106
VERY LARGE DATABASES ADM INIS TRATION @murilocmiranda http://www.sql.pt/ [email protected]

VLDB Administration Strategies

Embed Size (px)

DESCRIPTION

Presentation done by SQL Port event.

Citation preview

Page 1: VLDB Administration Strategies

VERY L

ARGE DAT

ABASES

AD

MI N

I ST

RA

TI O

N

@murilocmirandahttp://www.sql.pt/[email protected]

Page 2: VLDB Administration Strategies

AGENDA

Page 3: VLDB Administration Strategies

AGENDA

1. What is a VLDB?2. Typical Troubles3. OS Config4. Instance Config5. DB Config6. Maintenance

Page 4: VLDB Administration Strategies

VLDB??

Page 5: VLDB Administration Strategies

VLDB??

There’s no official definition.

Page 6: VLDB Administration Strategies

VLDB??

There’s no official definition.

Typically occupying TB range.

Page 7: VLDB Administration Strategies

VLDB??

There’s no official definition.

Typically occupying TB range.

Billions of rows.

Page 8: VLDB Administration Strategies

VLDB??

There’s no official definition.

Typically occupying TB range.

Billions of rows.

Typically: OLAP or OLTP with large amount of users.

Page 9: VLDB Administration Strategies

VLDB??

A very large database, or VLDB, is a database that contains an extremely high number of tuples (database rows), or occupies an extremely large physical filesystem storage space. The most common definition of VLDB is a database that occupies more than 1 terabyte or contains several billion rows, although naturally this definition changes over time.

Wikipedia…

Page 10: VLDB Administration Strategies

SQL VS. VLDB

Page 11: VLDB Administration Strategies

SQL VS. VLDB

Maximum database size

Page 12: VLDB Administration Strategies

SQL VS. VLDB

Maximum database size

524,272 TB

Page 13: VLDB Administration Strategies

SQL VS. VLDB

Maximum data file size16 TBMaximum log file size 2 TB

A limit of 32.767 files which can be distributed between 32.767 filegroups.

Page 14: VLDB Administration Strategies

TYPICAL TROUBLES

Page 15: VLDB Administration Strategies

TYPICAL TROUBLES

Maintenance

Page 16: VLDB Administration Strategies

TYPICAL TROUBLES

Maintenance

Backups

Page 17: VLDB Administration Strategies

TYPICAL TROUBLES

Maintenance

Backups

Indexes

Page 18: VLDB Administration Strategies

TYPICAL TROUBLES

Maintenance

Backups

Indexes

Statistics

Page 19: VLDB Administration Strategies

TYPICAL TROUBLES

Maintenance

Backups

Indexes

Statistics

Disaster Recovery

Page 20: VLDB Administration Strategies

TYPICAL TROUBLES

Maintenance

Backups

Indexes

Statistics

Disaster Recovery

Performance

Page 21: VLDB Administration Strategies

OS CONFIG

Page 22: VLDB Administration Strategies

OS CONFIG

Perform Volume Maintenance

Page 23: VLDB Administration Strategies

OS CONFIG

Turning on Instant Initialization to speed up data file growth and restores.

Page 24: VLDB Administration Strategies

OS CONFIG

Storage Layout

Page 25: VLDB Administration Strategies

OS CONFIG

Plan an efficient storage layout.

Page 26: VLDB Administration Strategies

OS CONFIG

Plan an efficient storage layout.

Normally, the more spread, the more effective.

Page 27: VLDB Administration Strategies

OS CONFIG

Plan an efficient storage layout.

Normally, the more spread, the more effective.

Suggestion:

SQL BINSQL DATASQL IDXSQL LOGS SQL TMP

Page 28: VLDB Administration Strategies

OS CONFIG

Mountpoints

Page 29: VLDB Administration Strategies

OS CONFIG

Mountpoints could be a good strategy.

Page 30: VLDB Administration Strategies

OS CONFIG

Mountpoints could be a good strategy.

Mountpoints are persistent directories that point to disk volumes.

