This is on Microsoft SQL Server. We have a query where we are trying to join two tables on fields containing numeric data.
One table has the field defined as numeric(18,2) and the other table has the field defined as decimal(24,4). When joining with the native data types, the query hangs and we run out of patience before it will finish (left it running 6 min…). So we tried casting the two fields to be both numeric(18,2) and the query finished in under 10 seconds. So we tried casting the two fields to be both decimal(18,2) and again the query hangs. Does anyone know the difference between the decimal and numeric data types that would make them perform so differently?
DECIMAL and NUMERIC datatypes are the one and the same thing in SQL Server.
Quote from BOL:
Numeric data types that have fixed
precision and scale.
decimal[ (p[ ,s] )] and numeric[ (p[
,s] )] Fixed precision and scale
numbers. When maximum precision is
used, valid values are from - 10^38 +1
through 10^38 - 1. The ISO synonyms
for decimal are dec and dec(p, s).
numeric is functionally equivalent to
decimal.
From that, I'm surprised to hear of a difference. I'd expect the execution plans to be the same between the 2 routes, can you check?
Why are you using two datatypes to begin with? If they contain the same type of data (and joining on them implies they do), they should be the same datatype. Fix this and all your problems go away. Why waste server resources continually casting to match two fields that should be defined the same?
You of course may need to adjust the input variables for any insert or update queries to match waht you chose as the datatype.
My guess is that it's not a matter of a specific difference between the two data types, but simply the fact that SQL Server needs to implicitly convert them to match for the join operation.
I don't know why there would be a difference from your first query and the second, where you explicitly convert, but I can see why there might be a problem when you convert to a datatype that doesn't match and then SQL Server has to implicitly convert them anyway (as in your third case). Maybe in the first case, SQL Server is implicitly converting both to decimal(24,4) so as not to lose data and that operation takes longer than converting the other way. Have you tried explicitly converting the numeric(18,2) to a decimal(24,4)?
The title pretty much frames the question. I have not used CHAR in years. Right now, I am reverse-engineering a database that has CHAR all over it, for primary keys, codes, etc.
How about a CHAR(30) column?
Edit:
So the general opinion seems to be that CHAR if perfectly fine for certain things. I, however, think that you can design a database schema that does not have a need for "these certain things", thus not requiring fixed-length strings. With the bit, uniqueidentifier, varchar, and text types, it seems that in a well-normalized schema you get a certain elegance that you don't get when you use encoded string values. Thinking in fixed lenghts, no offense meant, seems to be a relic of the mainframe days (I learned RPG II once myself). I believe it is obsolete, and I did not hear a convincing argument from you claiming otherwise.
I use char(n) for codes, varchar(m) for descriptions. Char(n) seems to result in better performance because data doesn't need to move around when the size of contents change.
Where the nature of the data dictates the length of the field, I use CHAR. Otherwise VARCHAR.
CHARs are still faster for processing than VARCHARs in the DBMS I know well. Their fixed size allow for optimizations that aren't possible with VARCHARs. In addition, the storage requirements are slightly less for CHARS since no length has to be stored, assuming most of the rows need to fully, or near-fully, populate the CHAR column.
This is less of an impact (in terms of percentage) with a CHAR(30) than a CHAR(4).
As to usage, I tend to use CHARs when either:
the fields will generally always be close to or at their maximum length (stock codes, employee IDs, etc); or
the lengths are short (less than 10).
Anywhere else, I use VARCHARs.
I use CHAR when length of the value is fixed. For example we are generating a code or something based on some algorithm which returns the code with the specific fixed lenght lets say 13.
Otherwise, I found VARCHAR better. One more reason to use VARCHAR is that when you get the value back in your application you don't need to trim that value. In the case of CHAR you will get the full length of the column whether the value is filling it fully or not. It would get filled by spaces and you end up trimming every value, and forgetting that would lead to errors.
For PostgreSQL, the documentation states that char() has no advantage in storage space over varchar(); the only difference is that it's blank-padded to the specified length.
