pg.ddx.io pgsql-admin@postgresql.org mailing list archive
help / color / mirror / Atom feedIndex (re)-creation speed
4+ messages / 2 participants
[nested] [flat]
* Index (re)-creation speed
@ 2004-02-06 04:36 Steve Lane <slane@moyergroup.com>
0 siblings, 1 reply; 4+ messages in thread
From: Steve Lane @ 2004-02-06 04:36 UTC (permalink / raw)
To: pgsql-admin
All:
Been looking at speed issues related to loading large data sets, and
experimenting with dropping and rebuilding indexes. I noticed some
interesting behavior on the table in question. Here's the table:
Table "test_responses_2"
Attribute | Type | Modifier
-------------------+-------------------+----------
id_response | integer |
id_code | integer |
id_survey | integer |
id_topic | integer |
id_item | integer |
id_administration | integer |
response_int | smallint |
response_txt | character varying |
rec_handle | character varying |
Indices: id_administration_test_key,
id_code_test_key,
id_item_test_key,
id_response_test_key,
id_survey_test_key,
id_topic_test_key
When I drop and rebuild the indexes, they take oddly varying amounts of time
to rebuild. I rebuilt them in the following order, with the following rough
times. I took a guess that the speed of the rebuild might be related to the
number of distinct values in the column -- this seems true in some cases but
not in others. Here are the times:
id_response 38 secs (86000 distinct)
id_topic 33 secs (6 distinct)
id_survey 13 secs (1 distinct)
id_code 39 secs (1444 distinct)
id_item 40 secs (65 distinct)
id_administration 13 secs (1 distinct)
Is there anything I should be learning from this?
-- sgl
=======================================================
Steve Lane
Vice President
The Moyer Group
14 North Peoria St Suite 2H
Chicago, IL 60607
Voice: (312) 433-2421 Email: slane@moyergroup.com
Fax: (312) 850-3930 Web: http://www.moyergroup.com
=======================================================
^ permalink raw reply [nested|flat] 4+ messages in thread
* Re: Index (re)-creation speed
@ 2004-02-06 06:30 Tom Lane <tgl@sss.pgh.pa.us>
parent: Steve Lane <slane@moyergroup.com>
0 siblings, 1 reply; 4+ messages in thread
From: Tom Lane @ 2004-02-06 06:30 UTC (permalink / raw)
To: Steve Lane <slane@moyergroup.com>; +Cc: pgsql-admin
Steve Lane <slane@moyergroup.com> writes:
> When I drop and rebuild the indexes, they take oddly varying amounts of time
> to rebuild. I rebuilt them in the following order, with the following rough
> times. I took a guess that the speed of the rebuild might be related to the
> number of distinct values in the column -- this seems true in some cases but
> not in others. Here are the times:
> id_response 38 secs (86000 distinct)
> id_topic 33 secs (6 distinct)
> id_survey 13 secs (1 distinct)
> id_code 39 secs (1444 distinct)
> id_item 40 secs (65 distinct)
> id_administration 13 secs (1 distinct)
How many rows altogether in this table? What sort_mem are you using
(and have you tried altering it)?
When I read your previous message I was wondering about different
datatypes having different comparison costs, but since these are all
integer columns that's clearly not the issue here. I think you may
be seeing something related to the number of initial runs created in the
sorting algorithm --- with only one distinct value, there'd always be
just one run and no need for any merge passes.