Page 31: VLDB Administration Strategies

OS CONFIG

Pros:

• Scalable.• Save drive letters (limited to 26).• Easy to add.• No need to restart SQL Server.

Page 32: VLDB Administration Strategies

OS CONFIG

Cons:

• Looks like a simple folder.• Need a different approach to monitor.

Page 33: VLDB Administration Strategies

OS CONFIG

So, if you don’t know the server….

Page 34: VLDB Administration Strategies

OS CONFIG

Partition Alignment

Page 35: VLDB Administration Strategies

OS CONFIG

Setting the partition offset properly can improve up to 30% the performance.

Page 36: VLDB Administration Strategies

OS CONFIG

Setting the partition offset properly can improve up to 30% the performance.Partition alignment increases

throughput (bytes/sec) and reduce disk queues.

Page 37: VLDB Administration Strategies

OS CONFIG

Setting the partition offset properly can improve up to 30% the performance.Partition alignment increases

throughput (bytes/sec) and reduce disk queues.

A partition that is track misaligned will occasionally cause 2 I/O operations instead of one.

Page 38: VLDB Administration Strategies

OS CONFIG

Unless performed at the time of partition creation, the default alignment offset (31,5 Kb) will result in unaligned partitions on versions of Windows up to and including Windows Server 2003.

Page 39: VLDB Administration Strategies

OS CONFIG

This offset is associated with hidden sectors,which basically store partition information.

Page 40: VLDB Administration Strategies

OS CONFIG

This offset is associated with hidden sectors,which basically store partition information.Considering that:

- Each disk sector has 512 bytes.- Win. 2003 has 63 hidden sectors.

Page 41: VLDB Administration Strategies

OS CONFIG

This offset is associated with hidden sectors,which basically store partition information.Considering that:

- Each disk sector has 512 bytes.- Win. 2003 has 63 hidden sectors.

512 * 63 = 31,5 Kb

Page 42: VLDB Administration Strategies

OS CONFIG

Example:

Stripe Unit Size: 64Kb*Allocation Unit Size: 64Kb

* Defined by storage team.

Optimal values

Page 43: VLDB Administration Strategies

OS CONFIG

Example:

Stripe Unit Size: 64Kb*Allocation Unit Size: 64Kb

* Defined by storage team.

Optimal values

Stripe Size

Data (Alloc. Unit Size)

Page 44: VLDB Administration Strategies

OS CONFIG

Optimal solution:

Stripe Size

Data (Alloc. Unit Size)

Page 45: VLDB Administration Strategies

OS CONFIG

Best Practice:

- Set an offset of 1024 Kb.- This value works for mostly disks out there.

- Allocation Unit Size = Stripe Unit Size.

The rule: Offset / Allocation unit = INTEGER Eg: 1024/64=16

Page 46: VLDB Administration Strategies

Some I/O subsystem vendors intercepting what Windows is trying to do and are still creating partitions with the incorrect offset – Even for Windows 2008+.

WARNIG

ALWAYS check!

Page 47: VLDB Administration Strategies

OS CONFIG

Anti-Virus in servers… is really a need?

Page 48: VLDB Administration Strategies

OS CONFIG

• Cost money to license.• Maintenance costs.• Can cause problems in Prod.• Can’t protect to zero-day exploits.

Page 49: VLDB Administration Strategies

OS CONFIG

What can we do instead?

Page 50: VLDB Administration Strategies

OS CONFIG

• Keep the servers patched.• Configure the firewall properly.• Restrict server’s access.• You can install AV… in workstations!

Page 51: VLDB Administration Strategies

OS CONFIG

What’s the big problem for SQL Server?

Page 52: VLDB Administration Strategies

OS CONFIG

• One more app fighting for resources.• SQL Server files can be locked.

Page 53: VLDB Administration Strategies

OS CONFIG

How can AV and SQL Server live together?

Page 54: VLDB Administration Strategies

OS CONFIG

Add exceptions!

