_millis"

WILLIAMS Dominic <>
Fri Jan 9 16:54:17 CET 2004


Patrik Nyblom wrote:

> TCP/IP is unfortunately not a protocol made for realtime
> applications... That's what all this long timeout hazzle comes from.
> [...]

Absolutely. My advice is not, under any circumstances, to reduce those
timeouts. Unfortunately, you must design your applications on the
assumption that it is not possible to detect their absence any faster.

I have played with those timers under various breeds of middleware,
lastly CORBA, and got bitten every time.

I think that over TCP/IP, it is even a mistake to try and design the
software to adapt to network faults and partitions. Design the network
itself to be more available than you need your software to be (using
redundant channels, AFT and such things) and just design the software
on the assumtion that the network works (i.e. let it fail when the
network doesn't work, don't try anything smart, then give up and wait
for the network to come back up.

Regards,

Dominic Williams.



More information about the erlang-questions mailing list