found a nice site with replication faq

http://www.replicationanswers.com/General.asp

Any detailed info after failed (like I/O, ...)?

What is the timeout on the snapshot agent?

The first link deals with the added RPC security of Windows 2003 (and MSDTC)

(starting at Set the appropriate MSDTC Security Configuration options on Windows Server 2003 SP1 and Windows XP...

Sorry, the extra dot was really tight

http://sql-server-performance.com/Community/forums/p/28131/151182.aspx

Have a look at

http://www.derkeiler.com/Newsgroups/microsoft.public.sqlserver.security/2003-05/0276.html

and

http://www.derkeiler.com/Newsgroups/microsoft.public.sqlserver.security/2003-05/0278.html

Have you checked the permissions of SQL Agent?

The autoupdating of stats usually improves execution plan.

Perhaps you can try this in a test-environment

CHECKPOINT --writes dirty blocks to disk

DBCC DROPCLEANBUFFERS --cleans the cache

DBCC FREEPROCCACHE --cleans the stored proc cache.

Then...

Hello,

Verify the permissions of the Windows accounts the replication jobs run under. (and also the linked server permissions)

It could be a oneway domain trust.

There is also something like the Windows...

Does the stored procedure changes set options (like set ANSI_NULL ON)

Verify the server SET OPTION.

Certain commands fail it some specific "options" aren't set. (like index on calculated columns etc)

Was this a one time hickup (maybe the machine shares resources with other programs)?

A wild guess:

Have a look at

http://sql-server-performance.com/Community/forums/p/28131/151182.aspx.

They mention verifying authentication setup and the linked server connection.

In my opinion it won't fire if you bypass the view. Would be quite a mess if the same table is referenced by multiple views.

Can be easily tested by creating...

your view also has the dbo prefix?

nice that you backup all database in case the hardware breaks (we had one by a defect harddisk shortcircuiting the san-backplane)

you could use the nolock hint

I don't know for sure but I thought there was a lengthlimit for the name of the primary key.

You could also try to remove the statistics and see what happens.