Thanks for the response. What I hear you saying is that the underlying
infrastructure that John is creating (i.e. the GribFeatureCollection)
and the fixes to what's broken in the identification of the data (e.g.
the break out of the variables on different accumlation times) will help
you provide consistent results. I agree that these changes are necessary.
However, I think the same thing can be achieved with the human readable
variable names. There is no guarantee that the VAR_* names won't change
in the future. As John discussed with me last week, if he finds a new
PDS variable that he thinks is important, it could be added to the
variable name and then we go through the pain again. That's no
different than changing the human readable names. The lookup for
creating consistent human readable names is already there to create the
Even with the human readable names, there will be pain for tool
developers that access the data, because some names will change. It
will require changes to the IDV, but at least they will be manageable.
The permalinks in the Godiva WMS viewer that is part of the TDS will
break because they use the variable name to get the data.
I think the human readable names serve the end users better than the
VAR_* names. For example, if I go to NOMADS now and go to a GRIB2 file
and choose the OPeNDAP view, I get a list of variables that I can
The variables that are selectable are in bold letters and easy to read.
I can quickly scroll through the page to find the variable I'm
interested in. While the long_name is listed in lesser print, it doesn't
stand out like the variable name does. In the new scheme, what will
stand out on the page is lots of VAR_* names which all look similar.
You could argue that no one uses this OPeNDAP interface, but I know that
there are some who do.
Or, if I go to the NetcdfSubsetService for a grib file on motherlode:
I see human readable names. In the end, I don't see that the VAR_ names
serve the end user.
As someone on the IDV users list said, "Hal, who do you serve: machines
or humans?" ;-)
On 2/29/12 7:16 AM, Glenn Rutledge wrote:
That is a very good question and I left that out in my response.
Long term access for users in archives means we constantly have to work
to fully document, understand, track down any data provenance issues,
and verifying (to a lessor degree), the data. What it says it is- it
actually is. Its just a form of quality assurance for users. Data
providers - especially 'real time' ones don't necessarily concern
themselves with these issues. They make a product- and move on. I'll
bet you are fully aware that the WOC/Gateway does not even provide a
complete DTG in the file name for many NWP products! I used to work w/
John Stackpole (great guy)- the original developer of Grib. He made grib
as a compact communications protocol- not, as I'll also bet you are also
aware, for archives.
NOMADS has about 1+ petabyte to manage for users- we serviced a growing
550TB last year and we need to scale. By aggregating the data most used
by users (common state variables, most popular, etc.) we can allow
streaming of files/records that allows the 50K+ users and ~300 million
downloads per year on NOMADS much better. Methods such as
pre-staging/caching most requested data on disk from tape, etc. etc.
What John is attempting to do will facilitate the access for multiple
users, requesting multiple files using aggregations and other streaming
caching (I don't quite understand the details there). Now- we can't even
ascertain with any degree of confidence what is what- in order to even
be able to aggregate- let alone feel comfortable about the accuracy of
the data we are serving to users.
It does not really help users find data- per se. It will help users
have more confidence that a aggregated monthly mean product from CFSR is
mean for each cycle (0, 6, 12, ..) for individual days of the month (the
diurnals)- rather then a typical monthly mean avg'ed over the entire day.
hope that makes sense. I'm not sure what other impacts this will have
for us here - LAS? our TDS to ESGF capabilities? It's kinda scary, but
John's radical change looks to solve a major archive problem I do know
that. We will run 4.2 and 4.3 in parallel I will tell you that for
Best regards, Glenn
On Tue, Feb 28, 2012 at 2:19 PM, Don Murray <don.murray@xxxxxxxx
On 2/28/12 11:43 AM, Glenn Rutledge wrote:
John and Community-
While I do not represent the NCDC Archive, for the NCDC NOMADS
and our users, I must agree that the changes John is proposing will
facilitate the long term use of grib data. While painful to
client (software | decoders), the proposed change will allow our
(with a more scalable way) to -better find and use our data. I'll
suggest that if this is adopted, NOMADS servers could provide
and 4.3 versions to (give software developers time to adapt)
client-side to adapt.
Could you elaborate on how you see that the new variable names will
allow the users to better find and use your data versus the human
readable names? For example, if I want to get the 500 hPa heights
from a model in your archive, how will the new names facilitate that?
NOAA/ESRL/PSD and CIRES
Glenn K. Rutledge
NOMADS Team Leader
National Climatic Data Center
Asheville, NC 28801
NOAA/ESRL/PSD and CIRES