VACUUM degrades performance significantly. Database becomes unusable!

[Stephen]

Hello,

Is it normal for plain VACUUM on large table to degrade performance by over
9 times? My database becomes unusable when VACUUM runs. From reading
newsgroups, I thought VACUUM should only slow down by 10% to 15%. Other MVCC
databases like MySQL InnoDB can even VACUUM discretely (runs internally). Is
it my Linux system or is it PostgreSQL?

The database is mostly read-only. There are 133,000 rows and each row is
about 2.5kB in size (mostly due to the bytea column holding a binary image).
The long row causes system to TOAST the table. VACUUM takes 5m20s to
complete. I repeatedly ran the following tests while system is idling:
In normal operation:
====================
tsdb=# explain analyze select * from table1 where id =
'33a4e9b6eae09634f4ff3e6fa9280f6e';
QUERY PLAN
----------------------------------------------------------------------------
----------------------------------------------------
Index Scan using table1_pkey on table1 (cost=0.00..6.01 rows=1 width=346)
(actual time=25.30..25.31 rows=1 loops=1)
Index Cond: (id = '33a4e9b6eae09634f4ff3e6fa9280f6e'::character varying)
Total runtime: 25.52 msec
(3 rows)

When VACUUM runs:
=================
tsdb=# explain analyze select * from table1 where id =
'336139b47b7faf09fc4d4f03680a4ce5';
QUERY PLAN
----------------------------------------------------------------------------
--------------------------------------------------------
Index Scan using table1_pkey on table1 (cost=0.00..6.01 rows=1 width=346)
(actual time=2290.07..2290.10 rows=1 loops=1)
Index Cond: (id = '336139b47b7faf09fc4d4f03680a4ce5'::character varying)
Total runtime: 2290.22 msec
(3 rows)
VACUUM output:
==============
tsdb=# VACUUM VERBOSE table1;
INFO: --Relation public.table1--
INFO: Pages 5887: Changed 0, Empty 0; Tup 132672: Vac 0, Keep 0, UnUsed
144.
Total CPU 0.28s/0.01u sec elapsed 36.08 sec.
INFO: --Relation pg_toast.pg_toast_12437088--
INFO: Pages 40495: Changed 0, Empty 0; Tup 197587: Vac 0, Keep 0, UnUsed
235.
Total CPU 1.73s/0.20u sec elapsed 233.91 sec.
VACUUM

vmstat while VACUUM'ing:
========================
procs memory swap io system
cpu
r b w swpd free buff cache si so bi bo in cs us sy
id
0 1 1 74420 6520 30616 405128 0 0 1280 0 287 487 0 1
99
0 1 0 74420 6520 30620 405168 0 0 1196 0 271 436 0 0
100
0 1 1 74420 6520 30620 405120 0 0 1496 4 289 491 0 3
97
0 1 1 74420 6520 30620 405208 0 0 1280 0 268 466 0 0
100
1 0 1 74420 6520 30620 405208 0 0 1280 0 288 482 0 1
99
1 0 1 74420 6520 30632 405200 0 0 1416 8 277 441 1 2
97
3 1 1 74416 6520 30632 405196 4 0 1284 0 284 473 0 3
97

PostgreSQL configuration (the only changes made):
=================================================
max_connections = 1024
shared_buffers = 2800
sort_mem = 8192
vacuum_mem = 8192
effective_cache_size = 32000
System:
=======
Hardware: AMD 1.2GHz Athlon 512MB SDRAM
OS: Redhat Linux 9.0 (kernel 2.4.20-8)
FS: EXT3 with Journalling mounted with noatime, UDMA5
Disk1 (Linux): IDE 20GB 7200 RPM Western Digital
Disk2 (PostgreSQL): IDE 120GB 7200 RPM Western Digital
PostgreSQL: 7.3.4
hdparm:
=======
/dev/hda: (Linux partition)
multcount = 16 (on)
IO_support = 1 (32-bit)
unmaskirq = 1 (on)
using_dma = 1 (on)
keepsettings = 0 (off)
readonly = 0 (off)
readahead = 8 (on)
geometry = 2498/255/63, sectors = 40132503, start = 0

