Yes, I think that's a pretty accurate answer - don't swap.
The problem is that on a steady-state network, where users tend to cycle through the
same sets of hosts, you see that deadly swapping pattern, where ntop is swapping out
something that will be referenced shortly to reference something that was swapped out.
While we could waste some space to make the structures multiples of page size, when we
do the scans (such as idle hosts), we walk the entire structure and there's really no
way to get around that, without the overhead of a doublely linked list...
-----Burton
---------- Original Message ----------------------------------
From: "Horta, Benny" <[EMAIL PROTECTED]>
Reply-To: [EMAIL PROTECTED]
Date: Wed, 25 Sep 2002 02:00:27 -0400
> Very stable so far, 2.1.50 is running solid on a WAN circuit that peaks at
>28mbit at times (averages 9mbit), no problems compiling, so far no memory
>leaks. hash size has been 8k and steady at 8k. not only is 2.1.50 solid its
>running on a box doing duty as a IDS, running in total.
>snort,mysql,httpd,acid,ntop 2.1.50 the box is a dell 1.5ghz p4 desktop,
>512mb memory. 2.1.50 is stable in my book :) but the magic rule is dont
>swap, if you swap you need memory put as many dimms as needed till swap
>shows zero. for big circuits VM is not acceptable for performance especially
>when the hashsizes start moving in the 40k range :) I learned the hard way
>trying to run it on a 90megabit circuit with only 512mb, I would venture 2GB
>of ram is required for a big line with lots of nodes.
>
>
__________________________________________________
D O T E A S Y - "Join the web hosting revolution!"
http://www.doteasy.com
_______________________________________________
Ntop mailing list
[EMAIL PROTECTED]
http://lists.ntop.org/mailman/listinfo/ntop
