Re: [nm-wg] standard timestamp format

[Adrian Cockcroft <Adrian.Cockcroft@Sun.COM>]

There are two distinct uses for timestamps, which have  
different/conflicting solutions.

a) universal time to allow correlation of events, especially  
correlation between endpoints of a link
If a network goes down, you may get an event from each end of the link,  
and it needs to be synchronized
well enough to resolve this as one failure rather than a series of  
failures.

b) linearly and monotonically increasing time to allow accurate  
summarization of data from a single
source, i.e. differencing of timestamps to obtain rates and response  
times.

An operating system such as Solaris uses two separate timing sources, a  
global universal time that
can be synchronized by NTP and set by the user. It can speed up, slow  
down and make forward and backward
large jumps. A second "hi-resolution" timer runs off the CPU/System  
clock crystal and is monotonic and
linear with better than microsecond resolution. This is used by all the  
performance metrics, for calculating
disk service time and CPU microstate accounting. See gethrtime(3C).

The discussion here is centered on the global/universal time format,  
but if it is to be used to do differencing
within a system, then it may be worth embedding a local hires timestamp  
in the format as well, which is only
valid when differencing times from the same source.

Adrian

On Jun 14, 2004, at 1:55 PM, Dan Gunter wrote:

> I'm not sure that a _single_ standard timestamp format is required;  
> what I would like to see is a minimal set of timestamp formats that  
> will satisfy everyone, with rules for converting between them  
> explicitly documented somewhere. I'm not sure *GGF* is the right place  
> for this, ultimately, but I don't see any harm in proposing something  
> in this context even if it just ends up becoming input into another  
> standards body down the line.
>
> I don't want to sidetrack NM-WG, but given that we need to exchange  
> timestamps and we've already had discussions on this, it looks like  
> we're already sidetracked, and waiting for another WG to form and  
> finish just to document this would be a real bump in the road. Not  
> documenting what decisions we make seems like a waste of a potentially  
> more broadly useful set of discussions and decisions.
>
> Therefore, I would propose that a few people interested in this get  
> together enough spare time to write a short-and-sweet document about  
> the timestamp formats we can agree on. AFAIK, these are currently:
>
> - NTP seconds (16 bytes, encoded either "raw", or for XML as hex or  
> decimal numbers)
> - ISO8601 (restricted to YYYY-MM-DDThh:mm:ss.ff..fZ variation)
>
> [ Note: I would propose that accuracy and resolution representation be  
> treated as optional extensions to be specified at some later date. As  
> much as I'd like to include these in any documentation we produce,  
> they tend to be controversial, and specifying these in a standard way  
> is not crucial to moving forward. ]
>
> Once this document is done, it could be reviewed by the group as a  
> whole and, if found satisfactory, thrown into the abyss of GGF  
> document approval. Whether or not final approval is given, our Web  
> Services will have some concrete, documented, agreement to go on for  
> the timestamp at least.
>
> The documents cited below could provide background material, but I  
> think what we need is much simpler, smaller, and easier to agree on.
>
> -Dan
>
> Brian Tierney wrote:
> > Hi all:
> > Richard mentioned that the issue of a standard timestamp has come up  
> > in  Mark/Paul's implementation work.
> > There were a couple of documents on this a few years ago:
> > http://www-didc.lbl.gov/GGF-PERF/GMA-WG/papers/GWD-GP-10-1.pdf
> > and
> > http://www.cs.tcd.ie/coghlan/meta/datagrid/PDPTAv7_2.pdf
> > I asked Steve Tueke what the WS-RF solution to timestamps is, and he   
> > said that they have not yet found a critical need for a standard   
> > timestamp.
> > Im not sure of the NM-WG is the right place for this, but I'd like to  
> >  see the above document
> > standardized. Other thoughts on this?
> > ---------------------------------------------------------------------- 
> > -- -------------------
> >   Brian L. Tierney,   Lawrence Berkeley National Laboratory (LBNL)
> >   1 Cyclotron Rd.  MS: 50B-2239,  Berkeley, CA  94720
> >   tel: 510-486-7381    fax: 510-495-2998   efax:  240-332-4065
> >   bltierney@lbl.gov   http://www-didc.lbl.gov/~tierney
> > ---------------------------------------------------------------------- 
> > -- ------------------