/dev/hdc: (PostgreSQL partition)
multcount = 16 (on)
IO_support = 1 (32-bit)
unmaskirq = 1 (on)
using_dma = 1 (on)
keepsettings = 1 (on)
readonly = 0 (off)
readahead = 8 (on)
geometry = 232581/16/63, sectors = 234441648, start = 0
Schema:
=======
CREATE TABLE table1 (
id varchar(32) DEFAULT '' NOT NULL,
colname1 varchar(10) DEFAULT 'http' NOT
NULL,
colname2 varchar(300) DEFAULT '' NOT NULL,
colname3 varchar(5) DEFAULT '' NOT
NULL,
colname4 varchar(300) DEFAULT '' NOT
NULL,
colname5 varchar(300) DEFAULT '' NOT NULL,
colname6 integer DEFAULT 0 NOT
NULL,
colname7 integer DEFAULT 0 NOT
NULL,
colname8 integer DEFAULT 200 NOT NULL,
colname9 varchar(10) DEFAULT '' NOT
NULL,
colname10 varchar(10) DEFAULT '' NOT NULL,
colname11 varchar(100) DEFAULT '' NOT
NULL,
colname12 varchar(100) DEFAULT '' NOT NULL,
colname13 varchar(100) DEFAULT '' NOT NULL,
colname14 varchar(20) DEFAULT '' NOT NULL,
colname15 integer DEFAULT 640 NOT
NULL,
colname16 integer DEFAULT 480 NOT
NULL,
colname17 integer DEFAULT 120 NOT
NULL,
colname18 integer DEFAULT 90 NOT
NULL,
colname19 timestamp DEFAULT CURRENT_TIMESTAMP NOT
NULL,
colname20 timestamp DEFAULT CURRENT_TIMESTAMP NOT
NULL,
colname21 integer DEFAULT 0 NOT NULL,
colname22 integer DEFAULT 0 NOT NULL,
colname23 timestamp DEFAULT CURRENT_TIMESTAMP NOT NULL,
colname24 integer DEFAULT 0 NOT
NULL,
colname25 integer DEFAULT 0 NOT NULL,
colname26 varchar(10) DEFAULT '' NOT
NULL,
colname28 varchar(10) DEFAULT '' NOT NULL,
colname29 varchar(10) DEFAULT 'jpeg' NOT
NULL,
colname30 varchar(20) DEFAULT '' NOT NULL,
colname31 bytea ,
PRIMARY KEY (id)
) WITHOUT OIDS

CREATE UNIQUE INDEX table1_idx_1 ON table1 (colname1, colname2, colname3,
colname4, colname5)

Sigh, :-(

Stephen

[Tom Lane]

"Stephen" <jl*****@xxxxxx.com> writes:

Is it normal for plain VACUUM on large table to degrade performance by over
9 times? My database becomes unusable when VACUUM runs. From reading
newsgroups, I thought VACUUM should only slow down by 10% to 15%.

We have heard reports of very significant slowdowns from people who were
already nearly saturating their disk I/O bandwidth, and then VACUUM
pushed their systems over the knee of the response curve. You haven't
said anything about what was happening in your system besides the VACUUM
and the test query, but I wonder how busy the disk drive was already.

Another possibility is that you've got the disk subsystem misconfigured
somehow, although your hdparm output looks okay to my not-expert eyes.

Possibly you should think about SCSI instead of IDE disk drives.
Consumer-grade IDE drives seem to be not very good about servicing
parallel loads. As an example, running "pgbench" with 10 client threads
on a fairly new Dell PC (P4, some IDE drive or other), I see about a 3x
slowdown in reported transactions-per-second when a background VACUUM
is running. The same test on my trusty old built-like-a-tank HP server
(with fast-for-its-time SCSI drives) shows only a 20% slowdown. I
attribute this to the SCSI drive being better able to handle concurrent
requests.

regards, tom lane

---------------------------(end of broadcast)---------------------------
TIP 1: subscribe and unsubscribe commands go to ma*******@postgresql.org

[Stephen]

The system is actually idling when I ran the tests (load average: 0.01,
0.02, 0.00). When VACUUM runs, load average increases quickly (load average:
1.77, 0.60, 0.21) just by running psql on command line and issuing 2
queries. I've been running PostgreSQL for over 3 years now and IIRC VACUUM
has always been a show-stopper with or without the FULL attribute on my
system (from linux 2.2 to linux 2.4, and from PostgreSQL 7.1 to PostgreSQL
7.3). The only thing that didn't change is the IDE drive. The problem is not
unique to this system, I have 2 machines with the same configuration and
VACUUM slows down both databases considerably. Unfortunately, I prefer to
implement a system on cheap IDE disks a la Google because my database is
expected to hit terabytes and SCSI cost becomes prohibitive.

Is anyone else getting performance degradation with VACUUM on Linux? I'm
thinking maybe Linux has a bad scheduler and high IO latency that other OSes
(eg. FreeBSD) do not suffer from. Although, someone did tell me that Redhat
9.0 Linux 2.4.20-8 has parts of the low latency patch applied but I
certainly don't feel the improvement on PostgreSQL.

[scott.marlowe]

It sounds like you might be I/O bound. if you drop the 5 way unique index
for a test, how do the vacuum and parallel select run?

On Wed, 15 Oct 2003, Stephen wrote:

Hello,

Is it normal for plain VACUUM on large table to degrade performance by over
9 times? My database becomes unusable when VACUUM runs. From reading
newsgroups, I thought VACUUM should only slow down by 10% to 15%. Other MVCC
databases like MySQL InnoDB can even VACUUM discretely (runs internally). Is
it my Linux system or is it PostgreSQL?

