Harry,
If cut-through is for H traffic only, I don't see any
conflict between cut-through
and preemption. Preemption is helping cut-through in a very
effective way.
The original goal for preemption is to reduce the packet
transfer delay and
jitter on the ring as much as
possible. Isn't this exactly what cut-through is
fighting for ?
Whether GFP belongs to L1 or L2, I still have doubts. But for
pure technical
discussion, I don't think that using GFP to eliminate the
chance of preemption
is appropriate here.
By the way, I totally understand and agree with your
definition of cut-through.
By your last statement, do you mean SONET clock distribution
is needed in
RPR by someone ?
Regards
William Dai
----- Original Message -----
Sent: Thursday, April 12, 2001 6:13
PM
Subject: RE: [RPRWG] Cut through
definition?
If
this scheme is adopted then I don't see how this would work with cut
through mode and GFP.
Hence it is no longer L1 agnostics. Should we not
support cut-through then?
Is
this a bad thing for RPR? I think I will let the someone else answer the
question!
I
would like to make it clear that we, Nortel, are like everybody else, that we
will support
GFP
on our equipment when the 3rd party silicon vendors sell GFP compliant
chips.
What
is the original goal for preemption? Are you providing a solution that is more
complicated than it's worth. What about test
equipment complexity.
Lastly, there are those who believe that RPR is more
than just for packet based network.
Harry
Although I'm not a big preemptive transfer fan, but I
think this topic deserves detailed
discussion before we rush into any conclusion. What
changes me is the discussion of
Jumbe Frame support on RPR, not long ago it was 2KB, now
it is 9KB, what about the
ultimate 64KB in the future ?
By saying that, I'm proposing neither ATM cell like
structure nor slotted ring structure,
and since RPR MAC is L1 agnostic, physical signalling
trick cannot be used either.
Let me give one example of preemptive transfer definition
here and let's discuss what
is so complicated (simple) about it.
1. There are 3 MAC
classes of traffic (H, M, L,).
2. Preemption is
allowed only for "Transit" H traffic to preempt "Transmit" M or L traffic.
3. Preempted segment
is not allowed to be prempted again.
4. Preempted
"Transmit" traffic will be scheduled to tranfer right after "Transit" H
traffic,
independent of classes.
5. Each Packet
transfer will be inserted an "IDLE/Escape" word for every 256 or 512
(for the
sake of alignment/padding concern) byte as the preemptive inserion point.
6. Jumbo frame is not
supported for H class.
By the way, SONET clock distribution is not needed. After
all, RPR is a packet based network.
Best Regards
William Dai
----- Original Message -----
Sent: Thursday, April 12, 2001 7:23
AM
Subject: RE: [RPRWG] Cut through
definition?
Exactly my point.
"we should keep it simple and not Segment
packets. " i.e. Do not preempt.
Regards,
Harry
I am not clear
how the proposed preemption method works.
Does a high priority transit packet preempt a low priority add
packet? Can a high priority add packet also preempt a low priority
transit packet? What happens if a previously preempted add packet
contends with a same priority packet that was also preempted in an
upstream node? What happens if a previously preempted add packet
contents with a same priority previously preempted transit packet that
follows a high priority preempting transit packet with a clock cycle gap
in between due to clock mismatch? Do we require a SONET clock to be
distributed on the ring? Is RPR MAC layer one agnostic?
Thanks.
Necdet
Harry Peng wrote:
Complexity what complexity:
In the tandem path, if a high priority packet can
preempt a low priority packet at arbitrary
boundary then the preempted logic will have to deal with a tandem
packet that is already pre-empted.
This means the fastest pre-emption response time is
on internal word size and the pre-empted packet will have to pad to word boundaries to make live
easier. Furthermore the tandem receiver will
have to respond to within one clock cycle as it is the
atomic size. What is the word size for 10G 64 bits
128 bits? What about for 40G or higher.
Unless, you are will to have cells. Then why not use
ATM.
I agree that we should keep it simple and not Segment
packets.
Regards,
Harry
-----Original Message----- From: Sushil Pandhi [mailto:Sushil.Pandhi@xxxxxxxxxxxxxxx]
Sent: Wednesday, April 11, 2001 10:33 AM
To: Leon Bruckman Cc:
'davidvja@xxxxxxxxxxx'; stds-802-17@xxxxxxxx Subject: Re: [RPRWG] Cut through definition?
I agree with Leon that 'pre-emption' will
increase the complexity. ATM solves this by segmenting the message into smaller size cells and reassembling cells, and using
this approach adds a lot of
complexity.
If we do not have preemption, and assuming 1522 byte
frame just starts transmission before
synchronouus traffic can be sent, 1522 byte
frame at OC-3 rate will take about 82.6 micro-seconds.
If we assume that in the ring, at 32
nodes the same situation arises then we have
about 2.6 msec delay because not doing
preemption. So I doubt preemption will give us much advantage.
-Sushil
Leon Bruckman wrote:
> DVJ > My personal
view is that preempting lower traffic in the middle of a packet
> adds complexity that is not really needed. At
1G, the transmission time for > a 1500
bytes packet is 12 usec, so the worst case for a 256 ring will be
3.1 > msec of added delay because of low
packets being transmitted and not >
preeempted. Furthermore, the probability of the worst case is very
small. We > did some simulations with the
following assumptions: > - There is always
a low priority packet being transmitted by the node > - High priority packet may arrive at any time during the
low priority packet > transmission (equal
probability) > Some of the results were
presented during the January interim (by Gal Mor). > > For a 128 nodes ring
operating at 1G the preeemption gain will still be in > the msec range with very high probability, and this can
easily be absorbed > by the jitter buffers
at the receiver. > Leon > > -----Original
Message----- > From: David V. James [mailto:davidvja@xxxxxxxxxxx]
> Sent: Friday, April 06, 2001 4:27 AM
> To: Carey Kloss; Devendra Tripathi
> Cc: stds-802-17@xxxxxxxx > Subject: RE: [RPRWG] Cut through definition?
> > All,
> > Relative to the
discussion of cut through, et. al. > My
perception is that a cutthrough node has two > insertion buffers, for classA (provisioned
synchronous) > and classB (provisioned
asynchronous). > > The preferred transmit order is as follows:
> a) classA insertion buffer
(always) > b) classA transmit
traffic (subject to provisioned rate) > c) asynchronous traffic. > The classA insertion buffer only needs to be the size
of > the maximum packets sent by this
node, plus (perhaps) some > extra symbols
to deal with hardware decoding latencies. > > The classB insertion buffer
is to deal with the accumulation >
asynchronous packets that occurs when (worst case) full
asynchronous > is coming in/out and
rate-limited synchronous is being transmitted. > The size of the classB buffer is on the order of several
upsteam-link > delays times
rateOfSynchronous/rateOfLink ratio. > > Order of the asynchronous
traffic (c) depends on the classB >
buffer-filled status, prenegotiated vs. consumed rates, and
> the size of the asynchronous backlog in the
client. > >
The asynchronous transmit buffer is a bit schitzophrenic on its
> behavior. It should be in the client (not the
MAC) because that > allows packets to be
reordered/inserted/deleted until the just > before transmission time. However, the amount of traffic
in the > asynchronous transmit queue may
influence the MAC queue-selection > and
throttle-signal assertion properties. > > I personally favor allowing
cut-through synchronous traffic to >
preempt asynchronous, even in the middle of a packet. That's
> yields the lowest possible jitter, but at some
encoding complexity > costs.
> >
DVJ
|