Quantifying the cost of compression - Presented by Thomas Kejser at SQL Bits

Thomas Kejser

[email protected]

http://blog.kejser.org

@thomaskejser

Quantifying the cost of Compression

Grade of the Steel

• Reduce Cost of Storage

• Reduce the number of IOPS

• Is this relevant?

• Squeeze more memory into DRAM

• But at what cost to CPU?

Why Compress Data?

SQL Server Compression Overview

Page 3-4x

Row 1-2x

Column 5-10x

Backup 5-10x

• It depends on WHAT your data contains

• There is NO WAY to tell until you try

• Anyone who tells you otherwise are lying!

How much will my Data Compress?

• The Information Entropy of the data

• The block size of compression

• Row: 1 column in one row

• Page: 8K

• Backup: up to 4MB

• Column store: 1M rows

• The algorithm you use

• = time you have to compression

• How the algorithm fits the data

What does it Depend on?

• VERY interesting subject

• A lot can be done and said about this

• But it would be another full length prezo

• See my PASS Nordic presentation

Nothing about Column Stores

How does row Compression Work

• Variable Length Encoding (new row format)

• Special Handling of NULL and 0s

0 0 0 1

4-byte integer

1 SQL Server 2008+ (Compression enabled)

4 Bytes SQL Server 2005

1 Byte

Consider These Two Examples

CREATE TABLE Numbers ( foo BIGINT NOT NULL , bar BIGINT NOT NULL ) INSERT INTO Numbers WITH (TABLOCK) (foo, bar) SELECT (n2.n - 1) * 1000 + (n1.n - 1) , (n2.n - 1) * 1000 + (n1.n - 1) FROM fn_nums(1000) n1 CROSS JOIN fn_nums(1000) n2 CREATE UNIQUE CLUSTERED INDEX CIX ON Numbers(foo) WITH (DATA_COMPRESSION = ROW) EXEC sp_spaceused 'Numbers'

CREATE TABLE Numbers ( foo BIGINT NOT NULL , bar BIGINT NOT NULL ) INSERT INTO Numbers WITH (TABLOCK) (foo, bar) SELECT (n2.n - 1) * 1000 + (n1.n - 1) , (n2.n - 1) * 1000000000 + (n1.n - 1) FROM fn_nums(1000) n1 CROSS JOIN fn_nums(1000) n2 CREATE UNIQUE CLUSTERED INDEX CIX ON Numbers(foo) WITH (DATA_COMPRESSION = ROW) EXEC sp_spaceused 'Numbers'

11MB 13MB

• Combination of

• Prefix Compression – common prefix inside column

• Dictionary – Common values across all columns

• Pages are kept compressed in buffer pool

• So: It should be more expensive to access them even there?

How does Page Compression Work?

Page Compression- Column Prefix

Page Header

0x5B8D80

0x41AABB

0x9A4041

0x112233

0x5CAABB

Lambert 5000000 NULL

4

5 20x41

20x43

20x42

Page Header

0x5B8D80

0x41AABB

0x9A4041

0x112233

0x5CAABB

Lambert 5000000 NULL

4

5 20x41

20x43

20x42

Column prefix

Page Dictionary

1

1

0

0

Decimal 6000000 is HEX 0x5B8D80

Page Compression - Dictionary

How fast is it?

Our Reasonably Priced Server

• 2 Socket Xeon E3645 • 2 x 6 Cores

• 2.4Ghz

• NUMA enabled, HT off

• 12 GB RAM

• 1 ioDrive2 Duo • 2.4TB Flash

• 4K formatted

• 64K AUS

• 1 Stripe

• Power Save Off

• Win 2008R2

• SQL 2012

Image Source: DeviantArt

Test: Table Scan with Page Compress

• Use TPC-H

• Scale factor 10

• 10GB total dataset

• Apply PAGE compress

Compression Size Build Time CPU Load

None 6.5 GB 18 sec 100%

ROW 4.9 GB 16 sec 100%

PAGE 3.9 GB 38 sec 100%

Gzip –5 2.5 GB NA NA

NTFS Compress (Best: 4K AUS) 3.7 GB NA NA

Windows Zipping 2.3 GB NA NA

• Fast Scan Through Table

• Table resident in memory first

• Scan NONE and PAGE

Compressed tables are faster, right?

Compression Size Scan Time CPU Load

None - Memory 6.5 GB 4 sec 100%

PAGE - Memory 3.9 GB 8 sec 100%

PAGE

Ahh.. But Thomas: you didn’t do I/O

Compression Size Scan Time CPU Load

None - Memory 6.5 GB 4 sec 100%

PAGE - Memory 3.9 GB 8 sec 100%

None – I/O 6.5 GB 6 sec 100%

PAGE – I/O 3.9 GB 9 sec 100%

NONE

• Our old friend xperf

Where does the time go?

