Forgive me if this is a naive question but it has been troubling me for
a while now... I do know a bit about this stuff but I can't get my head
round this...
How does oversubscription actually work in ATM based ADSL networks?
My understanding is that IP packets are launched into the ATM network on
AAL5 where they are split into cells and shovelled down virtual paths
towards DSLAM's where they are distributed to end users.
Now if the virtual path is oversubscribed, for example to offer 10:1 ADSL
service to subscribers, how is it possible to get any complete IP packets
through?
My first thought was that CLP=1 is used at ingress for the cells
corresponding to 90% of IP packets (1 in 10 packets are CLP=0), however
this would mean that in busy times the available bandwidth would quickly
drop to the 10% "CIR" even if there is spare capacity.
I just don't see how (under congestion conditions) you'd get many full IP
packets through since on average some of the cells in each AAL5 stream
will have been discarded???
For example 2 subscribers on the same DSLAM attempting to download large
files would each get 10% of the path capacity and the remaining 80% of
path capacity would be full of half complete AAL5 IP packets that can
never be reassembled.
The same problem happens on the upstream from the customers to the core,
though there seems to be even less control on the traffic since each end
user is sending ATM cells into the network with no policing or marking at
ingress at all.
Is there some mechanism I don't know about which protects cells within
AAL5 to keep L3 packets complete?
Aled
a while now... I do know a bit about this stuff but I can't get my head
round this...
How does oversubscription actually work in ATM based ADSL networks?
My understanding is that IP packets are launched into the ATM network on
AAL5 where they are split into cells and shovelled down virtual paths
towards DSLAM's where they are distributed to end users.
Now if the virtual path is oversubscribed, for example to offer 10:1 ADSL
service to subscribers, how is it possible to get any complete IP packets
through?
My first thought was that CLP=1 is used at ingress for the cells
corresponding to 90% of IP packets (1 in 10 packets are CLP=0), however
this would mean that in busy times the available bandwidth would quickly
drop to the 10% "CIR" even if there is spare capacity.
I just don't see how (under congestion conditions) you'd get many full IP
packets through since on average some of the cells in each AAL5 stream
will have been discarded???
For example 2 subscribers on the same DSLAM attempting to download large
files would each get 10% of the path capacity and the remaining 80% of
path capacity would be full of half complete AAL5 IP packets that can
never be reassembled.
The same problem happens on the upstream from the customers to the core,
though there seems to be even less control on the traffic since each end
user is sending ATM cells into the network with no policing or marking at
ingress at all.
Is there some mechanism I don't know about which protects cells within
AAL5 to keep L3 packets complete?
Aled