On 28 Jul 2011, at 15:51, Tony D'Amato wrote:
> 
> After reading the draft, and based upon the issues that Russ and Steven 
> brought up concerning NFSv4 interop and the epoch representation, I do not 
> support the draft in it's current form.

So, these would appear to be the two issues here, and I think it's worth 
considering them separately

1/ Time grandularity (aka NFSv4 interop)

I think it's important to remember here that AFS on the wire is very different 
from NFSv4 on the wire, and that the choices they made aren't necessarily a 
good guide for the choices we should make. In particular, we transmit a 
FetchStatus structure with the results from practically every RPC. Unnecessary 
bloat in the FetchStatus structure can have significant effects on the amount 
of data transmission required to do AFS file operations, and on the speed of 
those operations. There are currently two time fields in this response. We're 
already adding 64bits to this structure to support extended time, adding 
nano-second granularity would require a further 64bits on top of this.

I'd really like to see examples of shipping operating systems which expose 
nano-second time to the user through the VFS layer. Without examples of this, I 
can't see any point in adding the additional bloat.

2/ Choice of Epoch

Russ makes some very good points about the dangers of using the MS Windows 
epoch. Given that the epoch is an arbitrary choice, I do wonder if it might be 
easier and safer to keep the Unix epoch for the new time type.

Cheers,

Simon.

_______________________________________________
AFS3-standardization mailing list
[email protected]
http://lists.openafs.org/mailman/listinfo/afs3-standardization

Reply via email to