The database is mostly read-only. There are 133,000 rows and each row is
about 2.5kB in size (mostly due to the bytea column holding a binary image).
The long row causes system to TOAST the table. VACUUM takes 5m20s to
complete. I repeatedly ran the following tests while system is idling:
In normal operation:
====================
tsdb=# explain analyze select * from table1 where id =
'33a4e9b6eae09634f4ff3e6fa9280f6e';
QUERY PLAN
----------------------------------------------------------------------------
----------------------------------------------------
Index Scan using table1_pkey on table1 (cost=0.00..6.01 rows=1 width=346)
(actual time=25.30..25.31 rows=1 loops=1)
Index Cond: (id = '33a4e9b6eae09634f4ff3e6fa9280f6e'::character varying)
Total runtime: 25.52 msec
(3 rows)

When VACUUM runs:
=================
tsdb=# explain analyze select * from table1 where id =
'336139b47b7faf09fc4d4f03680a4ce5';
QUERY PLAN
----------------------------------------------------------------------------
--------------------------------------------------------
Index Scan using table1_pkey on table1 (cost=0.00..6.01 rows=1 width=346)
(actual time=2290.07..2290.10 rows=1 loops=1)
Index Cond: (id = '336139b47b7faf09fc4d4f03680a4ce5'::character varying)
Total runtime: 2290.22 msec
(3 rows)
VACUUM output:
==============
tsdb=# VACUUM VERBOSE table1;
INFO: --Relation public.table1--
INFO: Pages 5887: Changed 0, Empty 0; Tup 132672: Vac 0, Keep 0, UnUsed
144.
Total CPU 0.28s/0.01u sec elapsed 36.08 sec.
INFO: --Relation pg_toast.pg_toast_12437088--
INFO: Pages 40495: Changed 0, Empty 0; Tup 197587: Vac 0, Keep 0, UnUsed
235.
Total CPU 1.73s/0.20u sec elapsed 233.91 sec.
VACUUM

vmstat while VACUUM'ing:
========================
procs memory swap io system
cpu
r b w swpd free buff cache si so bi bo in cs us sy
id
0 1 1 74420 6520 30616 405128 0 0 1280 0 287 487 0 1
99
0 1 0 74420 6520 30620 405168 0 0 1196 0 271 436 0 0
100
0 1 1 74420 6520 30620 405120 0 0 1496 4 289 491 0 3
97
0 1 1 74420 6520 30620 405208 0 0 1280 0 268 466 0 0
100
1 0 1 74420 6520 30620 405208 0 0 1280 0 288 482 0 1
99
1 0 1 74420 6520 30632 405200 0 0 1416 8 277 441 1 2
97
3 1 1 74416 6520 30632 405196 4 0 1284 0 284 473 0 3
97

PostgreSQL configuration (the only changes made):
=================================================
max_connections = 1024
shared_buffers = 2800
sort_mem = 8192
vacuum_mem = 8192
effective_cache_size = 32000
System:
=======
Hardware: AMD 1.2GHz Athlon 512MB SDRAM
OS: Redhat Linux 9.0 (kernel 2.4.20-8)
FS: EXT3 with Journalling mounted with noatime, UDMA5
Disk1 (Linux): IDE 20GB 7200 RPM Western Digital
Disk2 (PostgreSQL): IDE 120GB 7200 RPM Western Digital
PostgreSQL: 7.3.4
hdparm:
=======
/dev/hda: (Linux partition)
multcount = 16 (on)
IO_support = 1 (32-bit)
unmaskirq = 1 (on)
using_dma = 1 (on)
keepsettings = 0 (off)
readonly = 0 (off)
readahead = 8 (on)
geometry = 2498/255/63, sectors = 40132503, start = 0

/dev/hdc: (PostgreSQL partition)
multcount = 16 (on)
IO_support = 1 (32-bit)
unmaskirq = 1 (on)
using_dma = 1 (on)
keepsettings = 1 (on)
readonly = 0 (off)
readahead = 8 (on)
geometry = 232581/16/63, sectors = 234441648, start = 0
Schema:
=======
CREATE TABLE table1 (
id varchar(32) DEFAULT '' NOT NULL,
colname1 varchar(10) DEFAULT 'http' NOT
NULL,
colname2 varchar(300) DEFAULT '' NOT NULL,
colname3 varchar(5) DEFAULT '' NOT
NULL,
colname4 varchar(300) DEFAULT '' NOT
NULL,
colname5 varchar(300) DEFAULT '' NOT NULL,
colname6 integer DEFAULT 0 NOT
NULL,
colname7 integer DEFAULT 0 NOT
NULL,
colname8 integer DEFAULT 200 NOT NULL,
colname9 varchar(10) DEFAULT '' NOT
NULL,
colname10 varchar(10) DEFAULT '' NOT NULL,
colname11 varchar(100) DEFAULT '' NOT
NULL,
colname12 varchar(100) DEFAULT '' NOT NULL,
colname13 varchar(100) DEFAULT '' NOT NULL,
colname14 varchar(20) DEFAULT '' NOT NULL,
colname15 integer DEFAULT 640 NOT
NULL,
colname16 integer DEFAULT 480 NOT
NULL,
colname17 integer DEFAULT 120 NOT
NULL,
colname18 integer DEFAULT 90 NOT
NULL,
colname19 timestamp DEFAULT CURRENT_TIMESTAMP NOT
NULL,
colname20 timestamp DEFAULT CURRENT_TIMESTAMP NOT
NULL,
colname21 integer DEFAULT 0 NOT NULL,
colname22 integer DEFAULT 0 NOT NULL,
colname23 timestamp DEFAULT CURRENT_TIMESTAMP NOT NULL,
colname24 integer DEFAULT 0 NOT
NULL,
colname25 integer DEFAULT 0 NOT NULL,
colname26 varchar(10) DEFAULT '' NOT
NULL,
colname28 varchar(10) DEFAULT '' NOT NULL,
colname29 varchar(10) DEFAULT 'jpeg' NOT
NULL,
colname30 varchar(20) DEFAULT '' NOT NULL,
colname31 bytea ,
PRIMARY KEY (id)
) WITHOUT OIDS

