This article has been archived. It is offered "as is" and will no longer be updated.
Bug #: 470933 (SHILOH_BUGS)
Microsoft SQL Server 2000 fixes are distributed as one downloadable file. Because the fixes are cumulative, each new release contains all the hotfixes and all the security fixes that were included in the previous SQL Server 2000 fix release.
You have a workload that is running on a multiprocessor computer that forces SQL Server to perform many memory allocation operations and memory free operations each second (several hundred small batches that require compilation each second, for example). You may notice that a query that typically runs in a certain time takes longer to complete, although the query processes the same amount of data, uses the same query plan, and performs the same amount of I/O. The only noticeable difference is an increase in the CPU time and the elapsed time that it takes to run the query. In extreme cases, you may find that simple queries (such as SET ROWCOUNT 0) might take several hundred milliseconds to run and several hundred milliseconds of CPU time.
Service pack information
To resolve this problem, obtain the latest service pack for Microsoft SQL Server 2000. For additional information, click the following article number to view the article in the Microsoft Knowledge Base:
290211 How to obtain the latest SQL Server 2000 service pack
The English version of this hotfix has the file attributes (or later) that are listed in the following table. The dates and times for these files are listed in coordinated universal time (UTC). When you view the file information, it is converted to local time. To find the difference between UTC and local time, use the Time Zone tab in the Date and Time tool in Control Panel.
Note Because of file dependencies, the most recent hotfix or feature that contains these files may also contain additional files.
Microsoft has confirmed that this is a problem in the Microsoft products that are listed in the "Applies to" section of this article.
This problem was first corrected in Microsoft SQL Server 2000 Service Pack 4.
For additional information, click the following article number to view the article in the Microsoft Knowledge Base:
824684 Description of the Standard Terminology That Is Used to Describe Microsoft Software Updates
Most memory allocation requests in SQL Server are handled by allocating a page from the buffer pool. SQL Server maintains counters to track how many pages have been stolen from the buffer pool to be use as a generic memory buffer instead of a buffer for a database page. The DBCC MEMORYSTATUS command makes part of this information available and is automatically logged in the error log in certain out-of-memory conditions. These numbers are also used to trigger other events in the server, such as the lazywriter process. For additional information about these counters and stolen memory, click the following article number to view the article in the Microsoft Knowledge Base:
271624 INF: Using DBCC MEMORYSTATUS to monitor SQL Server memory usage
On multiprocessor computers, access to these counter values must be serialized. SQL Server serializes the counter values by using a spinlock. With a spinlock, the thread that wants to acquire the spinlock tries to acquire the lock. If the lock is not available, the thread spins in a loop, periodically rechecking the availability of the resource. If the operation that occurs inside the spinlock protection has a very short duration, such as incrementing or decrementing a counter, using a spinlock can be much faster than using a kernel synchronization object and then switching from user mode to kernel mode and back. However, if the spinlock becomes highly contended, a thread may spend more time spinning and trying to acquire the lock than it does performing useful work. Because of this, spinlocks are not a good choice for protecting a heavily-contended data structure.
This hotfix uses a different synchronization mechanism to maintain these counter values. It requires a substantial amount of contention to cause any noticeable delay or increase in CPU time. Typically, you must have a server with 8 or more processors (so that 8 threads are running at the same time) to be able to process a workload that can generate this much contention. This much contention may be more likely to occur if hyperthreading is enabled because it effectively doubles the number of SQL threads that can be running.