Thread Links Date Links
Thread Prev Thread Next Thread Index Date Prev Date Next Date Index

Re: [EFM] Half-duplex deferral for MAC-PHY rate matching and compatibility with existing silicon


While I understand your concern over the "limitation" of 100Mbps half-duplex, I don't agree with your conclusion or even your definition of the problem.

Firstly regarding pause frames:

1. I believe that fewer existing MAC implementations will support pause frames than will support the "receive while transmit" that Shimon pointed out in the standard.

2. If you do use pause frames you will break the architectural division that expects PHYs to transmit frames - not source them.

3. You will have problems supporting the pause frame function between MACs (as it was intended) above your PHY layer implementation. Thus you will have taken 2 MACs that support .3x and taken away that function to make the PHYs work.

Secondly regarding the problem:

1. The difference between 100-half and hundred full is not as great as you seem to make out. For instance if you are aggregating standard VDSL links your downstream will exceed 100Mbps almost immediately after your aggregate exceeds 100Mbps - thus the extra bandwidth gained by full duplex is <20Mbps.

2. If you take a standard VDSL symmetric rate of 26/26, the payload rate is <25Mbps (somewhere around 23Mbps). In that case you can aggregate 4 lines using 100-half; 8 lines using 100-full; >8 lines using Gig. I think that 4 lines will be a sweet spot, above that it may be 24 lines as the next sweetspot.

3. If you have a "super PHY" capable of aggregating many high-speed links, you should consider using GMII.


Steven.Haas@xxxxxxxxxxxx wrote:

> Arthur,
> Our experience has shown that the half duplex method is the safest way and works with most MACs. The CRS can always be used to control the flow of frames from the MAC to the PHY, even at all the different rates proposed for EFM copper objectives.
> The only drawback is the "limitation" of 100Mbps. This can be overcome with most MACs using a flow control message (802.3x). This method is not watertight and not all MACs process this in a timely fashion so buffers are needed in the PHY. The case of an EFM PHY working above 100Mbps will most likely only happen in very short range loop aggregation scenarios when there are multiple lines going to a business customer. I expect fiber to be used in these cases so creating a new MAC seems unnecessary.
> I propose the following engineering tradeoff:
> 1. Use of the half duplex method for most modes of operation. This covers all rates up to 100Mbps aggregate.
> 2. Usage of flow control, combined with the appropriate buffers, in the rare cases of rates above 100Mbps.
> A third option is to create a jump in supported service rates from 100Mbps half duplex to 100Mbps full duplex.