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

Re: 802.3x flow control


Let us take the case of 10 G MAC and 9.58460 PHY.

a. Receive side
The received clock will be the retrieved 9.58460/n (n being the datawidth).
The data will 
be stored in the elasticity
buffer of MAC or Host receive FIFO depending upon where the receive clock
is terminated.
In any case, since input is slower, there is no problem of overflow (even
if elasticity buffer
is used). Thus there is no need of flow control (for this reason).

b. Transmit side
The PHY will be sending data at 9.58460 on the fiber. It is getting data at
10G from MAC.
It has to accumulate the difference. How long can it do so and using how
much local
buffer. As I wrote earlier 64 byte deep buffer can absorb this speed
difference for upto
1540 byte packet. After that, it has to request for long enough IPG so that
either its
buffer becomes empty or goes to certain threshold value. For this we
require local 
flow control which can be implemented using CRS / HOLD or similar mechanism

Thus in none of these cases 802.3x comes to help.

On the other hand, if we had 9.58460 MAC and 10G PHY:

a. On transmit side, there is no problem.
b. On receive side:

	b1. If received clock is terminated in receive Host FIFO. 802.3x can be used.
	b2. If elasticity buffer is used (which in my view is cleaner from design
               of view, especially keeping autonegotation in mind) we have
similar issue
	      as the storage in PHY. Again we require some local flow control BACK to
                PHY. Ovbiously this has to be out of standard, as known, so

If you were keeping the latter option in mind, I take my statement back
I would NOT like b1 from design point of view). 


At 05:44 PM 7/23/99 -0700, you wrote:
>With respect to 802.3x flow control, why do you say that you don't see how
it can
>help in the context of 10 vs 9.58460 Gbps?
>My assumption is that 802.3x flow control is speed independent. For
example, it
>can flow control a mixed 10, 100 and 1000 Mbps Ethernet devices. Is this
not true?
>Devendra Tripathi wrote:
>> Rich,
>> Given that we are discussing many flow controls here, it may be usefull to
>> mention which one we are talking and
>> when. In my understanding following flow controls are being mentioned here.
>> a. 802.3x : Useful  on recieve (wrt medium side) side FIFOs. In the context
>> of 10 vs 9.58460 I do
>>     not appreciate, how it can help.
>> b. Transmit side PHY to MAC flow control either using COL (as I suggested)
>> or HOLD (as Dan has suggested).
>>     This is applicable when PHY is at 9.58460 and MAC at 10.
>> c. Receive side MAC to PHY flow control. Applicable if one decides to have
>> 9.5846 as MAC speed and 10 as PHY. I am
>>     not sure if, ever, this has been context, though.
>> Is there anything else we are talking ?
>> Thanks,
>> Tripathi.
>Best Regards,
>Richard Taborek Sr.    Tel: 650 210 8800 x101 or 408 370 9233
>Principal Architect         Fax: 650 940 1898 or 408 374 3645
>Transcendata, Inc.           Email: rtaborek@xxxxxxxxxxxxxxxx
>1029 Corporation Way    
>Palo Alto, CA 94303-4305    Alt email: rtaborek@xxxxxxxxxxxxx