[c-nsp] ARP strangeness

Wed Jan 19 02:47:20 EST 2011

Keegan:

You're correct - without broadcast support, re-population initiated from the
7609 is impossible.  Once it's expired, the FTTH access gear's design, which
blocks broadcast traffic, makes it impossible for the CPE to respond to the
broadcast ARP.  The FTTH access gear never allows broadcast traffic to
ingress from the 7609.  So the only thing that can re-populate the 7609's
ARP cache is an ARP request by the CPE, *but* the CPE only does that after a
DHCP exchange after power on, never again, even after a full DHCP exchange.

Frank

From: Keegan Holley [mailto:keegan.holley at sungard.com] 
Sent: Tuesday, January 11, 2011 2:07 PM
To: rodunn at cisco.com
Cc: frnkblk at iname.com; cisco-nsp at puck.nether.net
Subject: Re: [c-nsp] ARP strangeness

On Tue, Jan 11, 2011 at 2:50 PM, Rodney Dunn <rodunn at cisco.com> wrote:

On 1/11/11 11:49 AM, Keegan Holley wrote:

Possibly a stupid question, but I thought ARP had to be broadcast
because the mac address of the destination was unknown.

That is true for the first request. For subsequent arp refreshes the most
efficient way is to unicast it.

That's what I said.  However, having an ethernet that doesn't support
broadcast would make that first entry impossible.  The problem would repeat
after reboots or power outages.

 If the CPE has

the correct mac address to unicast an ARP request, why would it need to
arp in the first place?

To make sure the CPE is still alive and responding.

RIght but it can only do that if it has already successfully arp'd so I
think we are saying the same thing.

 I suppose I can understand renewing the entry

via unicast, but there will always be a need for broadcast ARP.  It just
seems strange to implement an ethernet based device and block all
broadcast traffic. Am I missing something?

I didn't go back and read the entire thread to remember what was blocking
the broadcast and understand the thought logic of whey they do it. It may
would be, and possibly has some logic to it, that in a CPE environment you
always rely on traffic originated at the CPE to trigger the arp so you block
broadcast out to the CPE not coming from the CPE.

That doesn't make sense though.  The cpe will need to broadcast for the
initial request and after reboots regardless of what the provider router
does.  The device that was blocking broadcast was a third party FTTH device.
I get the feeling I'm missing something here though.  Maybe it only allows
broadcast for a specific interval after it detects a link down/link up. 

On Mon, Jan 10, 2011 at 5:27 PM, Frank Bulk <frnkblk at iname.com

<mailto:frnkblk at iname.com>> wrote:

   Thanks for explaining.

   Since the Linksys BEFRS41 does not ARP regularly, even after a DHCP
   RENEW
   and DHCP DISCOVER, and because the access gear blocks all broadcast
   traffic,
   the 7609-S will never (re-)populate its ARP entry.

   I'm going to see if the Linksys BEFRS41 has a configurable ARP
   expiration
   timer.  If so, dropping it to 10 minutes would cause it to unicast
   ARP for
   the default gateway, which would resolve the issue.

   Another possible option, I guess, is to extend the 7609-S ARP
   expiration to
   a longer time interval, but if the BEFRS41 is silent for even a second
   longer than the ARP timer, then I'm still stuck.

   I should really look at the behavior of other CPE to see how often they
   unicast ARP.

   Frank

   -----Original Message-----

   From: Rodney Dunn [mailto:rodunn at cisco.com <mailto:rodunn at cisco.com>]
   Sent: Monday, January 10, 2011 1:30 PM

   To: frnkblk at iname.com <mailto:frnkblk at iname.com>
   Cc: cisco-nsp at puck.nether.net <mailto:cisco-nsp at puck.nether.net>
   Subject: Re: [c-nsp] ARP strangeness

   It only gets updated on getting and ARP packet from the host.

   It is not updated based on L3 data level traffic flowing to/from the
   host.

   Rodney

   On 1/10/11 11:43 AM, Frank Bulk wrote:
    > Does the ARP cache get populated, or updated, if the traffic
   comes into an
    > L3 interface, or is it only populated upon a successful ARP response?
    >
    > Frank
    >
    > -----Original Message-----
    > From: cisco-nsp-bounces at puck.nether.net
   <mailto:cisco-nsp-bounces at puck.nether.net>
    > [mailto:cisco-nsp-bounces at puck.nether.net
   <mailto:cisco-nsp-bounces at puck.nether.net>] On Behalf Of Rodney Dunn
    > Sent: Wednesday, January 05, 2011 7:38 AM
    > To: Jeff Kell

    > Cc: cisco-nsp at puck.nether.net <mailto:cisco-nsp at puck.nether.net>
    > Subject: Re: [c-nsp] ARP strangeness
    >
    > On 1/4/11 11:43 PM, Jeff Kell wrote:
    >> On 1/4/2011 9:01 PM, Rodney Dunn wrote:
    >>> There were some changes to ARP at one point to provide some more
    >>> triggered capability. I don't recall exactly what that was but the
    >>> default behavior for many years was that we send a unicast arp
   to the
    >>> destination 60 seconds prior to the arp timer set to expire. If we
    >>> don't get a response we send it again when the timer pops and if no
    >>> response we invalidate the ARP entry.
    >>
    >> Umm, that sort of rocks my boat with regard to network monitoring
    >> assumptions...
    >>
    >> We have one of those NMS systems that periodically "reads L2
   devices for
    >> mac-address/port mapping" and "reads L3 devices ARP for mac-to-IP
    >> mapping".  Ideally, there should be no missing links (if the MAC is
    >> found, hopefully the ARP/IP is found, and vice-versa).
    >>
    >
    > That still holds true as long as a timer (sam cam ager) didn't pop
    > sooner than your arp refresh timer.
    >
    >> For the default mac-address aging time of 300 seconds, I had
   this notion
    >> that setting the ARP timeouts to 270 seconds would necessitate the
    >> router ARPing the device (assuming active traffic) prior to the
    >> mac-address aging out, keeping the mac-address table populated.
    >
    > Keep the other timers 60+ seconds out to be safe.
    >
    >>
    >> But if the Cisco L3 behavior is to gratuitously do this for me
   before
    >> the ARP timeout... that changes things a bit.
    >>
    >> Is this default behavior across all the Cats, or just the
   6500/7600?  Is
    >> it supervisor-specific?
    >>
    >
    > Traditionally generic to all of IOS. There may have been some
   platform
    > specific thing that changed here that I have missed in the last
   couple
    > of years though.
    >
    > Rodney
    >
    >> Jeff
    > _______________________________________________
    > cisco-nsp mailing list cisco-nsp at puck.nether.net

   <mailto:cisco-nsp at puck.nether.net>

    > https://puck.nether.net/mailman/listinfo/cisco-nsp
    > archive at http://puck.nether.net/pipermail/cisco-nsp/

   _______________________________________________
   cisco-nsp mailing list cisco-nsp at puck.nether.net

   <mailto:cisco-nsp at puck.nether.net>

   https://puck.nether.net/mailman/listinfo/cisco-nsp
   archive at http://puck.nether.net/pipermail/cisco-nsp/