Find the size of an index row

Hey Guys,
I would like to enumerate all the indexes in a table and find the size of
the index, when I mean the size not the total space occupied but the size of
each row of the index. Eg: Table EMP has id bigint, name varchar(64),
address varchar(64), phone bigint, zip int, etc. Index 1 (id, name), Index
2(zip, phone), Index 3(phone) Each row in this table could occupy 8+64+64+8+4
bytes. So each index associated with each row would occupy 8+64 + 8+8 + 8.
That's what I want. This would be used in later calculating the required
size based on number of records to be inserted.yodarules wrote:
> Hey Guys,
> I would like to enumerate all the indexes in a table and find the
> size of the index, when I mean the size not the total space occupied
> but the size of each row of the index. Eg: Table EMP has id bigint,
> name varchar(64), address varchar(64), phone bigint, zip int, etc.
> Index 1 (id, name), Index 2(zip, phone), Index 3(phone) Each row in
> this table could occupy 8+64+64+8+4 bytes. So each index associated
> with each row would occupy 8+64 + 8+8 + 8. That's what I want. This
> would be used in later calculating the required size based on number
> of records to be inserted.
See "Estimating the Size of a Table with a Clustered Index" in BOL.
--
David Gugick
Quest Software
www.quest.com|||Thanks David,
But thats not what I want, as I have explained in my post earlier. This
tells you the method to calcualte the space, which is what I might do when I
try to achieve my final goal. But the immediate goal is to dynamically find
the size of each non-clustered index on that table. The reason why I want to
do it dynamically is that, if I add or remove indexes, my SP should be able
to work with that rather than a hard coded value.
"David Gugick" wrote:
> yodarules wrote:
> > Hey Guys,
> >
> > I would like to enumerate all the indexes in a table and find the
> > size of the index, when I mean the size not the total space occupied
> > but the size of each row of the index. Eg: Table EMP has id bigint,
> > name varchar(64), address varchar(64), phone bigint, zip int, etc.
> > Index 1 (id, name), Index 2(zip, phone), Index 3(phone) Each row in
> > this table could occupy 8+64+64+8+4 bytes. So each index associated
> > with each row would occupy 8+64 + 8+8 + 8. That's what I want. This
> > would be used in later calculating the required size based on number
> > of records to be inserted.
> See "Estimating the Size of a Table with a Clustered Index" in BOL.
> --
> David Gugick
> Quest Software
> www.quest.com
>|||A couple of thoughts on your question.
1) SQL Server stores everything in 8KB pages and stores pages in 8 page
extents. Based on the fill factor and size of your index SQL Server will fit
as many rows from your index within the page as it can.
2) A non-clustered index always contains a reference to the key of the
clustered index or a pointer to a place in the heap if no clustered index
exists, in addition to the data in the non-clustered index. Additionally, a
varchar field does not always take up all 8 bytes. Varchars are variable
length and only stored with the bytes taken up by the value within the field.
So, your example is not actually accurate in how big a row of the index (and
it is further dependent on whether the index is clustered or non-clustered)
would be.
I'm not sure what exactly you are trying to do. But, in SQL 2000 you can
use DBCC SHOWCONTIG to see how many pages and extents an index takes up. It
will also show you if there is fragmentation in those pages and extents. In
SQL 2005 this command is replaced by the DMV dm_db_index_physical_stats.
dm_db_index_physical_stats also provides values for min,avg and max record
size. Which might be what you are looking for. I don't know if you can find
that data for SQL 2000.
The article in BOL that David pointed you to is a good illustration of how
that calculation would be done manually for a particular table and its
indexes.
HTH
--
Ryan Powers
Clarity Consulting
http://www.claritycon.com
"David Gugick" wrote:
> yodarules wrote:
> > Hey Guys,
> >
> > I would like to enumerate all the indexes in a table and find the
> > size of the index, when I mean the size not the total space occupied
> > but the size of each row of the index. Eg: Table EMP has id bigint,
> > name varchar(64), address varchar(64), phone bigint, zip int, etc.
> > Index 1 (id, name), Index 2(zip, phone), Index 3(phone) Each row in
> > this table could occupy 8+64+64+8+4 bytes. So each index associated
> > with each row would occupy 8+64 + 8+8 + 8. That's what I want. This
> > would be used in later calculating the required size based on number
> > of records to be inserted.
> See "Estimating the Size of a Table with a Clustered Index" in BOL.
> --
> David Gugick
> Quest Software
> www.quest.com
>|||I totally understand and am aware of the points mentioned. The main reason
why I want to do this to for size estimation. When we archive and delete
some records we store the number of rows deleted. When we reload the data
back, we want to do a space estimation and see if there's enough space to do
this operation. Its actaully much more complicated than what I have just
explained, but thats the idea. I don't want to do spaceused before and after
for various other reasons and this seems to be the best case for us. If I
cannot dynamically find index sizes, then I have to hard code it some where,
again we are happy with an approximation, hence don't really care about
varchar, etc. Actually the tables I'm interested in doing this don't have
varchar.
"Ryan Powers" wrote:
> A couple of thoughts on your question.
> 1) SQL Server stores everything in 8KB pages and stores pages in 8 page
> extents. Based on the fill factor and size of your index SQL Server will fit
> as many rows from your index within the page as it can.
> 2) A non-clustered index always contains a reference to the key of the
> clustered index or a pointer to a place in the heap if no clustered index
> exists, in addition to the data in the non-clustered index. Additionally, a
> varchar field does not always take up all 8 bytes. Varchars are variable
> length and only stored with the bytes taken up by the value within the field.
