Table display in Morpho is slow when many columns involved
A table with 176 columns but only 45 records took so long to open that I thought
Morpho had frozen. After 10 minutes or so it opened. The table I am working with
is in a data project that has not been put on the network yet. It is part of the
Data Package: "Data for the Habitat Conservation Planning for Endangered Species
Study" table PQdata-final.txt
X-chat notes: you know what, it just opened up, took about 10 minutes. I could
see the java working on it but I thought maybe it was stuck humm...
<sid> go to a faster machine?
<Callie> next time I'll time it. java was using about 50% processing power and
120 megs of ram
<sid> 50% probably because you have a dual proc machine
<Callie> is that a joke
<Callie> 3.19 GHz .99GB of RAM
just to help clarify, the 3.2 GHz machines have hyperthreading enabled, so 50%
cpu usage is pretty much equal to 100% usage on a non-HT machine
<dan> I worked to support very long tables, but not very wide ones!
<matt_> well, it won't be uncommon
<Callie> yes this one had 176 tables all needing attributes... thanks nick I
didn't realize that 50% CPU = 100 usage
<matt_> its not a good design
<matt_> but we'll have to deal
<matt_> callie -- can you enter a morpho bug to see if we can improve
performance on wide tables
#1 Updated by Callie Bowdish almost 17 years ago
08/12/05 checked opening the table in Morpho again today and it took 10 minutes
with the processor running at maximum. My task manager said Java2.exe was using
about 49% CPU. Nick reports that "hyperthreading enabled, so 50%
cpu usage is pretty much equal to 100% usage on a non-HT machine"
#2 Updated by Will Tyburczy almost 17 years ago
Increase in table load time seems to increase non-linearly as the number of
columns increases. Some sample load times using different tables:
table1, 244 cols: 15 min 30 sec
table1, 122 cols: 2 min 10 sec
table2, 256 cols: 27 min 00 sec
table2, 128 cols: 3 min 50 sec
#3 Updated by Dan Higgins almost 17 years ago
A major cause of this problem is in the section of code that builds the header
of the data table (i.e. column name, unit, etc). THe problem is that
AbstractDataPackage class is called to get that data and each column called a
method to get an array of Attribute information for all columns. The result was
that the attrbute information was called ncolumns number of times and getting
the data increased with ncolumns; so the amount of computation would increase
with square of the number of columns.
A fix was made to the AbstractDataPackage class to save the last list of
attribute nodes and used the saved data rather than regenerate it. This greatly
speeds up the opening of tables with large number of columns. We need to check
that this doesn't create problems (e.g. when a table is nmodified) before
closing this bug.
Dan Higgins --- Aug 30, 2005
#4 Updated by Dan Higgins almost 17 years ago
Added some additional fixes to the AbstractDataPackage class to force a rebuild
of the attributeArray after a column is added or deleted.
Marking this bug as 'fixed'. Tables now appear in a few seconds that previously
required ~1 hr to display! (Dan Higgins)