Page 55: VLDB Administration Strategies

OS CONFIG

Basically the AV should ignore:

• SQL Server data and log files (.mdf, .ndf and .ldf).• Backup files (.bak and .trn).• Full-text Catalog files.• Trace files (.trc).• ERRORLOG files.• SQL Server binaries folder.• Filestream folder.

More on: http://support.microsoft.com/kb/309422

Page 56: VLDB Administration Strategies

INSTANCE CONFIG

Page 57: VLDB Administration Strategies

INSTANCE CONFIG

Memory

Page 58: VLDB Administration Strategies

INSTANCE CONFIG

Memory

This is a very open subject.

Page 59: VLDB Administration Strategies

INSTANCE CONFIG

Memory

This is a very open subject.

There are lots of discussions about that…

Page 60: VLDB Administration Strategies

INSTANCE CONFIG

Memory

This is a very open subject.

There are lots of discussions about that…

There’s no perfect formula, because the correct awnser is….

Page 61: VLDB Administration Strategies

INSTANCE CONFIG

Memory

This is a very open subject.

There are lots of discussions about that…

There’s no perfect formula, because the correct answer is….

… it depends !!

Page 62: VLDB Administration Strategies

INSTANCE CONFIG

Memory

Baseline: 1 GB for the OS

Up to 16 GB available• 1 GB for each 4 GB

More than 16 GB• 1 GB for every 8 GB

An efficient general rule…

Page 63: VLDB Administration Strategies

INSTANCE CONFIG

Memory

This is for 64 bit servers…

For 32 bit, here is a good article to follow:http://www.eraofdata.com/understanding-and-configuring-sql-servers-memory-settings/

Page 64: VLDB Administration Strategies

INSTANCE CONFIG

TempDB

Page 65: VLDB Administration Strategies

INSTANCE CONFIG

TempDB

Two common behaviors:

Page 66: VLDB Administration Strategies

INSTANCE CONFIG

TempDB

Two common behaviors:• Ignore.• Overvalue.

Page 67: VLDB Administration Strategies

INSTANCE CONFIG

TempDB

As per Brent Ozar:

“TempDb is the SQL’s public toilet”

Page 68: VLDB Administration Strategies

INSTANCE CONFIG

TempDB

And this is true!

Page 69: VLDB Administration Strategies

INSTANCE CONFIG

TempDB

Page 70: VLDB Administration Strategies

INSTANCE CONFIG

TempDB

There’s a myth:• tempdb should always have one data file per processor core.

Page 71: VLDB Administration Strategies

INSTANCE CONFIG

TempDB

There’s a myth:• tempdb should always have one data file per processor core.

Again….

Page 72: VLDB Administration Strategies

INSTANCE CONFIG

TempDB

There’s a myth:• tempdb should always have one data file per processor core.

Again…. It depends!

Page 73: VLDB Administration Strategies

INSTANCE CONFIG

TempDB

Execute large operations, like a sort or store a huge temporary table,may be slowed down because of the round-robin operation.

The more files, the more costly.

Page 74: VLDB Administration Strategies

INSTANCE CONFIG

TempDB

Common wait types on TempDB:

• PAGELATCH_*: Contention for In-memory allocation bitmaps. • PAGEIOLATCH_*: Contention at the I/O subsystem level.

Page 75: VLDB Administration Strategies

INSTANCE CONFIG

TempDB

How many tempdb data files should we have?

Page 76: VLDB Administration Strategies

INSTANCE CONFIG

TempDB

How many tempdb data files should we have?A recommended approach is:• Up to 8 cores:

Number of files = Number of cores.

• More than 8 cores: 1. Add 8 files.2. Monitor PAGELATCH_*.3. Add 4 more files at a time, if necessary.

Page 77: VLDB Administration Strategies

INSTANCE CONFIG

TempDB

Other TempDB best practices:• Isolate the TempDB in a different storage system.• Depending of the load, you might need to separate

LDF and M(N)DF. • Use a fast drive (SSD :).• Set an initial size, equally to all the files.

• Set the auto-growth accordingly.• If you have a heavy operation using constantly the

TempDB, consider create a staging table into your own database.

