Text 1939, 53 rader
Skriven 2005-01-04 22:27:59 av Russell Tiedt (5:7105/1)
   Kommentar till text 1930 av Paul Rogers (1:105/360.0)
Ärende: Router/Firewall Advice
==============================
Hello Paul.

03 Jan 05 17:33, you wrote to me:

 PR>> My general assumption is the CPU is the last place to look for a
 PR>> bottleneck.  With the 3C509 it would probably experience some
 PR>> idle wait time that it now uses to drive the NE2000.  But even
 PR>> so, that time is coming during packet reception/transmission.
 PR>> The time it takes to process packets won't change appreciably.
 PR>> If it had something effective it could do while it's not driving
 PR>> the NE2000 so hard, then total computing throughput might
 PR>> benefit.  I doubt there will be a measurable difference in
 PR>> network response.  IOW, it ain't broke, why fix it?

 RT>> Well, I have a web server running on a PI-200MMX which responds
 RT>> miles faster  than the web server on the 486DX4-100, both have
 RT>> identical

 PR> When I said: "The time it takes to process packets won't change
 PR> appreciably," I was talking about for a given system!!!

That is understood, basically what I was saying is that the switching hub is 
not the bottleneck, that leaves the 486 ...

 RT>> NE2000 compatable  NIC's both connected to the same switching
 RT>> hub, ie. I get a almost immediate  respons from the webserver on
 RT>> the PI-200, while I have to  wait 25 - 30 seconds for a page to
 RT>> be served from the webserver on the 486, and then another 25-30
 RT>> seconds later the speaer beeps to indicate that a connection has
 RT>> been made.

 RT>> If that is not caused by lack of cpu cycles, then I don't know
 RT>> ...

 PR> But it's not during packet transmission.  You're experiencing the
 PR> packet processing time.  Try changing the NIC then.  I expect the
 PR> relative performance of each box with a NE2000 knockoff vs a 3C509
 PR> to remain relatively consistent, providing that's all they're doing.
 PR> The question was, and remains, is there anything useful the system
 PR> could have been doing while it was driving the NIC?  Your "data"
 PR> doesn't address that at all.

Well, if I put a DMA enabled 3C509 which frees up cpu cycles from packet 
processing, the cpu will have more cycles to serve the webpages the server 
gives it to push down the wire, after all webservers require cpu services to 
operate and serve web pages don't they, and if they have more cpu cycles, all 
things being equal, they should be served faster, that is unless I have 
seriously fallen of the bus somewhere down the road. :-)

Russell

--- GoldED+/LNX 1.1.4.7
 * Origin: Rusty's BBS - Bloemfontein, Free State, South Africa (5:7105/1)