Keywords: DataAccWG

Hi,

I published in docushare the updated spreadsheet with Database Size Estimates.

https://www.lsstcorp.org/docushare/dsweb/Get/Document-1779/lsst_storage_estimates_v10.xls

It is based on assumptions that we have one big Object table
(galaxies + stars) in deep db, and measurements for galaxies
per frame in time db.

The numbers are somewhat bigger than we expected: we are talking
about 400 TB of pure data (no overheads) in the first year now,
comparing to 70 TB before (Ouch!)


I tried to understand why the change is so big. Here it is:


DEEP DB: 3.09 --> 11.7 TB. Why?

 - we used to have row size 180 bytes (stars) & 880 bytes (galaxies)
   Now row size of the Object table is 1.8KB. The fact that there
   there are more stars than galaxies (4:3) in DR1 makes things worse.

 - Why is the row size so much bigger? We have to carry information
   for each filter (basically roughly what we had: 180 bytes
   x 6 filters), plus discussed 50% for reserved space


TIME DB: 19.1 --> 121 TB. Why?
  1) we used to have row size = 70, now it is 306
  2) We added sources for galaxies (ratio galaxy:stars in
     time db is 0.45).
     So that is 19.1 * 306/70 * 0.45 = 121

 - Why did the row size change so much? We were previously guessing
   we will need ~70 bytes for attributes related to stars in
   temp db. Now we are assuming size of Source row (204 bytes),
   plus +50% reserved space, gives 306


Jacek
_______________________________________________
LSST-data mailing list
[email protected]
http://www.lsstmail.org/mailman/listinfo/lsst-data

Reply via email to