Page 78: VLDB Administration Strategies

INSTANCE CONFIG

TempDB

From SQL Server 2012, local disk TempDB in SQL Server cluster.

Page 79: VLDB Administration Strategies

INSTANCE CONFIG

TempDB

From SQL Server 2012, local disk TempDB in SQL Server cluster.

• More flexibility.• Use PCIe bus instead of HBA, and have more

throughput.• Data and Log are in SAN, TempDB locally: Avoid

congestion or contention on a shared storage network or array.

Page 80: VLDB Administration Strategies

DB CONFIG

Page 81: VLDB Administration Strategies

DB CONFIG

• Don’t rely on auto-grow.• You can manage file growth and control the

free disk space and avoids performance problems.

Page 82: VLDB Administration Strategies

DB CONFIG

• Don’t rely on auto-grow.• You can manage file growth and control the

free disk space and avoids performance problems.

• Have page checksums turned on.• To detect damaged pages.

Page 83: VLDB Administration Strategies

DB CONFIG

• Don’t rely on auto-grow.• You can manage file growth and control the

free disk space and avoids performance problems.

• Have page checksums turned on.• To detect damaged pages.

• Make sure auto-stats update is turned on.• For OLTP consider turning auto-stats update

off only for heavily updated tables, and schedule a job that periodically updates the statistics for those tables.

Page 84: VLDB Administration Strategies

DB CONFIG

Page 85: VLDB Administration Strategies

DB CONFIG

• Make sure you’re managing the transaction log correctly:• Full recovery requires log backups.• No advantage in have multiple log files.• Control the file growth or this could

cause VLF fragmentation.• Performance issues.• Slow backup time.

• Don’t set the log file growth size to a multiple of 4 in older SQL Server versions.• http://connect.microsoft.com/SQLServer/feedback/details/481594/log-growth-

not-working-properly-with-specific-growth-sizes-vlfs-also-not-created-appropriately

Page 86: VLDB Administration Strategies

MAINTENACE

Page 87: VLDB Administration Strategies

MAINTENANCE

Few questions…

Page 88: VLDB Administration Strategies

MAINTENANCE

Is data-loss acceptable?

What about the recovery time?

How to meet your SLAs dealing with a TB database?

Are you able to UPDATE STATS, do INDEX MAINTENANCE and run a INTEGRITY CHECK in time and WITHOUT PROBLEMS?

Page 89: VLDB Administration Strategies

MAINTENANCE

DISASTER RECOVERY

Page 90: VLDB Administration Strategies

MAINTENANCE

First of all, think in a Disaster Recovery plan!

SQL Server is not Oracle, we have “free” included options:• Log Shipping (HA and DR)• Database Mirroring (HA and DR)

• DB Snapshot advantage

• Replication (HA, DR and LB)• AlwaysOn (HA, DR and LB)

• We can still be safe with a storage level replication.

Page 91: VLDB Administration Strategies

MAINTENANCE

PartitionCompressClean

Page 92: VLDB Administration Strategies

MAINTENANCE

Partition, Compress and Clean

Using the partitioning feature you can devise the maintenance.

Page 93: VLDB Administration Strategies

MAINTENANCE

Partition, Compress and Clean

Using the partitioning feature you can devise the maintenance.

• You can use the DBCC CHECKFILEGROUP command.• DBCC CHECKFILEGROUP and DBCC CHECKDB are. The main

difference is that DBCC CHECKFILEGROUP is limited to the single specified filegroup and required tables.

Page 94: VLDB Administration Strategies

MAINTENANCE

Partition, Compress and Clean

Using the partitioning feature you can devise the maintenance.

• Devising a filegroup architecture allows piecemeal restores with low TTR • Online piecemeal restore:

• After the PRIMARY FG restore the DB can be online.• The tables will come available while each FG is restored.

• Design the database accordingly:• Keep the necessary into the PRIMARY FG.

• Configuration tables, indispensable data, etc…• Think in the consistency: keep related tables in the same

FG.

