views:

19533

answers:

16

I took a look at the "Beginner's Guide to LINQ" post here on StackOverflow (http://stackoverflow.com/questions/8050/beginners-guide-to-linq), but had a follow-up question:

We're about to ramp up a new project where nearly all of our database op's will be fairly simple data retrievals (there's another segment of the project which already writes the data). Most of our other projects up to this point make use of stored procedures for such things. However, I'd like to leverage LINQ-to-SQL if it makes more sense.

So, the question is this: For simple data retrievals, which approach is better, LINQ-to-SQL or stored procs? Any specific pro's or con's?

Thanks.

+48  A: 

Linq to Sql.

Sql server will cache the query plans, so there's no performance gain for sprocs.

Your linq statements, on the other hand, will be logically part of and tested with your application. Sprocs are always a bit separated and are harder to maintain and test.

If I was working on a new application from scratch right now I would just use Linq, no sprocs.

Keith
I disagree regarding your comments on no gain for sprocs. I profiled a series of approaches to SQL Server access on a prod system, and using Prepare + stored procs had among the best results. Even across multiple calls of the same logic. Perhaps you are right on a DB w/ low usage, though.
torial
If you are, and will be, the most skilled database query developer, then use what you're most intimately familiar with. Unlike procedural code, you can't say "if it gives the right answer, ship it."
le dorfier
I think LINQ to SQL can cause recompilations when the size of a string parameter changes.
RussellH
@RussellH: This was true of .Net 3.5, they fixed that in .Net 4 (for both L2S and L2E)
BlueRaja - Danny Pflughoeft
@Kieth Not the case! Many more execution plans are created from Linq than SPs. Benefits exist with code to debug; but problems with recompiling for company deploy cycle; inflexibility to test different ACTION queries, WHERE, ORDER BY. Changing the order of WHERE statement can cause a different query to be processed; pre-fetching; this can't be easily done in Linq. Need to have the data layer available for tweaking. SPs are harder to maintain and test? If you're not used to it; but SPs are beneficial for corps and VLDBs, High-Availability, etc! Linq is for small projects, solo work; go for it.
SnapJag
@SnapJag - You're right that sprocs give you more fine grain control, but the fact is that 99% of the time (even in the large enterprise systems I work in) you don't need any additional optimisation. Sprocs are harder to maintain and test whether you're used to them or not - I'm very used to them (I was a DBA before I was a developer) and ensuring that the sproc for each CRUD operation for loads of different tables is up to date is a pain. The best practice (IMHO) is to code in the way that's easiest, then run performance checks and optimise only where its needed.
Keith
+2  A: 

I'm assuming you mean Linq To Sql

For any CRUD command it's easy to profile the performance of a stored procedure vs. any technology. In this case any difference between the two will be negligible. Try profiling for a 5 (simple types) field object over 100,000 select queries to find out if there's a real difference.

On the other hand the real deal-breaker will be the question on whether you feel comfortable putting your business logic on your database or not, which is an argument against stored procedures.

Jon Limjap
Since more places than the application can affect data - data imports, other applications, direct queries, etc. it is foolish not to put the business logic in the database. But if data integrity is not a concern to you, go ahead.
HLGEM
Business logic shouldn't go into the database. It should be in a programming language where it is easily debugged and managed. It can then be shared across applications as objects. Some rules might be in the database but not business logic.
4thSpace
+8  A: 

The best code is no code, and with stored procedures you have to write at least some code in the database and code in the application to call it , whereas with LINQ to SQL or LINQ to Entities, you don't have to write any additional code beyond any other LINQ query aside from instantiating a context object.

Mark Cidade
+38  A: 

I am generally a proponent of putting everything in stored procedures, for all of the reasons DBAs have been harping on for years. In the case of Linq, it is true that there will be no performance difference with simple CRUD queries.

But keep a few things in mind when making this decision: using any ORM couples you tightly to your data model. A DBA has no freedom to make changes to the data model without forcing you to change your compiled code. With stored procedures, you can hide these sorts of changes to an extent, since the parameter list and results set(s) returned from a procedure represent its contract, and the innards can be changed around, just so long as that contract is still met.

And also, if Linq is used for more complex queries, tuning the database becomes a much more difficult task. When a stored procedure is running slow, the DBA can totally focus on the code in isolation, and has lots of options, just so that contract is still satisfied when he/she is done.

I have seen many, many cases where serious problems in an application were addressed by changes to the schema and code in stored procedures without any change to deployed, compiled code.

Perhaps a hybird approach would be nice with Linq? Linq can, of course, be used to call stored procedures.

