Ashley, nothing is going to change in the code unless YOU are willing to
work the problem.

We've bounced this back and forth a couple of times, and you've never
responded to the questions about the # of hosts present on and the # of
hosts being accessed from your network.  (Yes, you indicated a /20 and /24,
but how many of the 17K are active?  How many active sessions?? etc.)

As somebody pointed out, one busy Kazza user can access 40K hosts searching
for a rare file.

Each host eats up a fairly amount of memory, because we have to track all of
the 64K possible ports.

Action items:

1. Post the data from textinfo.html as close to when it dies as possible...
I'm interested in

Host Memory Cache

Limit.....#define MAX_HOSTS_CACHE_LEN 512
Current Size.....52
Maximum Size.....69
# Entries Reused.....1099

and

Host Hash counts

Actual Hash Size.....512
Stored hosts.....20 [3 %]
Purge idle hosts.....Enabled
Purged hosts.....1151
Maximum hosts to purge per cycle.....512
DEFAULT_MAXIMUM_HOSTS_PURGE_PER_CYCLE.....512

2. Run under gdb and post the bt full trace so we can see where it's
dying...


3. Try using the large network flags ... -g | --track-local-hosts, -z
| --disable-sessions and/or -C | --large-network (the first 2 are documented
in docs/FAQ).


4. You can use a sneaky trick to estimate how big the host population ntop
is monitoring is by using rrd!

(Turn it on in the plugins menu and make sure to enable hosts - remember you
may have to create the directory and give the -u user w/r access to it).

If you delete all the files 'tween runs, then when ntop dies, the # of
entries in the rrd/hosts directory is an upper bound on the # of hosts ntop
is monitoring. If you subtract the # of purged items (Purged hosts.....) you
grabbed from info.html just before it dies, that will give you a lower
bound.



-----Burton


-----Original Message-----
From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED]]On Behalf Of
Ashley Gates
Sent: Thursday, January 23, 2003 4:52 PM
To: [EMAIL PROTECTED]
Subject: [Ntop] Hash size / memory gobbler


I've taken the advice to compile and use the latest source to ntop.

I now currently run:
ntop v.2.1.55 MT (SSL) [i686-pc-linux-gnu] (01/21/03 01:32:38 PM build)

The hash size problem still remains, now it takes even longer for ntop
to eat 1 gig of RAM and seg fault.

Previously it was 30-40 minutes, it now takes over an hour.

Good work so far on getting this gremlin under control, but alas I
cannot trust it for what it needs to be for me.. :-)

I will continue to compile the latest source once a week and keep
updating on the issue.

Thanks,

Ashley Gates
Director of Network Operations
Atlanta Broadband, Inc.
http://www.atlantabroadband.net
Office: 770.794.0767
_______________________________________________
Ntop mailing list
[EMAIL PROTECTED]
http://listgateway.unipi.it/mailman/listinfo/ntop

_______________________________________________
Ntop mailing list
[EMAIL PROTECTED]
http://listgateway.unipi.it/mailman/listinfo/ntop

Reply via email to