[RPRWG] Concern with Single Transit Buffer
I am very concerned of having a Single Transit Buffer (STB) MAC and
Dual Transit Buffer (DTB)
MAC coexist on the same ring
The fundamental problem is in the two distinct ways congestion is
detected on a RPR ring.
One being rate based and other based on HOL timer expiration. In a
Dual Transit Buffer (DTB) MAC
add LP and eMP gets opportunity to get on the ring because it can
hold LP and MP transit
traffic in the LPTB until this buffer fills up threatening imminent
packet loss. Since STB MACs
cannot hold any transit traffic for more than one MTU time unit the
same HOL timer setting, used at the DTB MACs,
will expire more frequently at the STB MAC nodes triggering
congestion on the ring while upstream DTB nodes are
in full compliance with their add and forward rates.
This scenario is similar to connecting two separate north bound, 2
lane highways with a one-way, east-bound lane
that has a T junction with a traffic light controlling local add
traffic trying to get on to this single lane road. Assume that
this traffic light ensures that add traffic does not wait for more
than certain amount of time (it has sensor that starts
a timer the moment a car arrives at the light) If the timer on this
traffic light has a fixed interval than pretty soon the
highway that is feeding traffic to this single lane road will have a
traffic build-up leading to a traffic jam. (Upstream DTB
nodes will suffer, including reserved services).
On the other hand if the timer on this traffic light is tied into
rate monitors on the highway
feeding traffic to this single lane road than as traffic rate on the
highway increase the timer on the traffic light
will change making the wait time vary for the add traffic. So, if
the add traffic has a high priority carpool lane
than the wait on this lane will also be variable. (HP or reserved
rate add traffic at the STB node will suffer variable delay
- hence will have no jitter bounds)
Just like the folks on the highway and the folks in this small town
trying to get to work on time will both
be frustrated - mixing STB and DTB MACs will be no different for
reserved and HP traffic on the ring.
The fairness algorithm was designed for a DTB MAC, shoe-horning a
STB MAC makes the standard complex and
very inconsistent. I request that we focus RPR on having one scheme
of transit path mechanism on the ring.
This will help the working group stay on time within the schedule we
have in front of us.
raj