[lug] Quote of the Day

[Sean Reifschneider]

On Tue, Jan 30, 2001 at 06:38:49PM -0700, D. Stimits wrote:
>Stateless is a better descrip. Hits are serviced, and it then
>disconnects (options can be used to force a maintained state, if the
>server allows it, but is not the norm here).

When people talk "connectionless", they typicly are talking about
UDP something like DNS or NTP or DHCP.  A request is sent to a particular
port on a remote box, and it uses the "source port" in the header to
fashion a reponse.  Any retransmits are done by the application.

With TCP, you must first establish a connection (which involves
several packets sent back and forth), THEN you can start sending
data over it.  The connection is held open until shortly after one
end closes it (there's some handshaking done on closes as well).

>Yes, this other person made the assumption that the design was because
>of an inability for unix to thread, or inferior threading...it was
>stated that windows uses threads in its IIS because of the superior
>implementation there (I was thinking about selling him some swamp land).

Linux has historicly had issues with threads, but that's mostly fairly
old news.

>This person stated that the pre-forking was an indication of the
>inferior ability of unix to properly thread. You are allowed to gag or

Creating a new thread requires some OS overhead, so unless IIS is doing
pre-threading it may be that the pre-forking is faster.  Having an existing
process handle the new connection being lighter-weight than creating a
new thread, which then handles the request...  I haven't done any benchmarks
here, but it seems possible.

Sean
-- 
 Whenever possible, steal code. 
                 -- Tom Duff
Sean Reifschneider, Inimitably Superfluous <jafo at tummy.com>
tummy.com - Linux Consulting since 1995. Qmail, KRUD, Firewalls, Python