Having said that, I still use char(1) or char(3) for one-character or three-character codes. I think that the clarity due to the type specifying what the column should contain provides value, even if there are no storage or performance advantages. And yes, I typically use check constraints or foreign key constraints as well. Apart from those cases, I generally just stick with text rather than using varchar(). Again, this is informed by the database implementation, which automatically switches from inline to out-of-line storage if the value is large enough, which some other database implementations don't do.
Char isn't obsolete, it just should only be used if the length of the field should never vary. In the average database, this would be very few fields mostly some kind of code field like State Abbreviations which are a standard 2 character filed if you use the postal codes. Using Char where the filed length is varaible means that there will be a lot of trimming going on and that is extra, unnecessary work and the database should be refactored.
I am working on a legacy ASP application. I am attempting to insert a value (40.33) into a field in SQL Server 2000 that happens to be a float type. Every place I can see (via some logging) in the application is sending 40.33 to the Stored Procedure. When I run SQL Profiler against the database while the call is happening, the value that I see in the trace is 4.033000183105469e+001
Where is all the extra garbage coming from (the 183105469)?
Why is it that when I pass in 40, or 40.25 there is nothing extra?
Is this just one of the weird side effects of using float? When I am writing something I normally use money or decimal or something else, so not that familiar with the float datatype.
Yes, this is a weird, although well-known, side effect of using FLOAT.
In Microsoft SQL Server, you should use exact numeric datatypes such as NUMERIC, DECIMAL, MONEY or SMALLMONEY if you need exact numerics with scale.
Do not use FLOAT.
I think this is probably just a precision issue - the 0.33 part of the number can't be represented exactly in binary - this is probably the closest that you can get to.
The problem is that floats are not 100% accurate. If you need your numbers to be exact (especially when dealing with monetary values)... you should use a Decimal type.
I realize that CHAR is recommended if all my values are fixed-width. But, so what? Why not just pick VARCHAR for all text fields just to be safe.
The general rule is to pick CHAR if all rows will have close to the same length. Pick VARCHAR (or NVARCHAR) when the length varies significantly. CHAR may also be a bit faster because all the rows are of the same length.
It varies by DB implementation, but generally, VARCHAR (or NVARCHAR) uses one or two more bytes of storage (for length or termination) in addition to the actual data. So (assuming you are using a one-byte character set) storing the word "FooBar"
CHAR(6) = 6 bytes (no overhead)
VARCHAR(100) = 8 bytes (2 bytes of overhead)
CHAR(10) = 10 bytes (4 bytes of waste)
The bottom line is CHAR can be faster and more space-efficient for data of relatively the same length (within two characters length difference).
Note: Microsoft SQL has 2 bytes of overhead for a VARCHAR. This may vary from DB to DB, but generally, there is at least 1 byte of overhead needed to indicate length or EOL on a VARCHAR.
As was pointed out by Gaven in the comments: Things change when it comes to multi-byte characters sets, and is a is case where VARCHAR becomes a much better choice.
A note about the declared length of the VARCHAR: Because it stores the length of the actual content, then you don't waste unused length. So storing 6 characters in VARCHAR(6), VARCHAR(100), or VARCHAR(MAX) uses the same amount of storage. Read more about the differences when using VARCHAR(MAX). You declare a maximum size in VARCHAR to limit how much is stored.
In the comments AlwaysLearning pointed out that the Microsoft Transact-SQL docs seem to say the opposite. I would suggest that is an error or at least the docs are unclear.
If you're working with me and you're working with Oracle, I would probably make you use varchar in almost every circumstance. The assumption that char uses less processing power than varchar may be true...for now...but database engines get better over time and this sort of general rule has the making of a future "myth".
Another thing: I have never seen a performance problem because someone decided to go with varchar. You will make much better use of your time writing good code (fewer calls to the database) and efficient SQL (how do indexes work, how does the optimizer make decisions, why is exists faster than in usually...).
Final thought: I have seen all sorts of problems with use of CHAR, people looking for '' when they should be looking for ' ', or people looking for 'FOO' when they should be looking for 'FOO (bunch of spaces here)', or people not trimming the trailing blanks, or bugs with Powerbuilder adding up to 2000 blanks to the value it returns from an Oracle procedure.