CREATE UNIQUE INDEX table1_idx_1 ON table1 (colname1, colname2, colname3,
colname4, colname5)

Sigh, :-(

Stephen


---------------------------(end of broadcast)---------------------------
TIP 6: Have you searched our list archives?

http://archives.postgresql.org

---------------------------(end of broadcast)---------------------------
TIP 4: Don't 'kill -9' the postmaster

[Stephen]

Scott,

I dropped the 5 way unique index and the VACUUM improved slightly. I ran
VACUUM, ANALYZE, VACUUM and queries repeatedly. The max response time seem
to have reduced to 1700 msec from 2300 msec. The higher load and vmstat
during VACUUM remained the same. It's still not enough to justify dropping
the index for my purposes.

tsdb=# explain analyze select * from table1 where id =
'3305b141837f065d673aa09cf382d331';
QUERY PLAN
----------------------------------------------------------------------------
--------------------------------------------------------
Index Scan using table1_pkey on table1 (cost=0.00..6.01 rows=1 width=346)
(actual time=1762.34..1762.37 rows=1 loops=1)
Index Cond: (id = '3305b141837f065d673aa09cf382d331'::character varying)
Total runtime: 1762.50 msec
(3 rows)

Regards,

Stephen

""scott.marlowe"" <sc***********@ihs.com> wrote in message
news:Pi**************************************@css1 20.ihs.com...

It sounds like you might be I/O bound. if you drop the 5 way unique index
for a test, how do the vacuum and parallel select run?

On Wed, 15 Oct 2003, Stephen wrote:

Hello,

Is it normal for plain VACUUM on large table to degrade performance by over 9 times? My database becomes unusable when VACUUM runs. From reading
newsgroups, I thought VACUUM should only slow down by 10% to 15%. Other MVCC databases like MySQL InnoDB can even VACUUM discretely (runs internally). Is it my Linux system or is it PostgreSQL?

The database is mostly read-only. There are 133,000 rows and each row is
about 2.5kB in size (mostly due to the bytea column holding a binary image). The long row causes system to TOAST the table. VACUUM takes 5m20s to
complete. I repeatedly ran the following tests while system is idling:
In normal operation:
====================
tsdb=# explain analyze select * from table1 where id =
'33a4e9b6eae09634f4ff3e6fa9280f6e';
QUERY PLAN

--------------------------------------------------------------------------

--

----------------------------------------------------
Index Scan using table1_pkey on table1 (cost=0.00..6.01 rows=1 width=346) (actual time=25.30..25.31 rows=1 loops=1)
Index Cond: (id = '33a4e9b6eae09634f4ff3e6fa9280f6e'::character varying) Total runtime: 25.52 msec
(3 rows)

When VACUUM runs:
=================
tsdb=# explain analyze select * from table1 where id =
'336139b47b7faf09fc4d4f03680a4ce5';
QUERY PLAN

--------------------------------------------------------------------------

--

--------------------------------------------------------
Index Scan using table1_pkey on table1 (cost=0.00..6.01 rows=1 width=346) (actual time=2290.07..2290.10 rows=1 loops=1)
Index Cond: (id = '336139b47b7faf09fc4d4f03680a4ce5'::character varying) Total runtime: 2290.22 msec
(3 rows)
VACUUM output:
==============
tsdb=# VACUUM VERBOSE table1;
INFO: --Relation public.table1--
INFO: Pages 5887: Changed 0, Empty 0; Tup 132672: Vac 0, Keep 0, UnUsed
144.
Total CPU 0.28s/0.01u sec elapsed 36.08 sec.
INFO: --Relation pg_toast.pg_toast_12437088--
INFO: Pages 40495: Changed 0, Empty 0; Tup 197587: Vac 0, Keep 0, UnUsed 235.
Total CPU 1.73s/0.20u sec elapsed 233.91 sec.
VACUUM

vmstat while VACUUM'ing:
========================
procs memory swap io system
cpu
r b w swpd free buff cache si so bi bo in cs us sy id
0 1 1 74420 6520 30616 405128 0 0 1280 0 287 487 0 1 99
0 1 0 74420 6520 30620 405168 0 0 1196 0 271 436 0 0 100
0 1 1 74420 6520 30620 405120 0 0 1496 4 289 491 0 3 97
0 1 1 74420 6520 30620 405208 0 0 1280 0 268 466 0 0 100
1 0 1 74420 6520 30620 405208 0 0 1280 0 288 482 0 1 99
1 0 1 74420 6520 30632 405200 0 0 1416 8 277 441 1 2 97
3 1 1 74416 6520 30632 405196 4 0 1284 0 284 473 0 3 97

