Neohapsis is currently accepting applications for employment. For more information, please visit our website www.neohapsis.com or email firstname.lastname@example.org
From: Yasuhiro Ohara (yasu_at_SFC.WIDE.AD.JP)
Date: Wed Feb 26 2003 - 07:36:02 CST
> for L2 detection....
> i have some questions...
> how long is it that when an interface goes down, does the kernel report it?
> is it instantaneous? i mean i lost clock for a microsec and i came back up..
> does it filter up immediately?or is there "a delay" to see if its really
> down there too?
> if it is instantaneous, could the routing daemon "lag" this by as many
> seconds rather than send out packets to test it?
I don't know how they intend to do but some kind of flap damping
technology as in BGP route flap damping will be necessary, at L2 level.
> > Second is about the control traffic load of Hello. In your example of
> > 100ms hello interval on 30 adjacencies, the load will be 403.2Kbps,
> > even with calculating standard OSPF hello format (which may be somewhat
> > redundant). I guess this cost is worthy to pay, in order to get
> > robust and fast convergence network.
> ?????????? err no, ISPs are usually not that accomodating in the developer's
> favour :-)
> choking up the bandwidth is a good point
do you mean the router's backplane by "bandwidth" ?
> (are u saying those hellos are out of band)?
I mean Hello calculation must not necessarily be done
on control-plane. Hello packet will be on the band, as far as I said.
and yes, 400K may be too much.
But, the size 400K is not caused by the fact that it's on L3,
it's caused by the fact that I calculated without changing
standard Hello packet.
If we make sufficient L2 detection, we need some kind of packet format,
which includes some kind of identifier, etc. Obviously detecting the
electricity is not enough, because we cascades many LAN switches.
> fine i need faster hellos on BMA networks.
> why? coz there is multicast happening, the only way i know that a router is
> out is either if "the arp is out and doesnt come back in" or i miss hellos
> and there we have "sufficent bandwidth" coz i assume BMA is primarily ethernet...
you mean most of your failures are L2, isn't it ? that's OK.
but how much you have sufficient bandwidth ?
don't you have extra 400K bandwidth ? ;p)
> and no one runs ospf on "extended lans" via media convertors...i
> hope (hell if u have a router put in a serial link)
> but on P2P, line protocol/PPP down is a good enuf bet to imply the other end
> is down. why do we need L3 there?
- L2 detection is not so better than L3 detection,
if both made by hardware
- L3 detection is (thought you're saying none) a little bit better
than L2 deteciton
- you must develop a detection method for each of L2 technology
(again I don't believe electricity checking is not enough)