In addition to performance benefits, CHAR can be used to indicate that all values should be the same length, e.g., a column for U.S. state abbreviations.
Char is a little bit faster, so if you have a column that you KNOW will be a certain length, use char. For example, storing (M)ale/(F)emale/(U)nknown for gender, or 2 characters for a US state.
Does NChar or Char perform better that their var alternatives?
Great question. The simple answer is yes in certain situations. Let's see if this can be explained.
Obviously we all know that if I create a table with a column of varchar(255) (let's call this column myColumn) and insert a million rows but put only a few characters into myColumn for each row, the table will be much smaller (overall number of data pages needed by the storage engine) than if I had created myColumn as char(255). Anytime I do an operation (DML) on that table and request alot of rows, it will be faster when myColumn is varchar because I don't have to move around all those "extra" spaces at the end. Move, as in when SQL Server does internal sorts such as during a distinct or union operation, or if it chooses a merge during it's query plan, etc. Move could also mean the time it takes to get the data from the server to my local pc or to another computer or wherever it is going to be consumed.
But there is some overhead in using varchar. SQL Server has to use a two byte indicator (overhead) to, on each row, to know how many bytes that particular row's myColumn has in it. It's not the extra 2 bytes that presents the problem, it's the having to "decode" the length of the data in myColumn on every row.
In my experiences it makes the most sense to use char instead of varchar on columns that will be joined to in queries. For example the primary key of a table, or some other column that will be indexed. CustomerNumber on a demographic table, or CodeID on a decode table, or perhaps OrderNumber on an order table. By using char, the query engine can more quickly perform the join because it can do straight pointer arithmetic (deterministically) rather than having to move it's pointers a variable amount of bytes as it reads the pages. I know I might have lost you on that last sentence. Joins in SQL Server are based around the idea of "predicates." A predicate is a condition. For example myColumn = 1, or OrderNumber < 500.
So if SQL Server is performing a DML statement, and the predicates, or "keys" being joined on are a fixed length (char), the query engine doesn't have to do as much work to match rows from one table to rows from another table. It won't have to find out how long the data is in the row and then walk down the string to find the end. All that takes time.
Now bear in mind this can easily be poorly implemented. I have seen char used for primary key fields in online systems. The width must be kept small i.e. char(15) or something reasonable. And it works best in online systems because you are usually only retrieving or upserting a small number of rows, so having to "rtrim" those trailing spaces you'll get in the result set is a trivial task as opposed to having to join millions of rows from one table to millions of rows on another table.
Another reason CHAR makes sense over varchar on online systems is that it reduces page splits. By using char, you are essentially "reserving" (and wasting) that space so if a user comes along later and puts more data into that column SQL has already allocated space for it and in it goes.
Another reason to use CHAR is similar to the second reason. If a programmer or user does a "batch" update to millions of rows, adding some sentence to a note field for example, you won't get a call from your DBA in the middle of the night wondering why their drives are full. In other words, it leads to more predictable growth of the size of a database.
So those are 3 ways an online (OLTP) system can benefit from char over varchar. I hardly ever use char in a warehouse/analysis/OLAP scenario because usually you have SO much data that all those char columns can add up to lots of wasted space.
Keep in mind that char can make your database much larger but most backup tools have data compression so your backups tend to be about the same size as if you had used varchar. For example LiteSpeed or RedGate SQL Backup.
