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

RE: call 3/4/2015 notes



A bit more on the timestamps.. so 1588 timestamp is:

struct Timestamp
{	
	UInteger48 secondsField;
	UInteger32 nanosecondsField;
};

The catch is that the nanosecondsField field is never greater than 10^9, which again is smaller than 2^30. That means the 2 most significant bits are always 0 or "unused". IMHO we should look into the possibility of just using the nanosecondsField in the RoE header..

The other possible time field could be the time interval:

struct TimeInterval
{
	Integer64 scaledNanoseconds;
}; 

The scaledNanoseconds is always left shifted by 16 bits giving 16 bits of sub-ns accuracy. I am not really sure whether we need such precision.

Just some thoughts after the call.

- Jouni


> -----Original Message-----
> From: stds-1904-3-tf@xxxxxxxx [mailto:stds-1904-3-tf@xxxxxxxx] On Behalf Of
> Jouni Korhonen
> Sent: Thursday, March 05, 2015 10:59 AM
> To: STDS-1904-3-TF@xxxxxxxxxxxxxxxxx
> Subject: call 3/4/2015 notes
> 
> Folks,
> 
> Start:	3/4/2015 8PM PST
> End:	3/4/2015 9PM PST
> 
> Present:
> 	Jouni Korhonen
> 	Bomin Li
> 	Kevin Bross
> 	Peter Ashwood-Smith
> 	Raz Gabe
> 	Jinri Huang
> 	??
> 
> Discussion:
> * EtherType
> * Packet sizes
> * CRB/VRB flows and their support
> * Sequence numbers & timestamps
> 
> Decisions (by consensus):
> * EthType - only one, if further "types" were needed provide equivalent
> functionality
>    using subtypes in RoE header.
> * Packet sizes - no recommendations unless someone has good data to backup
>    specific numbers. Generic recommendation not to provide numbers
> * CRB/VRB - need to be prepared for VBR flows as a preparation e.g. for
> foreseen
>    compressed flows.
> * Sequence Numbers and timestamp:
>   - both needed; sequence numbers essential to find out lost packets; field sizes
>     to be determined
>   - good to have 1588 alignment for timestamp format -> explore how and what
>   - timestamp not needed in every packet. More likely e.g. at the
>     beginning of each radio frame
>   - between timestamps rely on sequence numbers
>   - explore whether sequence number and timestamp could use
>     the same field; 80 bits is too much though
>   - to shorted the size of timestamp look for way to do so
>     One proposal to predefine the time reference -> e.g. TAI
>   - preferable to allow sub-ns accuracy in the timestamp field
>     (like 1588 allows)
> 
> Next call: 3/17/2015 at 6AM PST
> 
> - Jouni
> 
> --
> Jouni Korhonen, Ph.D, Associate Technical Director CTO Office, Networking,
> Broadcom Corporation
> O: +1-408-922-8135,  M: +1-408-391-7160