/dev/rob0 writes:
On Wednesday 15 September 2004 11:08, Arnt Gulbrandsen wrote:
Oh my heavens! A famous person! :) Did you know that you have been
immortalised in the /etc/hosts files of Slackware Linux?
Yes ;) People keep mentioning that. Hardly anyone notices e.g. 'man
mailaddr', but after eleven years I still get mail about that hosts
file. No idea why.
There's also a small performance difference. Which side is faster
depends on OS and version. Doesn't matter, anyway, they're both far
too fast to ever be a bottleneck.
I'm sure, but if you were running something like this on a large scale
you would surely want to choose correctly. 100 simultaneous TCP
connections to localhost might not be noticeable, but when you're in
the 1000 or 10000 arenas things are probably different.
Not really. It's a percentage matter, actually. If you have a connection
to e.g. a database server, then you'll need some CPU cycles to generate
a database query, the database server will need some CPU cycles to
process the query and generate the answer, and you'll need some CPU
cycles to parse the answer and use the result.
You and the database server will need much much much more CPU to do all
that than than the kernel needs to move the bytes. This is true whether
you have one connection or a thousand.
It's true that the kernel has to do a little more work if it has a _lot_
of TCP connections open. But I tried measuring it and couldn't - the
effect was too close to zero. (I did a little work speeding up the
TCP/UDP/IP code in the linux kernel years ago.)
Arnt