(Why are you bothering to index columns with only one or a few values in
them, anyway? Such an index will be useless for searches ...)
regards, tom lane
^ permalink raw reply [nested|flat] 4+ messages in thread
* Re: Index (re)-creation speed
@ 2004-02-06 07:03 Steve Lane <slane@moyergroup.com>
parent: Tom Lane <tgl@sss.pgh.pa.us>
0 siblings, 1 reply; 4+ messages in thread
From: Steve Lane @ 2004-02-06 07:03 UTC (permalink / raw)
To: pgsql-admin
On 2/6/04 12:30 AM, "Tom Lane" <tgl@sss.pgh.pa.us> wrote:
> Steve Lane <slane@moyergroup.com> writes:
>> When I drop and rebuild the indexes, they take oddly varying amounts of time
>> to rebuild. I rebuilt them in the following order, with the following rough
>> times. I took a guess that the speed of the rebuild might be related to the
>> number of distinct values in the column -- this seems true in some cases but
>> not in others. Here are the times:
>
>> id_response 38 secs (86000 distinct)
>> id_topic 33 secs (6 distinct)
>> id_survey 13 secs (1 distinct)
>> id_code 39 secs (1444 distinct)
>> id_item 40 secs (65 distinct)
>> id_administration 13 secs (1 distinct)
>
> How many rows altogether in this table? What sort_mem are you using
> (and have you tried altering it)?
Sorry, left out important info. I cleaned the table first (it's a temporary
copy), then imported one run of data, 86,000 rows.
I haven't checked sort_mem, will do so.
>
> When I read your previous message I was wondering about different
> datatypes having different comparison costs, but since these are all
> integer columns that's clearly not the issue here. I think you may
> be seeing something related to the number of initial runs created in the
> sorting algorithm --- with only one distinct value, there'd always be
> just one run and no need for any merge passes.
>
> (Why are you bothering to index columns with only one or a few values in
> them, anyway? Such an index will be useless for searches ...)
Again, I left out some useful details. Id_survey and id_administration will
be identical for a single run of data (which can generate 10-100K inserts).
And my tests were just one data set, imported into a cleaned table. Still,
the distinctness of these columns will be low -- on the order of one
distinct value per 10^4 rows. The others have levels of distinctness
proportional to what the above chart shows -- id_response is unique,
id_topic will have 4-8 values per 10^4 records, and so on.
And still, none of this explains to me why the index on id_item drove the
COPY cost up so apparently dramatically. I tried the COPY again both with
and without that one index and in both cases it caused COPY to go from 5 to
40+ seconds.
-- sgl
^ permalink raw reply [nested|flat] 4+ messages in thread
* Re: Index (re)-creation speed
@ 2004-02-06 14:50 Tom Lane <tgl@sss.pgh.pa.us>
parent: Steve Lane <slane@moyergroup.com>
0 siblings, 0 replies; 4+ messages in thread
From: Tom Lane @ 2004-02-06 14:50 UTC (permalink / raw)
To: Steve Lane <slane@moyergroup.com>; +Cc: pgsql-admin
Steve Lane <slane@moyergroup.com> writes:
> And still, none of this explains to me why the index on id_item drove the
> COPY cost up so apparently dramatically. I tried the COPY again both with
> and without that one index and in both cases it caused COPY to go from 5 to
> 40+ seconds.
That doesn't make a lot of sense to me either, particularly seeing that
id_item seems to be an intermediate case as far as uniqueness goes.
Do you still get 30+ seconds to COPY if that is the *only* index?
If the data you're using isn't too sensitive, would you mind sending me
a copy (off-list)? I'm interested to try profiling this behavior.
regards, tom lane
^ permalink raw reply [nested|flat] 4+ messages in thread
end of thread, other threads:[~2004-02-06 14:50 UTC | newest]
Thread overview: 4+ messages (download: mbox mbox.gz follow: Atom feed)
-- links below jump to the message on this page --
2004-02-06 04:36 Index (re)-creation speed Steve Lane <slane@moyergroup.com>
2004-02-06 06:30 ` Tom Lane <tgl@sss.pgh.pa.us>
2004-02-06 07:03 ` Steve Lane <slane@moyergroup.com>
2004-02-06 14:50 ` Tom Lane <tgl@sss.pgh.pa.us>
This inbox is served by DDX for PostgreSQL; see mirroring instructions
for how to clone and mirror all data and code used for this inbox