Re: survivability, rewriting

[Iljitsch van Beijnum <iljitsch@muada.com>]

On 5 nov 2003, at 11:05, Erik Nordmark wrote:

> > My question is why don't just send a replicated data packet instead of
> > pings?
> > I mean, i have a hint that something is wrong, so if i have data 
> > packet
> > available, i send it replicated as many times as possible source dest
> > address combinations i have available.

> That can work as long as you can tell from the returned packets which
> of the replicated data packets made it.
> I think this is hard. If you take a ULP packet and send it with 
> different
> src/dst locators and then after one RTT there is a response (a ULP ack 
> for
> instance) how can you tell which of the src/dst locator pairs worked?

Right. Also, data packets are presumably large, while probe packets can 
be small. Sending four 80 byte probes to test four combinations of two 
sources and two destinations is less expensive than sending one 500 or 
1500 byte data packet. On low bandwidth links having small probes also 
means the RTT for the reply is much faster: on a 56k link 80 vs 1500 
bytes makes a 400 ms difference.

These savings are especially significant when the active path is still 
available after all.