Another use is in views created for exporting data to a fixed width file. Let's say I have to export some data to a flat file to be read by a mainframe. It is fixed width (not delimited). I like to store the data in my "staging" table as varchar (thus consuming less space on my database) and then use a view to CAST everything to it's char equivalent, with the length corresponding to the width of the fixed width for that column. For example:
create table tblStagingTable (
pkID BIGINT (IDENTITY,1,1),
CustomerFirstName varchar(30),
CustomerLastName varchar(30),
CustomerCityStateZip varchar(100),
CustomerCurrentBalance money )
insert into tblStagingTable
(CustomerFirstName,CustomerLastName, CustomerCityStateZip) ('Joe','Blow','123 Main St Washington, MD 12345', 123.45)
create view vwStagingTable AS
SELECT CustomerFirstName = CAST(CustomerFirstName as CHAR(30)),
CustomerLastName = CAST(CustomerLastName as CHAR(30)),
CustomerCityStateZip = CAST(CustomerCityStateZip as CHAR(100)),
CustomerCurrentBalance = CAST(CAST(CustomerCurrentBalance as NUMERIC(9,2)) AS CHAR(10))
SELECT * from vwStagingTable
This is cool because internally my data takes up less space because it's using varchar. But when I use DTS or SSIS or even just a cut and paste from SSMS to Notepad, I can use the view and get the right number of trailing spaces. In DTS we used to have a feature called, damn I forget I think it was called "suggest columns" or something. In SSIS you can't do that anymore, you have to tediously define the flat file connection manager. But since you have your view setup, SSIS can know the width of each column and it can save alot of time when building your data flow tasks.
So bottom line... use varchar. There are a very small number of reasons to use char and it's only for performance reasons. If you have a system with hundrends of millions of rows you will see a noticeable difference if the predicates are deterministic (char) but for most systems using char is simply wasting space.
Hope that helps.
Jeff
There are performance benefits, but here is one that has not been mentioned: row migration. With char, you reserve the entire space in advance.So let's says you have a char(1000), and you store 10 characters, you will use up all 1000 charaters of space. In a varchar2(1000), you will only use 10 characters. The problem comes when you modify the data. Let's say you update the column to now contain 900 characters. It is possible that the space to expand the varchar is not available in the current block. In that case, the DB engine must migrate the row to another block, and make a pointer in the original block to the new row in the new block. To read this data, the DB engine will now have to read 2 blocks.
No one can equivocally say that varchar or char are better. There is a space for time tradeoff, and consideration of whether the data will be updated, especially if there is a good chance that it will grow.
There is a difference between early performance optimization and using a best practice type of rule. If you are creating new tables where you will always have a fixed length field, it makes sense to use CHAR, you should be using it in that case. This isn't early optimization, but rather implementing a rule of thumb (or best practice).
i.e. - If you have a 2 letter state field, use CHAR(2). If you have a field with the actual state names, use VARCHAR.
I would choose varchar unless the column stores fixed value like US state code -- which is always 2 chars long and the list of valid US states code doesn't change often :).
In every other case, even like storing hashed password (which is fixed length), I would choose varchar.
Why -- char type column is always fulfilled with spaces, which makes for column my_column defined as char(5) with value 'ABC' inside comparation:
my_column = 'ABC' -- my_column stores 'ABC ' value which is different then 'ABC'
false.
This feature could lead to many irritating bugs during development and makes testing harder.
CHAR takes up less storage space than VARCHAR if all your data values in that field are the same length. Now perhaps in 2009 a 800GB database is the same for all intents and purposes as a 810GB if you converted the VARCHARs to CHARs, but for short strings (1 or 2 characters), CHAR is still a industry "best practice" I would say.
Now if you look at the wide variety of data types most databases provide even for integers alone (bit, tiny, int, bigint), there ARE reasons to choose one over the other. Simply choosing bigint every time is actually being a bit ignorant of the purposes and uses of the field. If a field simply represents a persons age in years, a bigint is overkill. Now it's not necessarily "wrong", but it's not efficient.
But its an interesting argument, and as databases improve over time, it could be argued CHAR vs VARCHAR does get less relevant.
I would NEVER use chars. I’ve had this debate with many people and they always bring up the tired cliché that char is faster. Well I say, how much faster? What are we talking about here, milliseconds, seconds and if so how many? You’re telling me because someone claims its a few milliseconds faster, we should introduce tons of hard to fix bugs into the system?
So here are some issues you will run into:
Every field will be padded, so you end up with code forever that has RTRIMS everywhere. This is also a huge disk space waste for the longer fields.
