Re: Char and Varchar
From: Jacco Schalkwijk (NOSPAMjaccos_at_eurostop.co.uk)
Date: 05/13/04
- Next message: Narayana Vyas Kondreddi: "Re: Memory Crash?"
- Previous message: Benny Tordrup: "Re: Connecting to remote named instance"
- In reply to: Hugo Kornelis: "Re: Char and Varchar"
- Next in thread: Prabhat: "Re: Char and Varchar"
- Messages sorted by: [ date ] [ thread ]
Date: Thu, 13 May 2004 10:12:37 +0100
> >Changing the length frequently would required
> >rebuilding the data page - right?
> >
> >If changing an element to full size causes it to no longer fit on the
data
> >page, then page splits occur - that's alot of work for the engine to
handle.
>
> It's been a while since I read this part, but I think that's not
> entirely correct. The row's data itself will have to be rebuilt, as
> there can be no gaps between the contents of the varying length
> columns. If the row is shortened, it will stay where it is. If the
> row's size increases, it is replaced by a forwarding pointer and the
> new row's content is stored somewhere else; the other rows on the same
> page remain intact. If the row has to move again later, the original
> forwarding pointer is changed (to prevent long chains of pointers).
That is only correct for heaps, tables without a clustered index. In a table
with a clustered index this scenario will cause a page split, as the rows
have to stay in the same position in the clustered index.
-- Jacco Schalkwijk SQL Server MVP "Hugo Kornelis" <hugo@pe_NO_rFact.in_SPAM_fo> wrote in message news:bl45a0tjo06s6npdrm9rf83mg8o4n80uik@4ax.com... > On Wed, 12 May 2004 09:36:31 -0400, Steve Z wrote: > > >Hey guys, he was asking how it was stored - it's always good to know what > >goes on behind the scenes. > > I agree. That's why I replied to your post, acknowledging that you > were almost completely right, correcting a small error and adding an > important consideration. I have seen too many examples of people using > only fixed-length charachter data in order to avoid the overhead of > having to find the start and end of the data. Much as I'd like to > believe that people will realise by themselves that this overhead is > nothing compared to the increased I/O, I can't. > > > >Would you ever create a table with lets say 100 single character VARCHAR > >fields? Would there be value to that, if the columns were guaranteed to > >have a character in them (ie. not a space). > > Of course not. If the maximum length is short (<= 10 characters), I > always use CHAR. If it's longer, but most values use almost the full > maximum length, I also use CHAR. I use VARCHAR if long and short > strings are interspersed. > > > >How would a primary clustered key behave if it was built on a VARCHAR field > >vs. a CHAR field? > > It depends. If the column has a maximum length of 250 characters, but > most values are a lot shorter, using VARCHAR would reduce the number > of index pages on the intermediate levels. The leaf level would of > course not be affected (apart from the effect of any CHAR vs VARCHAR > column decision on the data pages). For an unindexed key, the leaf > level would need less pages as well. If the table holds many rows, I > think (I haven't tested it!) that the number of I/O's would be less, > hence the queries would return results faster. All this at the price > of the CPU having less idle time. > > > >And, please explain to me how the 100th VARCHAR fields byte position can be > >determined in a buffer of data without reading each of the prior 99 VARCHAR > >control fields? Without knowing the length of each of the prior 99 data > >elements in that row, how could the starting byte of the 100th element be > >determined? > > Easy. Take the end position of the 99th element (stored ar a fixed > position in the row, after the fixed length and before the varying > length data) and add 1. > > > > Since the length of the VARCHAR is limited to 8000 byte - that > >takes 14 of the 16 bits in the 2-byte control field. So, I can assume from > >that fact that the 2-byte control field is not storing the "position" of the > >element, but simply the length of each element. > > You seem to forget that the length of the complete row may not exceed > 8,060 bytes (check BOL, Maximum Capacity Specifications). The 2-byte > control field will hold the ending position of each element just fine. > > > >Inside MS SQL Server 2000, by Kalen Delany, on page 230 states that VARCHAR > >are most appropriate when you expect significant differences in the lengths > >of the data in the column and when the data length in the column wont > >frequently be changed. > > Apart from the page number, this has apparently not changed from the > previous edition (Inside MS SQL Server 7.0, by Ron Soukup and Kalen > Delany). > > > >Changing the length frequently would required > >rebuilding the data page - right? > > > >If changing an element to full size causes it to no longer fit on the data > >page, then page splits occur - that's alot of work for the engine to handle. > > It's been a while since I read this part, but I think that's not > entirely correct. The row's data itself will have to be rebuilt, as > there can be no gaps between the contents of the varying length > columns. If the row is shortened, it will stay where it is. If the > row's size increases, it is replaced by a forwarding pointer and the > new row's content is stored somewhere else; the other rows on the same > page remain intact. If the row has to move again later, the original > forwarding pointer is changed (to prevent long chains of pointers). > > > >Page 233 indicates that "no one answer is right" and "If you understand the > >tradeoffs; you'll be able to make the best choice". > > Yes - a terrific piece of advice that I support wholeheartedly! > > > Best, Hugo > -- > > (Remove _NO_ and _SPAM_ to get my e-mail address)
- Next message: Narayana Vyas Kondreddi: "Re: Memory Crash?"
- Previous message: Benny Tordrup: "Re: Connecting to remote named instance"
- In reply to: Hugo Kornelis: "Re: Char and Varchar"
- Next in thread: Prabhat: "Re: Char and Varchar"
- Messages sorted by: [ date ] [ thread ]
Relevant Pages
|
|