PostgreSQL configuration (the only changes made):
=================================================
max_connections = 1024
shared_buffers = 2800
sort_mem = 8192
vacuum_mem = 8192
effective_cache_size = 32000
System:
=======
Hardware: AMD 1.2GHz Athlon 512MB SDRAM
OS: Redhat Linux 9.0 (kernel 2.4.20-8)
FS: EXT3 with Journalling mounted with noatime, UDMA5
Disk1 (Linux): IDE 20GB 7200 RPM Western Digital
Disk2 (PostgreSQL): IDE 120GB 7200 RPM Western Digital
PostgreSQL: 7.3.4
hdparm:
=======
/dev/hda: (Linux partition)
multcount = 16 (on)
IO_support = 1 (32-bit)
unmaskirq = 1 (on)
using_dma = 1 (on)
keepsettings = 0 (off)
readonly = 0 (off)
readahead = 8 (on)
geometry = 2498/255/63, sectors = 40132503, start = 0

/dev/hdc: (PostgreSQL partition)
multcount = 16 (on)
IO_support = 1 (32-bit)
unmaskirq = 1 (on)
using_dma = 1 (on)
keepsettings = 1 (on)
readonly = 0 (off)
readahead = 8 (on)
geometry = 232581/16/63, sectors = 234441648, start = 0
Schema:
=======
CREATE TABLE table1 (
id varchar(32) DEFAULT '' NOT NULL, colname1 varchar(10) DEFAULT 'http' NOT
NULL,
colname2 varchar(300) DEFAULT '' NOT NULL, colname3 varchar(5) DEFAULT '' NOT NULL,
colname4 varchar(300) DEFAULT '' NOT
NULL,
colname5 varchar(300) DEFAULT '' NOT NULL, colname6 integer DEFAULT 0 NOT NULL,
colname7 integer DEFAULT 0 NOT NULL,
colname8 integer DEFAULT 200 NOT NULL, colname9 varchar(10) DEFAULT '' NOT
NULL,
colname10 varchar(10) DEFAULT '' NOT NULL, colname11 varchar(100) DEFAULT '' NOT NULL,
colname12 varchar(100) DEFAULT '' NOT NULL, colname13 varchar(100) DEFAULT '' NOT NULL, colname14 varchar(20) DEFAULT '' NOT NULL,
colname15 integer DEFAULT 640 NOT
NULL,
colname16 integer DEFAULT 480 NOT
NULL,
colname17 integer DEFAULT 120 NOT NULL,
colname18 integer DEFAULT 90 NOT NULL,
colname19 timestamp DEFAULT CURRENT_TIMESTAMP NOT
NULL,
colname20 timestamp DEFAULT CURRENT_TIMESTAMP NOT
NULL,
colname21 integer DEFAULT 0 NOT NULL,
colname22 integer DEFAULT 0 NOT NULL, colname23 timestamp DEFAULT CURRENT_TIMESTAMP NOT NULL,
colname24 integer DEFAULT 0 NOT NULL,
colname25 integer DEFAULT 0 NOT NULL, colname26 varchar(10) DEFAULT '' NOT NULL,
colname28 varchar(10) DEFAULT '' NOT NULL,
colname29 varchar(10) DEFAULT 'jpeg' NOT
NULL,
colname30 varchar(20) DEFAULT '' NOT NULL,
colname31 bytea , PRIMARY KEY (id)
) WITHOUT OIDS

CREATE UNIQUE INDEX table1_idx_1 ON table1 (colname1, colname2, colname3, colname4, colname5)

Sigh, :-(

Stephen


---------------------------(end of broadcast)---------------------------
TIP 6: Have you searched our list archives?

http://archives.postgresql.org

---------------------------(end of broadcast)---------------------------
TIP 4: Don't 'kill -9' the postmaster

[Matthew T. O'Connor]

On Wed, 2003-10-15 at 12:57, Tom Lane wrote:

"Stephen" <jl*****@xxxxxx.com> writes:

Is it normal for plain VACUUM on large table to degrade performance by over
9 times? My database becomes unusable when VACUUM runs. From reading
newsgroups, I thought VACUUM should only slow down by 10% to 15%.

We have heard reports of very significant slowdowns from people who were
already nearly saturating their disk I/O bandwidth, and then VACUUM
pushed their systems over the knee of the response curve.

Makes sense but, wouldn't vacuum saturate the I/O bandwidth by
definition? Especially with modern CPU's when vacuum runs it's going to
be reading from disk as fast as the disk can possibly supply the data.
For this reason, I think the suggestion you made recently of putting in
a small delay in the main vacuum loop might be helpful.

