[c-nsp] GSR12008|GRP-B|4OC12/ATM-MM-SC|3GE-GBIC-SC throughput?

[Jason Lixfeld]

On 15-Apr-09, at 2:58 PM, Lamar Owen wrote:

> Incidentally, the 'show fabric' undocumented command shows internal  
> latencies
> across the fabric.

Highest latency on the fabric is 84ms, over two months ago.

> The ATM SAR tax may be hitting you, too.

Not being an ATM guru, I hope someone will clue-bat me if I get too  
far gone with my calculations below...

An ATM cell payload is 48 bytes long.  On top of each cell, there's a  
5 byte ATM header.

If my average packet size is 690 bytes, one packet would be stuffed  
into 15 cells.  Each of those 15 cells would have an additional 5  
bytes of overhead for the header.

So, 5 bytes header for 15 cells = 75 bytes per 1 690 byte packet = 765  
bytes/6120 bits.

At the time the sample was taken, I was pulling in 27131pps over my  
two GSR ATM interfaces and pushing 32478pps over the same two  
interfaces.  If my cell tax calculations are right, that would equal  
166041720bps in and 198765360bps out across the GSR, but equally  
importantly, that would equal the same amount being put over the OC12  
to Toronto.

Correct me if I'm wrong, but based on this estimation, ATM cell tax  
wouldn't be an issue, would it?