Eric Z Beard
One feature you neglected is security. With Linq, you have to expose the tables directly to the application. With stored procedures you can limit access to those procs and enforce your business requirements.
Chris Lively
I like your hybrid approach idea: if it's a simple CRUD that does not require joins and locking hints, especially if you're just querying one field, then you know the schema is not going to change and you won't be getting in fights with the DBA.
alord1689
"A DBA has no freedom to make changes to the data model without forcing you to change your compiled code." Why would you advocate this? No-one should have "freedom" to make changes, that's the point of configuration management. Changes should go through dev, test, etc before deployment regardless.
Neil Barnwell
@Neil: Yes, but he can't make changes in the development environment without forcing the developer to change the code.
Adam Robinson
I've got to say that I've seen the argument about tightly coupling to data bases a whole lot, and I'm not actually sure it's that valid. I've never actually come across a situation (beyond trivial examples) where I want to change the database that doesn't require a change in code higher up anyway. And really, how is Linq any different from stored procs or parameterized queries anyway. Your data access layer should still be nicely separated and abstracted regardless of whether you use an ORM or something simpler. That is what gives you the loose coupling, not what technology you use. Just my 2c
Simon P Stevens
@Chris: the security issue you speak of is easily solved using views and roles.
Gern Blandston
@Eric - Great comments. I'm a developer and DBA. I wear both hats for consulting purposes and advocate the same. LINQ has its place, it doesn't replace SPs. What is does enhance, is the ability for smaller projects/teams to bring product to market. But low will be the days when the company/project goes big and has to adapt for high-availability, VLDBs, tweaking, hints, complex queries, etc. I could never thing of using Linq in my high-volume, high-availability solutions. Especially when there are teams of programmers AND DBAs cooperating to provide big robust solutions.
SnapJag
+25  A: 

For basic data retrieval I would be going for Linq without hesitation.

Since moving to Linq I've found the following advantages:

  1. Debugging my DAL has never been easier.
  2. Compile time safety when your schema changes is priceless.
  3. Deployment is easier because everything is compiled into DLL's. No more managing deployment scripts.
  4. Because Linq can support querying anything that implements the IQueryable interface, you will be able to use the same syntax to query XML, Objects and any other datasource without having to learn a new syntax
lomaxx
for me compile time safety is key!
bbqchickenrobot
For deployment however, if you're on a corporate team that has a deployment cycle and a bug exists that needs to be out quickly, your deployment of DLLs through QA, etc, is probably more strict than the path to deployment of a single database script.Your advantages seem to better represent a small team/project scenario.
SnapJag
+10  A: 

@Eric LINQ to ADO.NET Entities doesn't tie you to a data model. With the entity framework, you define a conceptual data model that is mapped to the storage model, which you can change at any time, even without compiling (provided the mapping is in a text file as opposed to being part of the binary).

Mark Cidade
Hi. The comment button is right down there \/
BlueRaja - Danny Pflughoeft
+6  A: 

A DBA has no freedom to make changes to the data model without forcing you to change your compiled code. With stored procedures, you can hide these sorts of changes to an extent, since the parameter list and results set(s) returned from a procedure represent its contract, and the innards can be changed around, just so long as that contract is still met.

I really don't see this as being a benefit. Being able to change something in isolation might sound good in theory, but just because the changes fulfil a contract doesn't mean it's returning the correct results. To be able to determine what the correct results are you need context and you get that context from the calling code.

lomaxx
+59  A: 

Some advantages of LINQ over sprocs:

  1. Type safety: I think we all understand this.
  2. Abstraction: This is especially true with LINQ-to-Entities. This abstraction also allows the framework to add additional improvements that you can easily take advantage of. PLINQ is an example of adding multi-threading support to LINQ. Code changes are minimal to add this support. It would be MUCH harder to do this data access code that simply calls sprocs.
  3. Debugging support: I can use any .NET debugger to debug the queries. With sprocs, you cannot easily debug the SQL and that experience is largely tied to your database vendor (MS SQL Server provides a query analyzer, but often that isn't enough).
  4. Vendor agnostic: LINQ works with lots of databases and the number of supported databases will only increase. Sprocs are not always portable between databases, either because of varying syntax or feature support (if the database supports sprocs at all).
  5. Deployment: Others have mentioned this already, but it's easier to deploy a single assembly than to deploy a set of sprocs. This also ties in with #4.
  6. Easier: You don't have to learn T-SQL to do data access, nor do you have to learn the data access API (e.g. ADO.NET) necessary for calling the sprocs. This is related to #3 and #4.