I also acknowledge that SCSI makes a big difference here, but LOTS of
people run PG on cheap linux boxes with IDE drives, so if there is
something we can do to help this setup, it would be "a good thing" IMHO.
---------------------------(end of broadcast)---------------------------
TIP 2: you can get off all lists at once with the unregister command
(send "unregister YourEmailAddressHere" to ma*******@postgresql.org)

[Stephen]

Good news,

I partially fixed the problem on Linux 2.4. It appears the responsiveness
can be improved significantly by tuning the disk IO elevator in Linux using
"elvtune" in util-linux. The elevator in Linux is used to re-order
read/write requests to reduce disk seeks by ordering requests according to
disk sectors. Unfortunately, the elevator in kernel 2.4 is not very smart
(or flexible I should say depending on your needs) and can starve a
read/write request for a long time if not properly tuned.

See my older thread I wrote long time ago. Back then I didn't have too much
time to test different values especially for VACUUM:

http://groups.google.com/groups?q=li...+postgresql&hl
=en&lr=&ie=UTF-8&oe=UTF-8&selm=F92Znj0TrJIPT6nhdBf00021ae6%40hotmail.com&r nu
m=1

See also:

http://strasbourg.linuxfr.org/jl3/features-2.3-1.html
Below are the results using different elvtune values running repeatedly when
VACUUM'ing.

elvtune -r 2048 -w 8192 /dev/hdc (default Redhat 9):
================================================== ==

tsdb=# explain analyze select * from table1 where id =
'336139b47b7faf09fc4d4f03680a4ce5';
QUERY PLAN
----------------------------------------------------------------------------
--------------------------------------------------------
Index Scan using table1_pkey on table1 (cost=0.00..6.01 rows=1 width=346)
(actual time=2290.07..2290.10 rows=1 loops=1)
Index Cond: (id = '336139b47b7faf09fc4d4f03680a4ce5'::character varying)
Total runtime: 2290.22 msec
(3 rows)
elvtune -r 128 -w 8192 /dev/hdc:
================================

tsdb=# explain analyze select * from table1 where id =
'008ab286d725d2ea0b3269c89fc01ce2';
QUERY PLAN
----------------------------------------------------------------------------
------------------------------------------------------
Index Scan using table1_pkey on table1 (cost=0.00..6.01 rows=1 width=346)
(actual time=450.29..450.32 rows=1 loops=1)
Index Cond: (id = '008ab286d725d2ea0b3269c89fc01ce2'::character varying)
Total runtime: 450.46 msec
(3 rows)
elvtune -r 64 -w 8192 /dev/hdc:
===============================

tsdb=# explain analyze select * from table1 where id =
'0078997ac809877c1a0d1f76af753608';
QUERY PLAN
----------------------------------------------------------------------------
------------------------------------------------------
Index Scan using table1_pkey on table1 (cost=0.00..6.01 rows=1 width=346)
(actual time=206.00..206.01 rows=1 loops=1)
Index Cond: (id = '0078997ac809877c1a0d1f76af753608'::character varying)
Total runtime: 206.14 msec
(3 rows)
elvtune -r 32 -w 8192 /dev/hdc:
===============================

tsdb=# explain analyze select * from table1 where id =
'00c45490c9f24858c17d7dfb98c5def5';
QUERY PLAN
----------------------------------------------------------------------------
------------------------------------------------------
Index Scan using table1_pkey on table1 (cost=0.00..6.01 rows=1 width=346)
(actual time=210.61..210.62 rows=1 loops=1)
Index Cond: (id = '00c45490c9f24858c17d7dfb98c5def5'::character varying)
Total runtime: 210.75 msec
(3 rows)
elvtune -r 8 -w 8192 /dev/hdc:
==============================

tsdb=# explain analyze select * from table1 where id =
'006ea95ef6b9b8f0ddcb1f33c40190ec';
QUERY PLAN
----------------------------------------------------------------------------
------------------------------------------------------
Index Scan using table1_pkey on table1 (cost=0.00..6.01 rows=1 width=346)
(actual time=338.04..338.05 rows=1 loops=1)
Index Cond: (id = '006ea95ef6b9b8f0ddcb1f33c40190ec'::character varying)
Total runtime: 338.18 msec
(3 rows)
elvtune -r 1 -w 8192 /dev/hdc:
==============================

tsdb=# explain analyze select * from table1 where id =
'0025a7a9182d5456474a72f773433c01';
QUERY PLAN
----------------------------------------------------------------------------
------------------------------------------------------
Index Scan using table1_pkey on table1 (cost=0.00..6.01 rows=1 width=346)
(actual time=390.41..390.42 rows=1 loops=1)
Index Cond: (id = '0025a7a9182d5456474a72f773433c01'::character varying)
Total runtime: 390.55 msec
(3 rows)

elvtune -r 0 -w 8192 /dev/hdc:
==============================

tsdb=# explain analyze select * from table1 where id =
'0023783eda7e2a5f434e55a66c3a0a11';
QUERY PLAN
----------------------------------------------------------------------------
------------------------------------------------------
Index Scan using table1_pkey on table1 (cost=0.00..6.01 rows=1 width=346)
(actual time=246.29..246.32 rows=1 loops=1)
Index Cond: (id = '0023783eda7e2a5f434e55a66c3a0a11'::character varying)
Total runtime: 246.44 msec
(3 rows)

tsdb=# explain analyze select * from table1 where id =
'001eacb36161ac6a1f860bd391dce5c2';
QUERY PLAN
----------------------------------------------------------------------------
----------------------------------------------------
Index Scan using table1_pkey on table1 (cost=0.00..6.01 rows=1 width=346)
(actual time=14.11..14.12 rows=1 loops=1)
Index Cond: (id = '001eacb36161ac6a1f860bd391dce5c2'::character varying)
Total runtime: 14.25 msec
(3 rows)

