SQL Server: the maximum number of rows in table

I develop software that stores a lot of data in one of its database tables (SQL Server version 8, 9 or 10). Let's say, about 100,000 records are inserted into that table per day. This is about 36 million records per year. For fear that I would lose on performance, I decided to create a new table everyday (a table with current date in its name) to lower the number of records per table.

Could you please tell me, whether it was a good idea? Is there a record limit for SQL server tables? Or do you know how many records (more or less) can be stored in a table before performance is lowered significantly?

234867 次浏览
小开

I don't know MSSQL specifically, but 36 million rows is not large to an enterprise database - working with mainframe databases, 100,000 rows sounds like a configuration table to me :-).

While I'm not a big fan of some of Microsoft's software, this isn't Access we're talking about here: I assume they can handle pretty substantial database sizes with their enterprise DBMS.

I suspect days may have been too fine a resolution to divide it up, if indeed it needs dividing at all.

小开

I do not know of a row limit, but I know tables with more than 170 million rows. You may speed it up using partitioned tables (2005+) or views that connect multiple tables.

小开
最佳答案

It's hard to give a generic answer to this. It really depends on number of factors:

  • what size your row is
  • what kind of data you store (strings, blobs, numbers)
  • what do you do with your data (just keep it as archive, query it regularly)
  • do you have indexes on your table - how many
  • what's your server specs

etc.

As answered elsewhere here, 100,000 a day and thus per table is overkill - I'd suggest monthly or weekly perhaps even quarterly. The more tables you have the bigger maintenance/query nightmare it will become.

小开

You can populate the table until you have enough disk space.

For better performance you can try migration to SQL Server 2005 and then partition the table and put parts on different disks (if you have RAID configuration that could really help you). Partitioning is possible only in enterprise version of SQL Server 2005. You can look at the partitioning example at this link.

Also you can try to create views for most used data portion, that is also one of the solutions.

小开

It depends, but I would say it is better to keep everything in one table for that sake of simplicity.

100,000 rows a day is not really that much of an enormous amount. (Depending on your server hardware). I have personally seen MSSQL handle up to 100M rows in a single table without any problems. As long as your keep your indexes in order it should be all good. The key is to have heaps of memory so that indexes don't have to be swapped out to disk.

On the other hand, it depends on how you are using the data, if you need to make lots of query's, and its unlikely data will be needed that spans multiple days (so you won't need to join the tables) it will be faster to separate out it out into multiple tables. This is often used in applications such as industrial process control where you might be reading the value on say 50,000 instruments every 10 seconds. In this case speed is extremely important, but simplicity is not.

小开 
SELECT Top 1 sysobjects.[name], max(sysindexes.[rows]) AS TableRows,
CAST(
CASE max(sysindexes.[rows])
WHEN 0 THEN -0
ELSE LOG10(max(sysindexes.[rows]))
END
AS NUMERIC(5,2))
AS L10_TableRows
FROM sysindexes INNER JOIN sysobjects ON sysindexes.[id] = sysobjects.[id]
WHERE sysobjects.xtype = 'U'
GROUP BY sysobjects.[name]
ORDER BY max(rows) DESC
小开

These are some of the Maximum Capacity Specifications for SQL Server 2008 R2

  • Database size: 524,272 terabytes
  • Databases per instance of SQL Server: 32,767
  • Filegroups per database: 32,767
  • Files per database: 32,767
  • File size (data): 16 terabytes
  • File size (log): 2 terabytes
  • Rows per table: Limited by available storage
  • Tables per database: Limited by number of objects in a database
小开

We have tables in SQL Server 2005 and 2008 with over 1 Billion rows in it (30 million added daily). I can't imagine going down the rats nest of splitting that out into a new table each day.

Much cheaper to add the appropriate disk space (which you need anyway) and RAM.

小开

We overflowed an integer primary key once (which is ~2.4 billion rows) on a table. If there's a row limit, you're not likely to ever hit it at a mere 36 million rows per year.

小开

Largest table I've encountered on SQL Server 8 on Windows2003 was 799 million with 5 columns. But whether or not it's good will is to be measured against the SLA and usage case - e.g. load 50-100,000,000 records and see if it still works.

小开

Partition the table monthly.That is the best way to handle tables with large daily influx ,be it oracle or MSSQL.

小开

I have a three column table with just over 6 Billion rows in SQL Server 2008 R2.

We query it every day to create minute-by-minute system analysis charts for our customers. I have not noticed any database performance hits (though the fact that it grows ~1 GB every day does make managing backups a bit more involved than I would like).

Update July 2016

Row count

We made it to ~24.5 billion rows before backups became large enough for us to decide to truncate records older than two years (~700 GB stored in multiple backups, including on expensive tapes). It's worth noting that performance was not a significant motivator in this decision (i.e., it was still working great).

For anyone who finds themselves trying to delete 20 billion rows from SQL Server, I highly recommend this article. Relevant code in case the link dies (read the article for a full explanation):

ALTER DATABASE DeleteRecord SET RECOVERY SIMPLE;
GO


BEGIN TRY
BEGIN TRANSACTION
-- Bulk logged
SELECT  *
INTO    dbo.bigtable_intermediate
FROM    dbo.bigtable
WHERE   Id % 2 = 0;


-- minimal logged because DDL-Operation
TRUNCATE TABLE dbo.bigtable;


-- Bulk logged because target table is exclusivly locked!
SET IDENTITY_INSERT dbo.bigTable ON;
INSERT INTO dbo.bigtable WITH (TABLOCK) (Id, c1, c2, c3)
SELECT Id, c1, c2, c3 FROM dbo.bigtable_intermediate ORDER BY Id;
SET IDENTITY_INSERT dbo.bigtable OFF;
COMMIT
END TRY
BEGIN CATCH
IF @@TRANCOUNT > 0
ROLLBACK
END CATCH


ALTER DATABASE DeleteRecord SET RECOVERY FULL;
GO

Update November 2016

If you plan on storing this much data in a single table: don't. I highly recommend you consider table partitioning (either manually or with the built-in features if you're running Enterprise edition). This makes dropping old data as easy as truncating a table once a (week/month/etc.). If you don't have Enterprise (which we don't), you can simply write a script which runs once a month, drops tables older than 2 years, creates next month's table, and regenerates a dynamic view that joins all of the partition tables together for easy querying. Obviously "once a month" and "older than 2 years" should be defined by you based on what makes sense for your use-case. Deleting directly from a table with tens of billions of rows of data will a) take a HUGE amount of time and b) fill up the transaction log hundreds or thousands of times over.


提交答案