Multiple/single instance of Linq to SQL DataContext

I have a project with a number of different classes querying and modifying data in a common set of tables. I've set up a .dbml file which provides us with a DataContext class. My question is whether a single instance of the DataContext should be used by all objects, or whether multiple instances are safe to use. I'm also wondering about thread safety in the case of a single DataContext, and whether access to it's methods should be synchronized.

Rick Strahl has a nice article about your options: http://www.west-wind.com/weblog/posts/246222.aspx.

See also: LINQ to SQL - where does your DataContext live?.

You may want a slightly different strategy for each type of deployment - web, desktop, windows service...

Summarized, your options are:

• Global DataContext - dangerous in multi-threaded environments (including web apps). Remember that instance members are not guaranteed to be thread-safe (from Bradley Grainger's answer above).
• DataContext per thread - complicated. If your DataContext is tracking changes you must be sure to flush them at the appropriate time. Instantiating, storing, and retrieving the DataContext is a pain.
• DataContext per atomic action - you lose the ability to track changes since one DataContext creates an object while another updates or deletes it. Attaching a data object to a new DataContext may not work like you expect.
• DataContext per data object - seems inelegant because you have to fuss with the DataContext on instantiation(create and attach) and update/delete (pull it off the data object and use it).

I opted for a DataContext per data object. It may not be the fanciest solution but it works in all deployment environments.

I use an new instance of DataContext for every transaction.

Reusing old instances can be troublesome, and will bloat the content of the DataContext, since any item that has been loaded at some time, will have to be tracked - your app will get slower and slower, bloating up memory.

If you need an item longer than for a transaction, you can detach it from the DataContext by cloning the item, and can reattach it later to a new and fresh DataContext using Attach().
I even can clone an item, send it over the network with WCF, get it back in some later call, attach it to a new DataContext and save the changes (of course I need a timestamp column for this).