Re: [8023-CMSG] Technical proposals

[Arthur Marris <arthurm@xxxxxxxxxxx>]

Ben,

   My assumption is that this would be a Clause 4 change similar to ifsStretch. This is where it belongs from the implementation and architecture point of view. There would be no impact to PHYs or the RS if it was done here. It could be made optional so as not to impact existing MAC implementations. Having a clean and generic way of controlling IPG for rate control seems a very useful feature. The need for this became apparent in both 802.3ae and 802.3ah.

 

Arthur.

 

---

From: owner-stds-802-3-cm@listserv.ieee.org [mailto:owner-stds-802-3-cm@listserv.ieee.org] On Behalf Of Benjamin Brown
Sent: Friday, December 10, 2004 1:21 PM
To: STDS-802-3-CM@listserv.ieee.org
Subject: Re: [8023-CMSG] Technical proposals

 

All,

(Arthur, I don't mean to pick on you but your message gives
me a good excuse for bringing this up!)

I'm very happy to see this amount of discussion on these topics.

In addition to considering the kinds of rate limiting you'd like to
see, please also consider how these changes might be worded
in the 802.3 standard. Would they impact Clause 4 / Annex 4A
in a fashion similar to the way ifsStretch impacted Clause 4 or in
another new way? Would they impact the RS/PHY below the
MAC? Would there be a new Annex 31C? Would we create
another MAC sublayer to be inserted above or below the MAC
Control sublayer? How would this new sublayer play with the
OAM and Link Aggregation sublayers?

Adding functionality to a single implementation is very different
from modifying the standard in a way that doesn't imply any
particular implementation.

Good luck!

Ben

Arthur Marris wrote:

Hugh,

   I would be happy to review and comment on your presentation.

   I have implemented a solution for limited bit rate. I believe the

best way to do this is to define two 8 bit parameters to control the

IPG. The first is multiplied with the length of the previously

transmitted frame and the second plus 1 is divided with its length to

produce the IPG. The plus 1 is to prevent a divide by zero.

   This meets both Pat's requirement to fractionally increase the frame

size based on data length and the requirement to slow down traffic more

substantially. Using 8 bits for each parameter allows the rate to be

slowed down by up to a factor of 255 with a granularity of 1/256. Also

this is easy to implement.

Arthur Marris

Cadence Design Foundry UK Limited

The Alba Campus

Livingston, UK, EH54 7HH

+44 (0)1506 595104 (desk)

+44 (0)7901 855100 (cell)

-----Original Message-----

From: owner-stds-802-3-cm@listserv.ieee.org

[mailto:owner-stds-802-3-cm@listserv.ieee.org] On Behalf Of Pat Thaler

Sent: Thursday, December 09, 2004 10:47 PM

To: STDS-802-3-CM@listserv.ieee.org

Subject: Re: [8023-CMSG] Technical proposals

Hugh,

There is also the case of lumpy per data overhead - cases where there is

x overhead added to a packet per every y bytes. An example would be

adding FEC protection at the physical layer to packets as EPON does. It

is different from limited bit rate because there is extra overhead when

the packet isn't a multiple of y. If we are willing to accept a solution

that has a small lost throughput cost this could be handled by allowing

the constant per packet overhead and limited bit rate modes to be used

in together.

Regards,

Pat

-----Original Message-----

From: owner-stds-802-3-cm@listserv.ieee.org

[mailto:owner-stds-802-3-cm@listserv.ieee.org]On Behalf Of Hugh Barrass

Sent: Thursday, 09 December, 2004 11:28 AM

To: STDS-802-3-CM@listserv.ieee.org

Subject: Re: [8023-CMSG] Technical proposals

Ben & Peter,

This was a very important point that I was planning to raise in January

(if I get the chance :-)

I consider that there are 3 distinct types of static rate limiting that

we should consider. We may decide to support only 1 or 2 of them but I

think it is important to consider all 3 carefully before we make a

decision:

1. Constant (per packet) overhead.

This is the same as the applications that Kevin (daines_cmsg_1_0409.pdf)

outlined where a tag or encapsulation is added to each packet. This

requires that the MAC control layer should force a larger minimum IPG

between all packets transmitted.

2. Limited (payload) bit rate.

This is the situation with where a media convertor is dealing with a

medium that doesn't support the "power of 10" speeds that are

traditional for (most) Ethernet interfaces. Examples of this might be

Ethernet over DSL (including EFM) or Ethernet over SONET. Another class

of application for this rate limiting would be Network Interface Cards

that have limited bus bandwidth into the host system (this was the

example that I drew in barrass_1_0704.pdf slide 26). In this case the

MAC control layer would adjust the IPG in a manner dependent on the

packet length.

3. Limited packet rate.

This application has not been proposed or considered as yet but may be

worth including. This rate limiting mechanism would be used to protect a

device that is unable to process packet headers at the full line rate

possible. This may be because it is unable to perform address lookup or

packet classification or may be due to another reason altogether. A MAC

control layer could easily implement this type of rate control.

I suggest that we could define all three types of rate control although

a "smart" definition should allow "non optimal" implementations to

sacrifice corner case performance to reduce complexity.

If I get the go ahead, I will prepare a presentation and I would welcome

assistance with reviewing and modifying it prior to the meeting.

Hugh.

Benjamin Brown wrote:

Peter,

I hope you don't mind that I opened your questions up

to the entire reflector. It helps me make an important

point.

Yes, I think that it is absolutely appropriate to consider

a solution for a pseudo-static rate limiter that accounts

for a difference in bit rate (e.g. 100M transmitting to 2M)

as well as a constant overhead (e.g. untagged transmitting

to tagged). Anything less than this would probably not

solve the problems of both of these kinds of networks.

Further, I advise participants that the Sacramento meeting

is indeed the appropriate time to bring technical proposals

for just such a solution. There is still plenty of time to put

some ideas on some slides. There is even time to start

sharing those ideas with other members of this task force

(yes, we're now a task force and no longer a study group)

to begin forming consensus. Consensus is what drives this

process and the responsibility for creating a consensus lies

entirely on the shoulders of every member of this task force.

The sooner we get proposals and start building consensus,

the sooner we can accept a baseline proposal, start writing

drafts and getting to the ballot process.

For now, I'll continue to be the repository for presentation

requests. I'll make sure that whoever stands in front of the

room in Sacramento is aware of your plans.

Regards,

Ben

Peter Jones wrote:

Benjamin,

I have a question based on what you said when you came to visit

802.17 (which is where I spend my time), and in conversation

afterwards.

It seems to me that are two main uses for the simple "rate limiting"

approach:

1)     a physical bit subrated link - e.g. only 45Mb/s on a 100Mb/s

link from a ethernet to DS3 media converter

2)     an encapsulating device - something that adds a fixed overhead

per packet like a provider bridge or a security device.

Do you think that these are both going to be addressed during the CM

work?

regards

peter

--

-----------------------------------------

Benjamin Brown

178 Bear Hill Road

Chichester, NH 03258

603-491-0296 - Cell

603-798-4115 - Office

benjamin-dot-brown-at-ieee-dot-org

(Will this cut down on my spam???)

-----------------------------------------