Compress and Decompress in SQL Server 2016

Question

Compress and Decompress in SQL Server 2016

David Fundakowski

Ten Centuries

Points: 1328
More actions
September 30, 2018 at 10:11 pm

#408577

Comments posted to this topic are about the item Compress and Decompress in SQL Server 2016

Viewing 10 posts - 1 through 9 (of 9 total)

You must be logged in to reply to this topic. Login to reply

BrainDonor SSCoach Points: 19254 More actions · Answer 1

Nice article but testing of this function is a must. When looking into these commands a few months ago, although the general trend was a reduction in data length I did have some data that actually increased in size when fed through COMPRESS().

Steve Hall
Linkedin
Blog Site

Jonathan AC Roberts SSCoach Points: 18361 More actions · Answer 2

BrainDonor - Monday, October 1, 2018 3:11 AM
Nice article but testing of this function is a must. When looking into these commands a few months ago, although the general trend was a reduction in data length I did have some data that actually increased in size when fed through COMPRESS().

I guess if you compress an already compressed value it will make it bigger.

BrainDonor SSCoach Points: 19254 More actions · Answer 3

Jonathan AC Roberts - Monday, October 1, 2018 4:52 AM
I guess if you compress an already compressed value it will make it bigger.

That is one way but this was just imported text - Compress and Decompress Basic Examples.

Steve Hall
Linkedin
Blog Site

deroby SSC-Addicted Points: 416 More actions · Answer 4

Interesting article.
As a side note, if I remember correctly, MSSQL will 'ignore' out-of-page data (LOB_DATA) when it comes to PAGE compression; I guess COMPRESS() and DECOMPRESS() are the 'workarounds' for this? I do hope they'll add compression to those sooner than later though.
Anyway, as such I think it would still be interesting to see a figure 3-3 that shows the effect of table-compression on the test-data.

André Lozeau Mr or Mrs. 500 Points: 546 More actions · Answer 5

quite interesting.

Like Deroby, I'd like to see comparison between COMPRESS() and compression at the table level

Solomon Rutzky SSCoach Points: 16430 More actions · Answer 6

deroby - Monday, October 1, 2018 6:07 AM
As a side note, if I remember correctly, MSSQL will 'ignore' out-of-page data (LOB_DATA) when it comes to PAGE compression; I guess COMPRESS() and DECOMPRESS() are the 'workarounds' for this? I do hope they'll add compression to those sooner than later though.
Anyway, as such I think it would still be interesting to see a figure 3-3 that shows the effect of table-compression on the test-data.

My testing shows that:

both ROW and PAGE compression definitely do not work with off-row data (LOB and Row-Overflow)
ROW compression does not work at all with MAX types, even if the data is stored in-row
PAGE compression can work with in-row values of MAX types, but not over a certain size (not sure why, though)

You can try for yourself using the queries below. In the second set, even though the value is only 6000 bytes, and hence in-row, PAGE compression does nothing really. But, truncate the table, change the replicate amount to be 4000, and then PAGE compression works wonders.

USE [tempdb];

-- DROP TABLE dbo.CompressionTest; CREATE TABLE dbo.CompressionTest ( [ID] INT NOT NULL IDENTITY(1, 1) PRIMARY KEY, [SomeValue] VARCHAR(MAX) NOT NULL );

INSERT INTO dbo.CompressionTest ([SomeValue]) SELECT REPLICATE(CONVERT(VARCHAR(MAX), 'a'), 9000) -- off-row value FROM master.sys.all_columns;

ALTER TABLE dbo.CompressionTest REBUILD WITH (FILLFACTOR = 100, DATA_COMPRESSION = NONE);

EXEC sys.sp_spaceused N'dbo.CompressionTest'; /* name rows reserved data index_size unused dbo.CompressionTest 10539 97376 KB 97024 KB 72 KB 280 KB */

ALTER TABLE dbo.CompressionTest REBUILD WITH (DATA_COMPRESSION = ROW);

EXEC sys.sp_spaceused N'dbo.CompressionTest'; /* name rows reserved data index_size unused dbo.CompressionTest 10539 97120 KB 96960 KB 72 KB 88 KB */

ALTER TABLE dbo.CompressionTest REBUILD WITH (DATA_COMPRESSION = PAGE);