Page 95: VLDB Administration Strategies

MAINTENANCE

Partition, Compress and Clean

Compress backups Vs. Compress Data

• Backup compression:• More CPU usage to backup/restore (avg ~20%).• Less time to backup/restore (avg ~40%).• Good compression ratio.

• SELECT backup_size/compressed_backup_size FROM msdb..backupset;

• A backup set will not be able to contain both compressed and uncompressed backups.

• No advantage with TDE enabled.

Page 96: VLDB Administration Strategies

MAINTENANCE

Partition, Compress and Clean

Compress backups Vs. Compress Data

• Data compression (ROW and PAGE):• TDE and Data Compression play together!• Backup and Data Compression can coexist!

Page 97: VLDB Administration Strategies

MAINTENANCE

Partition, Compress and Clean

Purge and Archive the data

• Purging data:• If data is needed no more…• Save storage.• Faster backups.• Improves the performance.

Page 98: VLDB Administration Strategies

MAINTENANCE

Partition, Compress and Clean

Purge and Archive the data

• Archiving data:• If data is still needed…• Isolate in a different FG.• Set as Read-Only: Avoids locking.

• For faster scans: 100% fill factor.• Update statistics with FULLSCAN.

• You can adapt the backup strategy.• You can adapt the backup strategy using Partial

Backups.• This allows you to exclude read-only filegroups.

Page 99: VLDB Administration Strategies

MAINTENANCE

More about DBCC CHECKDB• CHECKDB takes time and uses resources.

• Run a DBCC CHECKDB using the WITH PHYSICAL_ONLY option.• Limits the checking to the integrity of the physical structure of

the page and record headers and the allocation consistency of the database.

• Faster, but a full CHECKDB is required periodically.

Page 100: VLDB Administration Strategies

MAINTENANCE

More about DBCC CHECKDB• We can divide up the consistency checking over several

days, Paul Randal’s prescription is:• Divide tables in two buckets (bigger ones and the rest)

• On Sunday:• Run a DBCC CHECKALLOC• Run a DBCC CHECKCATALOG• Run a DBCC CHECKTABLE on each table in the first bucket

• On Monday, Tuesday, Wednesday:• Run a DBCC CHECKTABLE on each table in the 2nd, 3rd, 4th

buckets, respectively• On Thursday:

• Run a DBCC CHECKALLOC• Run a DBCC CHECKTABLE on each table in the 5th bucket

• On Friday and Saturday:• Run a DBCC CHECKTABLE on each table in the 6th and 7th

buckets, respectively

More on: http://www.sqlskills.com/blogs/paul/checkdb-from-every-angle-consistency-checking-options-for-a-vldb/

Page 101: VLDB Administration Strategies

MAINTENANCE

More about BACKUPS• Besides doing PARTIAL BACKUPS we have more options…• A MULTISTREAM BACKUP is an option to run faster:

DB

File 1

File 2

File 3

E:

G:

F:

Page 102: VLDB Administration Strategies

MAINTENANCE

More about BACKUPS• To make sure it will be well stored, we can use a MIRROR.

DB

File 1

File 2

File 3

E:

G:

F:

File 1

File 2

File 3

Page 103: VLDB Administration Strategies

MAINTENANCE

More about BACKUPS• If storing to the network:

• Use a separate network card to avoid network congestion.

• Don’t forget about T-LOG backups!• Create a good backup strategy.

• Verify the backups periodically.

Page 104: VLDB Administration Strategies

MAINTENANCE

INDEXES MAINTENANCE• Only rebuild/defrag indexes that are really fragmented

(avoid unnecessary work in short maintenance windows)

• If you defrag instead of rebuild, make sure you manually update stats.

• Be wary of doing large index maintenance jobs if you use log shipping or DBM• They contribute to large log backups• Index rebuilds are always full-logged when DBM is

present

Page 105: VLDB Administration Strategies

QUESTIONS?

Page 106: VLDB Administration Strategies

OBRIGADO!@murilocmirandahttp://www.sql.pt/[email protected]