Remarks:
========

In all cases, VACUUM completed in the same amount of time, process load
average is reduced slightly and vmstat IO is the same.

It turns out that as the -r read value in elvtune is made smaller, the more
responsive the read becomes up to a certain point when the disk needs to do
too many unoptimized seeks. A too small read value can cause the response
time to fluctuate more than a large value.

Understandably, the elvtune values are quite specific depending on what you
do. In my case, where the table is mostly read-only, "elevtune -r 64 -w 8192
/dev/hdc" works best. It should also depend on your expected load, purpose,
disk type, size and settings.

VACUUM is still disk IO intensive. Even with "elvtune -r 64 -w 8192
/dev/hdc" and VACUUM, the response time has gone down to 200 msec from 2200
msec (10 times factor), is still high compared to normal queries at 25 msec.
VACUUM needs to clamped down much more!

Regards,

Stephen

[Vivek Khera]

>>>>> "S" == Stephen <jl*****@xxxxxx.com> writes:

S> The system is actually idling when I ran the tests (load average: 0.01,
S> 0.02, 0.00). When VACUUM runs, load average increases quickly (load average:
S> 1.77, 0.60, 0.21) just by running psql on command line and issuing 2
S> queries. I've been running PostgreSQL for over 3 years now and IIRC VACUUM

Your CPU usage is irrelevent here. What's your *disk* utilization.
Does linux have systat and/or iostat to tell you how busy your disks
are?

--
=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
Vivek Khera, Ph.D. Khera Communications, Inc.
Internet: kh***@kciLink.com Rockville, MD [phone]
AIM: vivekkhera Y!: vivek_khera http://www.khera.org/~vivek/

---------------------------(end of broadcast)---------------------------
TIP 3: if posting/reading through Usenet, please send an appropriate
subscribe-nomail command to ma*******@postgresql.org so that your
message can get through to the mailing list cleanly

[Stephen]

Good news,

I partially fixed the problem on Linux 2.4. It appears the responsiveness
can be improved significantly by tuning the disk IO elevator in Linux using
"elvtune" in util-linux. The elevator in Linux is used to re-order
read/write requests to reduce disk seeks by ordering requests according to
disk sectors. Unfortunately, the elevator in kernel 2.4 is not very smart
(or flexible I should say depending on your needs) and can starve a
read/write request for a long time if not properly tuned.

See my older thread I wrote long time ago. Back then I didn't have too much
time to test different values especially for VACUUM:

http://groups.google.com/groups?q=li...+postgresql&hl
=en&lr=&ie=UTF-8&oe=UTF-8&selm=F92Znj0TrJIPT6nhdBf00021ae6%40hotmail.com&r nu
m=1

See also:

http://strasbourg.linuxfr.org/jl3/features-2.3-1.html
Below are the results using different elvtune values running repeatedly when
VACUUM'ing.

elvtune -r 2048 -w 8192 /dev/hdc (default Redhat 9):
================================================== ==

tsdb=# explain analyze select * from table1 where id =
'336139b47b7faf09fc4d4f03680a4ce5';
QUERY PLAN
----------------------------------------------------------------------------
--------------------------------------------------------
Index Scan using table1_pkey on table1 (cost=0.00..6.01 rows=1 width=346)
(actual time=2290.07..2290.10 rows=1 loops=1)
Index Cond: (id = '336139b47b7faf09fc4d4f03680a4ce5'::character varying)
Total runtime: 2290.22 msec
(3 rows)
elvtune -r 128 -w 8192 /dev/hdc:
================================

tsdb=# explain analyze select * from table1 where id =
'008ab286d725d2ea0b3269c89fc01ce2';
QUERY PLAN
----------------------------------------------------------------------------
------------------------------------------------------
Index Scan using table1_pkey on table1 (cost=0.00..6.01 rows=1 width=346)
(actual time=450.29..450.32 rows=1 loops=1)
Index Cond: (id = '008ab286d725d2ea0b3269c89fc01ce2'::character varying)
Total runtime: 450.46 msec
(3 rows)
elvtune -r 64 -w 8192 /dev/hdc:
===============================

tsdb=# explain analyze select * from table1 where id =
'0078997ac809877c1a0d1f76af753608';
QUERY PLAN
----------------------------------------------------------------------------
------------------------------------------------------
Index Scan using table1_pkey on table1 (cost=0.00..6.01 rows=1 width=346)
(actual time=206.00..206.01 rows=1 loops=1)
Index Cond: (id = '0078997ac809877c1a0d1f76af753608'::character varying)
Total runtime: 206.14 msec
(3 rows)
elvtune -r 32 -w 8192 /dev/hdc:
===============================