Now let’s say you have the quintessential example of a char field of just one character but the field is optional. If somebody passes an empty string to that field it becomes one space. So when another application/process queries it, they get one single space, if they don’t use rtrim. We’ve had xml documents, files and other programs, display just one space, in optional fields and break things.
So now you have to ensure that you’re passing nulls and not empty string, to the char field. But that’s NOT the correct use of null. Here is the use of null. Lets say you get a file from a vendor
Name|Gender|City
Bob||Los Angeles
If gender is not specified than you enter Bob, empty string and Los Angeles into the table. Now lets say you get the file and its format changes and gender is no longer included but was in the past.
Name|City
Bob|Seattle
Well now since gender is not included, I would use null. Varchars support this without issues.
Char on the other hand is different. You always have to send null. If you ever send empty string, you will end up with a field that has spaces in it.
I could go on and on with all the bugs I’ve had to fix from chars and in about 20 years of development.
I stand by Jim McKeeth's comment.
Also, indexing and full table scans are faster if your table has only CHAR columns. Basically the optimizer will be able to predict how big each record is if it only has CHAR columns, while it needs to check the size value of every VARCHAR column.
Besides if you update a VARCHAR column to a size larger than its previous content you may force the database to rebuild its indexes (because you forced the database to physically move the record on disk). While with CHAR columns that'll never happen.
But you probably won't care about the performance hit unless your table is huge.
Remember Djikstra's wise words. Early performance optimization is the root of all evil.
Many people have pointed out that if you know the exact length of the value using CHAR has some benefits. But while storing US states as CHAR(2) is great today, when you get the message from sales that 'We have just made our first sale to Australia', you are in a world of pain. I always send to overestimate how long I think fields will need to be rather than making an 'exact' guess to cover for future events. VARCHAR will give me more flexibility in this area.
I think in your case there is probably no reason to not pick Varchar. It gives you flexibility and as has been mentioned by a number of respondants, performance is such now that except in very specific circumstances us meer mortals (as opposed to Google DBA's) will not notice the difference.
An interesting thing worth noting when it comes to DB Types is the sqlite (a popular mini database with pretty impressive performance) puts everything into the database as a string and types on the fly.
I always use VarChar and usually make it much bigger than I might strickly need. Eg. 50 for Firstname, as you say why not just to be safe.
It's the classic space versus performance tradeoff.
In MS SQL 2005, Varchar (or NVarchar for lanuagues requiring two bytes per character ie Chinese) are variable length. If you add to the row after it has been written to the hard disk it will locate the data in a non-contigious location to the original row and lead to fragmentation of your data files. This will affect performance.
So, if space is not an issue then Char are better for performance but if you want to keep the database size down then varchars are better.
Fragmentation. Char reserves space and VarChar does not. Page split can be required to accommodate update to varchar.
There is some small processing overhead in calculating the actual needed size for a column value and allocating the space for a Varchar, so if you are definitely sure how long the value will always be, it is better to use Char and avoid the hit.
when using varchar values SQL Server needs an additional 2 bytes per row to store some info about that column whereas if you use char it doesn't need that
so unless you
Using CHAR (NCHAR) and VARCHAR (NVARCHAR) brings differences in the ways the database server stores the data. The first one introduces trailing blanks; I have encountered problem when using it with LIKE operator in SQL SERVER functions. So I have to make it safe by using VARCHAR (NVARCHAR) all the times.
For example, if we have a table TEST(ID INT, Status CHAR(1)), and you write a function to list all the records with some specific value like the following:
CREATE FUNCTION List(#Status AS CHAR(1) = '')
RETURNS TABLE
AS
RETURN
SELECT * FROM TEST
WHERE Status LIKE '%' + #Status '%'
In this function we expect that when we put the default parameter the function will return all the rows, but in fact it does not. Change the #Status data type to VARCHAR will fix the issue.
In some SQL databases, VARCHAR will be padded out to its maximum size in order to optimize the offsets, This is to speed up full table scans and indexes.
Because of this, you do not have any space savings by using a VARCHAR(200) compared to a CHAR(200)