From: Karthik Balaguru on
Hi,
I wonder the need for ACKs at various layers.
If the Network Layer has the ACK feature,
why is the ACK mechanism required at the
Data Link Layer too ? Any ideas ?

Thx in advans,
Karthik Balaguru
From: Aaron Leonard on
~ Hi,
~ I wonder the need for ACKs at various layers.
~ If the Network Layer has the ACK feature,
~ why is the ACK mechanism required at the
~ Data Link Layer too ? Any ideas ?
~
~ Thx in advans,
~ Karthik Balaguru

The lower layer has better knowledge of the transmission
expecations of its link than do the upper layers. Therefore
it can tune its retransmission scheme much more precisely
to yield better performance.

For example, TFTP has a 5-second (let's say) ACK timeout,
because it can make no assumptions about the latency of the
underlying network path (which could be any arbitrary
topology IP path.) However, 802.11 has a (let's say)
1 millisecond timeout, because it knows that its transmissions
travel at the speed of radio waves, and never more than
(let's say) one mile. Therefore, if an 802.11 transmission
should be lost or corrupted in transit (a not uncommon
event), the transmitter can resend after a millisecond,
and so get the data through pretty quickly, despite the error.

If 802.11 didn't retransmit, but you had to rely upon the
upper layer to do the retransmission, you'd have to wait
5 seconds. The resulting throughput would be much worse.

Quite some years ago, I recall reading _An Engineering Approach
to Computer Networking_ by S. Keshav. Right now it's buried
pretty deep in a box at my feet, but it might well still be
worth a read. There's also Radia Perlman's _Interconnections_.

There's still a lot to be said for sitting down and reading
a book or two.

Aaron
From: Karthik Balaguru on
On Feb 18, 3:21 am, Aaron Leonard <Aa...(a)Cisco.COM> wrote:
> ~ Hi,
> ~ I wonder the need for ACKs at various layers.
> ~ If the Network Layer has the ACK feature,
> ~ why is the ACK mechanism required at the
> ~ Data Link Layer too ? Any ideas ?
> ~
> ~ Thx in advans,
> ~ Karthik Balaguru
>
> The lower layer has better knowledge of the transmission
> expecations of its link than do the upper layers.  Therefore
> it can tune its retransmission scheme much more precisely
> to yield better performance.
>
> For example, TFTP has a 5-second (let's say) ACK timeout,
> because it can make no assumptions about the latency of the
> underlying network path (which could be any arbitrary
> topology IP path.)  However, 802.11 has a (let's say)
> 1 millisecond timeout, because it knows that its transmissions
> travel at the speed of radio waves, and never more than
> (let's say) one mile.  Therefore, if an 802.11 transmission
> should be lost or corrupted in transit (a not uncommon
> event), the transmitter can resend after a millisecond,
> and so get the data through pretty quickly, despite the error.
>
> If 802.11 didn't retransmit, but you had to rely upon the
> upper layer to do the retransmission, you'd have to wait
> 5 seconds.  The resulting throughput would be much worse.
>

Okay !

Thx,
Karthik Balaguru