[Fwd: VOTimeSeries]

[Doug Tody]

On Tue, 3 Jul 2007, Rob Seaman wrote:

> Yeah, I've gently asserted this idea at various points in the past (in fact,
> somebody else mentioned something similar at a telecon yesterday).  It has not
> been welcomed so far.  But relying purely on references stored on remote
> servers opens unattractive possibilities like later triggering a massive
> assault on those servers - especially in the era of 10^5 LSST events nightly
> batched and (it now appears) eagerly replicated.

This is one reason it is necessary to be able bundle certain classes of 
information in the event.  Doing so makes it relatively straightforward
to scale up, allowing many copies of the same event to be processed in
parallel on a cluster.  External references can scale too, but are not
high performance, and should only be used for non-critical stuff.

> > There are important questions which have not been discussed here, such as
> > where this time series data comes from, how it is used,
> 
> For VOEvent the answers are as broad or broader than any other VO science
> cases.

If so, that pretty much clinches it; you will need a general time
series implementation.  Data can be either included directly in the
event, or pointed to as external data; the same data model should
be used for both cases (most of it can probably be omitted for the
simpler cases of directly included data).  The external version will
require more complete dataset metadata, similar to what we already
have for SSA.  While formally specifying the data model and supporting
multiple serializations may not be a requirement (for VOEvent at
least), it is highly desirable.  Commonality with related standards
such as SSA is highly desirable if not a requirement.

> > how we avoid problems of source confusion if the data is merely retrieved
> > from some archive, and so forth.
> 
> This seems a little less pertinent.  VOEvent is more tied to the
> pipeline side of the business than to "static" archives.  A bigger
> question is shelf life.

Sure, but where does this time series data come from?  Is this observed
as part of the event detection, or is it retrieved from some archive
under the assumption that we know what source the event refers to?

	- Doug