tsdb=# explain analyze select * from table1 where id =
'00c45490c9f24858c17d7dfb98c5def5';
QUERY PLAN
----------------------------------------------------------------------------
------------------------------------------------------
Index Scan using table1_pkey on table1 (cost=0.00..6.01 rows=1 width=346)
(actual time=210.61..210.62 rows=1 loops=1)
Index Cond: (id = '00c45490c9f24858c17d7dfb98c5def5'::character varying)
Total runtime: 210.75 msec
(3 rows)
elvtune -r 8 -w 8192 /dev/hdc:
==============================

tsdb=# explain analyze select * from table1 where id =
'006ea95ef6b9b8f0ddcb1f33c40190ec';
QUERY PLAN
----------------------------------------------------------------------------
------------------------------------------------------
Index Scan using table1_pkey on table1 (cost=0.00..6.01 rows=1 width=346)
(actual time=338.04..338.05 rows=1 loops=1)
Index Cond: (id = '006ea95ef6b9b8f0ddcb1f33c40190ec'::character varying)
Total runtime: 338.18 msec
(3 rows)
elvtune -r 1 -w 8192 /dev/hdc:
==============================

tsdb=# explain analyze select * from table1 where id =
'0025a7a9182d5456474a72f773433c01';
QUERY PLAN
----------------------------------------------------------------------------
------------------------------------------------------
Index Scan using table1_pkey on table1 (cost=0.00..6.01 rows=1 width=346)
(actual time=390.41..390.42 rows=1 loops=1)
Index Cond: (id = '0025a7a9182d5456474a72f773433c01'::character varying)
Total runtime: 390.55 msec
(3 rows)

elvtune -r 0 -w 8192 /dev/hdc:
==============================

tsdb=# explain analyze select * from table1 where id =
'0023783eda7e2a5f434e55a66c3a0a11';
QUERY PLAN
----------------------------------------------------------------------------
------------------------------------------------------
Index Scan using table1_pkey on table1 (cost=0.00..6.01 rows=1 width=346)
(actual time=246.29..246.32 rows=1 loops=1)
Index Cond: (id = '0023783eda7e2a5f434e55a66c3a0a11'::character varying)
Total runtime: 246.44 msec
(3 rows)

tsdb=# explain analyze select * from table1 where id =
'001eacb36161ac6a1f860bd391dce5c2';
QUERY PLAN
----------------------------------------------------------------------------
----------------------------------------------------
Index Scan using table1_pkey on table1 (cost=0.00..6.01 rows=1 width=346)
(actual time=14.11..14.12 rows=1 loops=1)
Index Cond: (id = '001eacb36161ac6a1f860bd391dce5c2'::character varying)
Total runtime: 14.25 msec
(3 rows)

Remarks:
========

In all cases, VACUUM completed in the same amount of time, process load
average is reduced slightly and vmstat IO is the same.

It turns out that as the -r read value in elvtune is made smaller, the more
responsive the read becomes up to a certain point when the disk needs to do
too many unoptimized seeks. A too small read value can cause the response
time to fluctuate more than a large value.

Understandably, the elvtune values are quite specific depending on what you
do. In my case, where the table is mostly read-only, "elevtune -r 64 -w 8192
/dev/hdc" works best. It should also depend on your expected load, purpose,
disk type, size and settings.

VACUUM is still disk IO intensive. Even with "elvtune -r 64 -w 8192
/dev/hdc" and VACUUM, the response time has gone down to 200 msec from 2200
msec (10 times factor), is still high compared to normal queries at 25 msec.
VACUUM needs to clamped down much more!

Regards,

Stephen

[Shridhar Daithankar]

Vivek Khera wrote:

>>"S" == Stephen <jl*****@xxxxxx.com> writes:

S> The system is actually idling when I ran the tests (load average: 0.01,
S> 0.02, 0.00). When VACUUM runs, load average increases quickly (load average:
S> 1.77, 0.60, 0.21) just by running psql on command line and issuing 2
S> queries. I've been running PostgreSQL for over 3 years now and IIRC VACUUM

Your CPU usage is irrelevent here. What's your *disk* utilization.
Does linux have systat and/or iostat to tell you how busy your disks
are?

It has a rather verbose vmstat which gives everything.

Shridhar
---------------------------(end of broadcast)---------------------------
TIP 4: Don't 'kill -9' the postmaster

[Greg Stark]

Vivek Khera <kh***@kcilink.com> writes:

Your CPU usage is irrelevent here. What's your *disk* utilization.
Does linux have systat and/or iostat to tell you how busy your disks
are?

I normally look at the bi/bo columns of "vmstat 1".

There does appear to be an iostat command, on debian in the sysstat package.
No idea how good it is.

--
greg
---------------------------(end of broadcast)---------------------------
TIP 5: Have you checked our extensive FAQ?

http://www.postgresql.org/docs/faqs/FAQ.html