gistfile1.txt

Loading tool:

    #!/usr/bin/env ruby

    require 'pg'
    require 'digest/sha1'

    # Output a table of current connections to the DB
    conn = PG.connect(dbname: 'indexloadtesting', host: 'localhost')

    1000000.times do
      string = (0...60).map { (65 + rand(26)).chr }.join
      hash = Digest::SHA1.hexdigest(string)
      conn.exec("insert into hashtable values ('#{hash}')")
    end
    
On a table with only 1 column and no indexes. Column type was a varchar(40):

    # time ./indexes_last.rb
    ./indexes_last.rb  44.00s user 21.45s system 42% cpu 2:34.20 total
    
On a table with 1 column and 1 index (again, column type is varchar(40)):

    # time ./indexes_last.rb
    ./indexes_last.rb  43.52s user 21.71s system 31% cpu 3:24.41 total
    
Which is no real surprise, index first takes longer. But this proves nothing on its own.

Indexing the varchar(40) column on the first unindexed table:

    indexloadtesting=# create index foo on hashtable (hash);
    CREATE INDEX
    Time: 32503.838 ms

So wall clock time ... indexes up front is roughly 3:24, indexes last is roughtly 3:06.

As far as comparing them in size:

indexloadtesting=# \di+
                                     List of relations
 Schema |         Name          | Type  | Owner |       Table        | Size  | Description
--------+-----------------------+-------+-------+--------------------+-------+-------------
 public | foo                   | index | ken   | hashtable          | 65 MB |
 public | hashtableindexes_pkey | index | ken   | hashtablewithindex | 85 MB |
(2 rows)