EXEC tempdb.sys.sp_spaceused N'dbo.CompressionTest'; /* name rows reserved data index_size unused dbo.CompressionTest 10539 97120 KB 96960 KB 72 KB 88 KB */

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

TRUNCATE TABLE dbo.CompressionTest;

INSERT INTO dbo.CompressionTest ([SomeValue]) SELECT REPLICATE(CONVERT(VARCHAR(MAX), 'a'), 6000) -- in-row value FROM master.sys.all_columns;

ALTER TABLE dbo.CompressionTest REBUILD WITH (FILLFACTOR = 100, DATA_COMPRESSION = NONE);

EXEC sys.sp_spaceused N'dbo.CompressionTest'; /* name rows reserved data index_size unused dbo.CompressionTest 10539 85128 KB 84328 KB 232 KB 568 KB */

ALTER TABLE dbo.CompressionTest REBUILD WITH (DATA_COMPRESSION = ROW);

EXEC sys.sp_spaceused N'dbo.CompressionTest'; /* name rows reserved data index_size unused dbo.CompressionTest 10539 85128 KB 84328 KB 232 KB 568 KB */

ALTER TABLE dbo.CompressionTest REBUILD WITH (DATA_COMPRESSION = PAGE);

EXEC sys.sp_spaceused N'dbo.CompressionTest'; /* name rows reserved data index_size unused dbo.CompressionTest 10539 85128 KB 84328 KB 232 KB 568 KB */

SELECT * FROM sys.dm_db_database_page_allocations(DB_ID(), OBJECT_ID(N'dbo.CompressionTest'), 1, 1, 'DETAILED');

Converting the table to a Clustered Columnstore Index sometimes helps with large types, even off row values.

So yes, I think COMPRESS / DECOMPRESS is definitely an alternative for this scenario (that or Clustered Columnstore, depending on the scenario).

On a related note: I brought up this very point (these two work-arounds) in my post about the new UTF-8 Collations in SQL Server 2019:
Native UTF-8 Support in SQL Server 2019: Savior or False Prophet?

Take care, Solomon...

SQL# — https://SQLsharp.com/ ( SQLCLR library ofover 340 Functions and Procedures)
Sql Quantum Lift — https://SqlQuantumLift.com/ ( company )
Sql Quantum Leap — https://SqlQuantumLeap.com/ ( blog )
Info sites — Collations • Module Signing • SQLCLR

Solomon Rutzky SSCoach Points: 16430 More actions · Answer 7

BrainDonor - Monday, October 1, 2018 5:00 AM
Jonathan AC Roberts - Monday, October 1, 2018 4:52 AM
I guess if you compress an already compressed value it will make it bigger.
That is one way but this was just imported text - Compress and Decompress Basic Examples.

There is a certain minimum size of the resulting "compressed" value (I guess due to "overhead"), so input values under a certain size (depending on compression algorithm, I believe) will actually have negative gains from the compression.

Also, compressing relies upon reducing patterns. If the input value has no discernible patterns (such as a value that has already been compressed), it won't compress plus you still get that overhead.

Take care, Solomon...

SQL# — https://SQLsharp.com/ ( SQLCLR library ofover 340 Functions and Procedures)
Sql Quantum Lift — https://SqlQuantumLift.com/ ( company )
Sql Quantum Leap — https://SqlQuantumLeap.com/ ( blog )
Info sites — Collations • Module Signing • SQLCLR

frederico_fonseca SSCoach Points: 16167 More actions · Answer 8

compress definitely works a lot better for particular data - As an example one of my tables (700Million rows, 900GB + on standard format and page compression), when I compress the main blob on it , with a mix of text data and pdf/images it goes down to 400GB. not more because of PDF data which is already a zip format.
This is using a gzipstream - which includes a header of 128 bytes. If it was deflatestream it would not include this header.

deroby SSC-Addicted Points: 416 More actions · Answer 9

@Solomon Rutzky:
thanks for working this out, my tests had shown similar results; I simply wondered what it would be like in this particular tests as it "fits" the subject.
(and I was slightly hoping someone would say that version xyz now includes LOB_DATA when applying PAGE compression =)

Kind regards,
Roby