Some disadvantages of LINQ vs sprocs:

  1. Network traffic: sprocs need only serialize sproc-name and argument data over the wire while LINQ sends the entire query. This can get really bad if the queries are very complex. However, LINQ's abstraction allows Microsoft to improve this over time.
  2. Less flexible: Sprocs can take full advantage of a database's featureset. LINQ tends to be more generic in it's support. This is common in any kind of language abstraction (e.g. C# vs assembler).
  3. Recompiling: If you need to make changes to the way you do data access, you need to recompile, version, and redeploy your assembly. Sprocs can sometimes allow a DBA to tune the data access routine without a need to redeploy anything.

Security and manageability are something that people argue about too.

  1. Security: For example, you can protect your sensitive data by restricting access to the tables directly, and put ACLs on the sprocs. With LINQ, however, you can still restrict direct access to tables and instead put ACLs on updatable table views to achieve a similar end (assuming your database supports updatable views).
  2. Manageability: Using views also gives you the advantage of shielding your application non-breaking from schema changes (like table normalization). You can update the view without requiring your data access code to change.

I used to be a big sproc guy, but I'm starting to lean towards LINQ as a better alternative in general. If there are some areas where sprocs are clearly better, then I'll probably still write a sproc but access it using LINQ. :)

Chris Gillum
"LINQ works with lots of databases "But LINQTOSQL only supports SQL Server.
RussellH
LINQ to Entities works with Postgres and MySql in addition to MSSQL. Not sure, but I thought I read there was something for Oracle around.
bbqchickenrobot
devart dotConnect has an Oracle thing, but it's buggy as hell. Also, I cannot get over the performance deficit that is introduced by having to select data out of the database in order to update it (Attach() is also possible, but it's rather poopey)
Ed Woodcock
+4  A: 

LINQ doesn't prohibit the use of stored procedures. I've used mixed mode with LINQ-SQL and LINQ-storedproc. Personally, I'm glad I don't have to write the stored procs....pwet-tu.

kenny
+14  A: 

LINQ will bloat the procedure cache

If an application is using LINQ to SQL and the queries involve the use of strings that can be highly variable in length, the SQL Server procedure cache will become bloated with one version of the query for every possible string length. For example, consider the following very simple queries created against the Person.AddressTypes table in the AdventureWorks2008 database:

        var p = 
            from n in x.AddressTypes 
            where n.Name == "Billing" 
            select n;

        var p = 
            from n in x.AddressTypes 
            where n.Name == "Main Office" 
            select n;

If both of these queries are run, we will see two entries in the SQL Server procedure cache: One bound with an NVARCHAR(7), and the other with an NVARCHAR(11). Now imagine if there were hundreds or thousands of different input strings, all with different lengths. The procedure cache would become unnecessarily filled with all sorts of different plans for the exact same query.

More here: https://connect.microsoft.com/VisualStudio/feedback/ViewFeedback.aspx?FeedbackID=363290

SQLMenace
Excellent bit of information. Thank you for sharing.
Martin Clarke
What a silly argument -- just use a variable and your problem is solved.
JohnOpincar
@John, it wouldn't help if you used a validate instead of a literal string since at the end you are sending a literal string to the database. it might look like a variable in your code, but it will be a string in the generated T-SQL that gets sent to the DB.
Keivan
SQLMenace: According to the page you linked to, the issue is solved in VS2010.
Mark Byers
+1  A: 

Stored Procs vs Code (Previous discussion)

liammclennan
+10  A: 

I think the pro LINQ argument seems to be coming from people who don't have a history with database development (in general).

Especially if using a product like VS DB Pro or Team Suite, many of the arguments made here do not apply, for instance:

Harder to maintain and Test: VS provides full syntax checking, style checking, referential and constraint checking and more. It also provide full unit testing capabilities and refactoring tools.

LINQ makes true unit testing impossible as (in my mind) it fails the ACID test.

Debugging is easier in LINQ: Why? VS allows full step-in from managed code and regular debugging of SPs.

Compiled into a single DLL rather than deployment scripts: Once again, VS comes to the rescue where it can build and deploy full databases or make data-safe incremental changes.

Don't have to learn TSQL with LINQ: No you don't, but you have to learn LINQ - where's the benefit?

I really don't see this as being a benefit. Being able to change something in isolation might sound good in theory, but just because the changes fulfil a contract doesn't mean it's returning the correct results. To be able to determine what the correct results are you need context and you get that context from the calling code.

Um, loosely coupled apps are the ultimate goal of all good programmers as they really do increase flexibility. Being able to change things in isolation is fantastic, and it is your unit tests that will ensure it is still returning appropriate results.

