Um... sorry, I was thinking you were talking about the netflow packet
timeout (missed the asymetric nature of your connection)...

CPU 100% could be a symptom of two bugs we've found - one in the dns
resolution thread and another in the netflow (sflow?) setup if the listening
socket is zero.  Both are fixed in the latest CVS, but there is a big
problem in there with daemonizing that renders ntop unusable (after the
04Apr2002 snapshot).

I think I just tagged the daemonizing bug - I'm testing right now on another
session, so... give me a few minutes and (if I'm right) I will post the
patch that would let you use the versions w/o the cpu usage bug... (e.g.
latest CVS)... that way you'll really see what your CPU usage is...

The things to look for are:

1) ntop should not be using swap space - if you don't have enough memory to
keep the structures in memory, then reporting (which walks the structures)
will be very painful

and

2) < 100% CPU usage

-----Burton

-----Original Message-----
From: Christian Hammers [mailto:[EMAIL PROTECTED]]
Sent: Thursday, April 11, 2002 9:30 AM
To: Burton M. Strauss III
Cc: [EMAIL PROTECTED]
Subject: Re: [Ntop-dev] still have missing data in netflow export


On Thu, Apr 11, 2002 at 09:20:55AM -0500, Burton M. Strauss III wrote:
> That said, a big enough box should be able to handle 3x100 Mbps, but be
> aware that with three SATURATED links you're getting close to the memory
> bandwidth of many Intel Pentium III computers. (But my memory is fuzzy
about
> the details here).
How will I notice that? I guess at least the CPU will be busy handling that
amount of data and the interrupts per seconds number will raise very high,
too, so I am far away right now:

  procs                      memory    swap          io     system cpu
r  b  w   swpd   free   buff  cache  si  so    bi    bo   in    cs  us  sy
id
1  0  1      0   5588  11852  74068   0   0     0    45  218   119   1   1
98
0  0  0      0   5588  11852  74068   0   0     0     4  283   117   0   5
95

bye,

-christian-

P.S.: The timeout thing I was looking for is in session.c's
      scanTimeoutedTCPSessions(), esp. the following thing, as I have
      only "lost packets" in asymetric routing:

      if(((thisSession->sessionState == STATE_TIMEOUT)
         && ((thisSession->lastSeen+TWO_MSL_TIMEOUT) < myGlobals.actTime))
         || /* The branch below allows to flush sessions which have not been
               terminated properly (we've received just one FIN (not two).
               It might be that we've lost some packets (hopefully not). */

--
Christian Hammers    WESTEND GmbH - Aachen und Dueren     Tel 0241/701333-0
[EMAIL PROTECTED]     Internet & Security for Professionals    Fax 0241/911879
          WESTEND ist CISCO Systems Partner - Authorized Reseller

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

Reply via email to