> So, your example is not actually accurate in how big a row of the index (and
> it is further dependent on whether the index is clustered or non-clustered)
> would be.
> I'm not sure what exactly you are trying to do. But, in SQL 2000 you can
> use DBCC SHOWCONTIG to see how many pages and extents an index takes up. It
> will also show you if there is fragmentation in those pages and extents. In
> SQL 2005 this command is replaced by the DMV dm_db_index_physical_stats.
> dm_db_index_physical_stats also provides values for min,avg and max record
> size. Which might be what you are looking for. I don't know if you can find
> that data for SQL 2000.
> The article in BOL that David pointed you to is a good illustration of how
> that calculation would be done manually for a particular table and its
> indexes.
> HTH
> --
> Ryan Powers
> Clarity Consulting
> http://www.claritycon.com
>
> "David Gugick" wrote:
> > yodarules wrote:
> > > Hey Guys,
> > >
> > > I would like to enumerate all the indexes in a table and find the
> > > size of the index, when I mean the size not the total space occupied
> > > but the size of each row of the index. Eg: Table EMP has id bigint,
> > > name varchar(64), address varchar(64), phone bigint, zip int, etc.
> > > Index 1 (id, name), Index 2(zip, phone), Index 3(phone) Each row in
> > > this table could occupy 8+64+64+8+4 bytes. So each index associated
> > > with each row would occupy 8+64 + 8+8 + 8. That's what I want. This
> > > would be used in later calculating the required size based on number
> > > of records to be inserted.
> >
> > See "Estimating the Size of a Table with a Clustered Index" in BOL.
> >
> > --
> > David Gugick
> > Quest Software
> > www.quest.com
> >|||Okay.
So, it sounds like you are using sql 2000, and you are looking to
programmatically access the indexes, the columns that make up those indexes
and the size of the columns. So, you can estimate size. I could be jumping
to the wrong conclusions.
There are some undocumented SPs that could help you get there.
The following all seem to give you parts of what you want,
sp_MStablespace
sp_MSindexspace
sp_MShelpindex
sp_MShelpcolumns
sp_columns_rowset
For specifics about using any/ all of the above. Follow this link.
http://www.sql-server-performance.com/ac_sql_server_2000_undocumented_sp.asp
Note that since these are undocumented, they won't necessarily work in SQL
2005.
HTH
--
Ryan Powers
Clarity Consulting
http://www.claritycon.com
"yodarules" wrote:
> I totally understand and am aware of the points mentioned. The main reason
> why I want to do this to for size estimation. When we archive and delete
> some records we store the number of rows deleted. When we reload the data
> back, we want to do a space estimation and see if there's enough space to do
> this operation. Its actaully much more complicated than what I have just
> explained, but thats the idea. I don't want to do spaceused before and after
> for various other reasons and this seems to be the best case for us. If I
> cannot dynamically find index sizes, then I have to hard code it some where,
> again we are happy with an approximation, hence don't really care about
> varchar, etc. Actually the tables I'm interested in doing this don't have
> varchar.
> "Ryan Powers" wrote:
> > A couple of thoughts on your question.
> >
> > 1) SQL Server stores everything in 8KB pages and stores pages in 8 page
> > extents. Based on the fill factor and size of your index SQL Server will fit
> > as many rows from your index within the page as it can.
> > 2) A non-clustered index always contains a reference to the key of the
> > clustered index or a pointer to a place in the heap if no clustered index
> > exists, in addition to the data in the non-clustered index. Additionally, a
> > varchar field does not always take up all 8 bytes. Varchars are variable
> > length and only stored with the bytes taken up by the value within the field.
> > So, your example is not actually accurate in how big a row of the index (and
> > it is further dependent on whether the index is clustered or non-clustered)
> > would be.
> >
> > I'm not sure what exactly you are trying to do. But, in SQL 2000 you can
> > use DBCC SHOWCONTIG to see how many pages and extents an index takes up. It
> > will also show you if there is fragmentation in those pages and extents. In
> > SQL 2005 this command is replaced by the DMV dm_db_index_physical_stats.
> > dm_db_index_physical_stats also provides values for min,avg and max record
> > size. Which might be what you are looking for. I don't know if you can find
> > that data for SQL 2000.
> >
> > The article in BOL that David pointed you to is a good illustration of how
> > that calculation would be done manually for a particular table and its
> > indexes.
> >
> > HTH
> >
> > --
> > Ryan Powers
> > Clarity Consulting
> > http://www.claritycon.com
> >
> >
> > "David Gugick" wrote:
> >
> > > yodarules wrote:
> > > > Hey Guys,
> > > >
> > > > I would like to enumerate all the indexes in a table and find the
> > > > size of the index, when I mean the size not the total space occupied
> > > > but the size of each row of the index. Eg: Table EMP has id bigint,
> > > > name varchar(64), address varchar(64), phone bigint, zip int, etc.
> > > > Index 1 (id, name), Index 2(zip, phone), Index 3(phone) Each row in
> > > > this table could occupy 8+64+64+8+4 bytes. So each index associated
> > > > with each row would occupy 8+64 + 8+8 + 8. That's what I want. This
> > > > would be used in later calculating the required size based on number
> > > > of records to be inserted.
> > >
> > > See "Estimating the Size of a Table with a Clustered Index" in BOL.
> > >
> > > --
> > > David Gugick
> > > Quest Software
> > > www.quest.com
> > >