xperf –on base –stackwalk profile

Function NONE PAGE

MinMaxStep 15.9% 9.2%

IndexDataSetSession::GetNextRowValuesInternal 14.6% 17.0%

CEsExec::GeneralEval 11.8% 6.7%

CValXVarTable::GetDataX 8.8% 5.2%

CXVariant::CopyDeep 7.8% 4.6%

memcpy 6.5% .4%

CValXVarTableRow::SetDataX 6.0% 3.3%

GetDataFromXvar8 5.2% 2.9%

RowsetNewSS::FetchNextRow 2.7% 1.6%

ps_dl_sqlhilo 2.7% 1.5%

GetData 2.1% 1.2%

GetDataFromXvar 1.8% 1.0%

CTEsCompare<122;122>::BlCompareXcArgArg 1.5% .9%

CTEsCompare<58;58>::BlCompareXcArgArg 1.2% .6%

CTEsCompare<167;167>::BlCompareXcArgArg 1.1% .7%

CTEsCompare<56;56>::BlCompareXcArgArg 1.1% .6%

SetMultData 1.1% .6%

CQScanTableScanNew::GetRow 1.0% .6%

CQScanStreamAggregateNew::GetRowHelper 1.0% .6%

CTEsCompare<52;52>::BlCompareXcArgArg .8% .7%

ScalarCompression::AddPadding .7%

PageComprMgr::DecompressColumn 4.7%

DataAccessWrapper::DecompressColumnValue 9.9%

CDRecord::LocateColumnInternal 14.4%

AnchorRecordCache::LocateColumn 2.3%

DataAccessWrapper::StoreColumnValue 4.3%

Additional Runtime of GetNextRowValuesInternal 2.4%

Total 38.7%

• Sample from LINEITEM

• Force loop join with index seeks

• Do 1.4M seeks

Test: Singleton Row Fetch

Singleton seeks – Cost of compression

Compression Seek (1.4M seeks) CPU Load

None - Memory 13 sec 100% one core

PAGE - Memory 24 sec 100% one core

None – I/O 21 sec 100% one core

PAGE – I/O 32 sec 100% one core

Function % Weight

CDRecord::LocateColumnInternal 0.82%

DataAccessWrapper::DecompressColumnValue 0.47%

SearchInfo::CompareCompressedColumn 0.28%

PageComprMgr::DecompressColumn 0.24%

AnchorRecordCache::LocateColumn 0.18%

ScalarCompression::AddPadding 0.04%

ScalarCompression::Compare 0.11%

Additional Runtime of GetNextRowValuesInternal 0.14%

Total Compression 2.28%

Total CPU (single core) 8.33%

Compression % 27.00%

xperf –on base –stackwalk profile

Test: Updates of pages

Compression Update 1.4M CPU Load

None - Memory 13 sec 100% one core

PAGE - Memory 54 sec 100% one core

None – I/O 17 sec 100% one core

PAGE – I/O 59 sec 100% one core

L_QUANTITY is NOT NULL i.e. in place UPDATE

Function CPU %

qsort 0.86

CDRecord::Resize 0.84

CDRecord::LocateColumnInternal 0.36

perror 0.36

Page::CompactPage 0.36

ObjectMetadata::`scalar deleting destructor' 0.27

SearchInfo::CompareCompressedColumn 0.24

CDRecord::InitVariable 0.19

CDRecord::LocateColumnWithCookie 0.18

memcmp 0.16

PageDictionary::ValueToSymbol 0.16

Record::DecompressRec 0.14

PageComprMgr::DecompressColumn 0.14

CDRecord::InitFixedFromOld 0.1

SOS_MemoryManager::GetAddressInfo64 0.08

AnchorRecordCache::LocateColumn 0.08

CDRecord::GetDataForAllColumns 0.08

ScalarCompression::Compare 0.07

PageComprMgr::CompressColumn 0.07

Record::CreatePageCompressedRecNoCheck 0.06

memset 0.05

PageComprMgr::ExpandPrefix 0.04

PageRef::ModifyColumnsInternal 0.04

Page::ModifyColumns 0.03

DataAccessWrapper::ProcessAndCompressBuffer 0.03

SingleColAccessor::LocateColumn 0.03

CDRecord::BuildLongRegionBulk 0.02

ChecksumSectors 0.02

Page::MCILinearRegress 0.02

DataAccessWrapper::DecompressColumnValue 0.02

SOS_MemoryManager::GetAddressInfo 0.02

CDRecord::FindDiff 0.02

AnchorRecordCache::Init 0.02

PageComprMgr::CombinePrefix 0.01

Total 5.17

UPDATE Compression burners

Out of 8.55 … Approx: 60%

• I am going to do a demo!

• Lets see xperf in action

And now, for something unusual!

Benchmark Test: HammerDb

Compression TPM CPU Load

NONE Compress 16.7 M 70% all cores

PAGE Compress 14.5 M 70% all cores

Benchmark Simulating TPC type workloads It is NOT TPC, but it helps you set up something similar Below are “TPC-C like” workload

What about locks and contention?

Xevent Trace Lock Acquire/Release

How long are locks held?

0

100

200

300

400

500

600

PAGE NONE

CPU KCycles

Lock Held Cycle Count

Avg

StdDev

&