ETC Meeting Minutes 05/15/2012

Doug went over recent ingest pauses:

• GSFCS4PA - Ops Ingest paused due to invalid characters on 5/08/12 and 5/10/12.
• LARC - Ops Ingest paused due to invalid characters on 5/03/12.
• LPDAAC - Ops Ingest paused due to invalid characters on 5/03/12 and 5/09/12.
• GHRC - Ops Ingest paused due to inventory synch problems on 5/08/12
• USGS_EROS - Ops Ingest paused due to inventory synch problems on 5/11/12

Other Ops Information:

• Reverb-beta: Extended planned outage
• Expected to be down for 4 hours on 05/14/12 but required extension until 05/15/12
• Testbed: No Issues
• Partner Test: Unplanned Outage
• Order Completion to ECS Providers unavailable on 5/2/12: caused by blocked port (18201) in NetOPS
• EIAT experienced outage due to host name changes from Ops/Dev split activities
• Operations: Unplanned Outage
• Ops Reverb down 5/8/12: caused by issue with URS LDAP permissions, resumed service within 1hour of detection

Previous Week's Reverb Query Metrics

Partner Test performance improvement plans are still in work after the PM of 5/15. We will update as soon as the new machine is successfully put in place.

Multiple Products ordered per granule is still in test by SDPS development. 

ECHO Development put together a summary of orbital search changes:

Major bugs found and resolved:

1. The core of the backtracking algorithm involves finding the orbit swath's edges. It takes a point and asks "If the orbit passes directly over this point, where are its swath edges at the same latitude as the point?" For points with latitudes slightly lower than the orbit's maximum latitude (or slightly higher than its minimum latitude), the legacy code was finding the wrong answer.

It's very difficult to explain the problem in text, so I've attached a picture. It shows the incorrectly-computed point. The legacy code finds all 4 swath edges at the given latitude, then chooses the two that surround the target point. In this case, it chooses incorrectly. The new code chooses the correct points.

Impact on search results: For spatial searches that include regions near the orbit's inflection points (minimum / maximum latitudes), we will potentially filter out granules that the legacy code erroneously included.

2. For orbits with an integer "number of orbits" field, the legacy ECHO code (not the backtrack algorithm) was erroneously considering one too many orbit passes. So, for instance, if the orbit parameters set "number of orbits" to 1, ECHO would return granules for orbits crossing the search area as well as granules for orbits which would cross the search area on their next pass.

Impact on search results: For spatial searches where the dataset's orbit parameters have an integer "number of orbits" field, wewill filter additional granules that the legacy ECHO code erroneously included.

Minor changes to results:

1. The legacy code used a mix of floating point and double precision numbers and often performed more calculations than necessary. This adds a small amount of accumulated error to calculations. The new code consistently uses double precision numbers and performs fewer calculations, resulting in less error. As a result, our calculations differ by a small amount, no more than 100m at the equator and usually much less. We believe the new calculations are more accurate, but in most cases the precision of both results exceeds that of the orbit parameters.

Impact on search results: In rare cases, a granule will be found by the legacy code which is not found by the new code or vice versa. These granules will have swaths whose edges almost exactly touch the edge of a spatial search condition.

2. The new code has not yet been optimized and takes longer to execute than the legacy code. This should only be noticeable when searching very large spatial regions against orbits with a very narrow swath. We know of calculations we can optimize which may allow us to more than make up this gap.

Impact on search results: For searches over very large spatial regions (say, a quarter of the Earth or more) run against orbits with a very narrow swath, searches may take a couple of seconds longer to complete.

Looking to set up a meeting with LPDAAC next week to layout work required to bring functionality to operations.