Before you all get upset, I think LINQ has its place and has a grand future. But for complex, data-intensive applications I do not think it is ready to take the place of stored procedures. This was a view I had echoed by an MVP at TechEd this year (they will remain nameless).

EDIT: The LINQ to SQL Stored Procedure side of things is something I still need to read more on - depending on what I find I may alter my above diatribe ;)

Dr8k
Learning LINQ is no big deal -- it's just syntactic sugar. TSQL is an entirely different language. Apples and Oranges.
Adam Lassek
The benefit of learning LINQ is that it isn't tied to a single technology - the querying semantics can be used for XML, objects and so on, and this will expand over time.
Dave R.
SnapJag
+6  A: 

LINQ definitely has its place in application-specific databases and in small businesses.

But in a large enterprise, where central databases serve as a hub of common data for many applications, we need abstraction. We need to centrally manage security and show access histories. We need to be able to do impact analysis: if I make a small change to the data model to serve a new business need, what queries need to be changed and what applications need to be re-tested? Views and Stored Procedures give me that. If LINQ can do all that, and make our programmers more productive, I'll welcome it -- does anyone have experience using it in this kind of environment?

You bring up a good point; LINQ is not an alternative in this case.
Meta-Knight
Your various applications should all be using a *common* data access layer (of course, that isn't always the case). If so, then, you can make a single change to the common library and redeploy. You can also use something like WCF to handle the data acess for you. There's a nice layer of abstraction that could technically use linq to access data behind the scenes.I guess it all just depends on what your guys come up with for your requirements in regards to using SPs vs Linq.
bbqchickenrobot
+4  A: 

Also there is the issue of possible 2.0 rollback. Trust me it has happened to me a couple times so I am sure it has happened to others.

I also agree that abstraction is best. Along with the fact the the original purpose of any ORM is to make RDBMS match up nicely to the OO concepts. However if everything worked fine before LINQ by having to deviate a bit from OO concepts then screw 'em. Concepts and reality don't always fit well together. There is no room for militant zealots in IT.

preach brother!
alchemical
+7  A: 

LINQ is new and has its place. LINQ is not invented to replace stored procedure.

Here I will focus on some performance myths & CONS, just for "LINQ to SQL", of course I might be totally wrong ;-)

(1)People say LINQ statment can "cache" in SQL server, so it doesn't lose performance. Partially true. "LINQ to SQL" actually is the runtime translating LINQ syntax to TSQL statment. So from the performance perspective,a hard coded ADO.NET SQL statement has no difference than LINQ.

(2)Given an example, a customer service UI has a "account transfer" function. this function itself might update 10 DB tables and return some messages in one shot. With LINQ, you have to build a set of statements and send them as one batch to SQL server. the performance of this translated LINQ->TSQL batch can hardly match stored procedure. Reason? because you can tweak the smallest unit of the statement in Stored procedue by using the built-in SQL profiler and execution plan tool, you can not do this in LINQ.

The point is, when talking single DB table and small set of data CRUD, LINQ is as fast as SP. But for much more complicated logic, stored procedure is more performance tweakable.

(3)"LINQ to SQL" easily makes newbies to introduce performance hogs. Any senior TSQL guy can tell you when not to use CURSOR (Basically you should not use CURSOR in TSQL in most cases). With LINQ and the charming "foreach" loop with query, It's so easy for a newbie to write such code:

foreach(Customer c in query)
{
  c.Country = "Wonder Land";
}
ctx.SubmitChanges();

You can see this easy decent code is so attractive. But under the hood, .NET runtime just translate this to an update batch. If there are only 500 lines, this is 500 line TSQL batch; If there are million lines, this is a hit. Of course, experienced user won't use this way to do this job, but the point is, it's so easy to fall in this way.

it's easy to fail with pointers in C/C++/C# does it mean it should never be used?
bbqchickenrobot
+4  A: 

IMHO, RAD = LINQ, RUP = Stored Procs. I worked for a large Fortune 500 company for many years, at many levels including management, and frankly, I would never hire RUP developers to do RAD development. They are so siloed that they very limited knowledge of what to do at other levels of the process. With a siloed environment, it makes sense to give DBAs control over the data through very specific entry points, because others frankly don't know the best ways to accomplish data management.

But large enterprises move painfully slow in the development arena, and this is extremely costly. There are times when you need to move faster to save both time and money, and LINQ provides that and more in spades.

Sometimes I think that DBAs are biased against LINQ because they feel it threatens their job security. But that's the nature of the beast, ladies and gentlemen.

Craig