SQL Server for Windows NT is a Win32 application that runs on the WindowsNT or Windows NT Advanced Server operating systems. In some uncommonsituations, when SQL Server is running a compute-bound operation, theinteractive console responsiveness of Windows NT may be diminished. Thisarticle discusses characteristics of the Windows NT process scheduler(often called thread scheduler or thread dispatcher) that relate to thisbehavior.
Windows NT and Windows NT Advanced Server use the same system of schedulingWin32 processes and threads. In other operating systems, this is usuallycalled process scheduling. With Windows NT, the unit of scheduling is thethread; hence, the term thread scheduling or dispatching. However, the termprocess will be used in this discussion, where appropriate. The unit ofscheduling time is called the time slice or quantum.
Windows NT uses 32 numerical thread priorities, ranging from 1 (the lowest)to 31 (the highest) with 0 being reserved for system use.
Threads 1 through 15 are variable priority in that the scheduler adjuststhe priority based on thread behavior. The base priority of a thread is thebase level from which these upward adjustments are made. The currentpriority of a thread is called its dynamic priority. Interactive threadsthat yield before their time slice is up will tend to be adjusted upward inpriority from their base priority. Compute-bound threads that do not yield,consuming their entire time slice, will tend to have their prioritydecreased, but not below the base level. This arrangement is often calledheuristic scheduling. It provides better interactive performance and tendsto lessen the system impact of "CPU hog" threads.
Windows NT also uses additional priority adjustments based on other events,such as momentarily boosting thread priority when it returns from an I/Ocall, when returning from a user-mode sub-system call, and when receivingkeyboard input.
Priority threads 16 through 31 are termed real-time, and do not vary inpriority based on behavior. This provides more deterministic scheduling,which is useful for near-real-time systems.
Although there is a numerical priority for each thread, these are usuallyreferred to within a Win32 program by their process priority flag or threadpriority flag. There is a mapping of these flags to numerical threadpriorities, which is detailed in the Win32 SDK, Windows NT Resource KitVol. 3, and Knowledge Base article 106253
. In this article we primarilyrefer to these by their numerical equivalent to reduce confusion.
The key to understanding Windows NT thread scheduling and resultantapplication behavior is knowing the central algorithm used. This algorithmis very simple, and is the same one a number of other operating systemsuse. It is "highest priority ready thread gets run." A list of readythreads or processes exist which is often called the "dispatch queue" or"eligible queue." The queue entries are in order based on their individualpriority. A hardware-driven real-time clock or interval timer willperiodically interrupt, passing control to a device driver that calls theprocess or thread scheduler. The thread scheduler will take the highestpriority entry from the queue and dispatch it to run.
Thus, a hardware event drives the regular examination of thread states, andthe evaluation of which is to run next. The frequency of this operation isquite rapid, often from 10 to 32 milliseconds on most systems.
Neither Windows NT nor many other operating systems use CPU quotas. Thescheduling algorithm merely selects the highest priority ready-to-runthread and allows it to run uninterrupted until the next clock tick, atwhich time the scheduler gets control and reevaluates which thread is thehighest priority and ready to run. However, if the previously runningthread is still highest priority, and still ready to run (has not blockedon I/O or otherwise changed states), it will be run again.
Also there is essentially no gradual or proportional control over theamount of CPU time a compute-bound thread receives. Often the misconceptionexists that by lowering the priority of a compute bound thread a certainamount, this will result in a proportional decrease in the CPU time itreceives. In actuality, it will receive just as much time as before, aslong as it is higher in priority than other threads. For example, a computebound thread could be diminished from priority 31 to 16, and as long as allother threads are at 15 or below, it will receive just as much time at 16as at 31, and its system impact will be just as great.
The scheduler attempts to minimize this situation for threads in thevariable class, which ranges from 1 to 15. However, highly compute-boundthreads can still degrade overall system responsiveness in somecircumstances. You can usually investigate these situations by usingPerformance Monitor to identify which threads are consuming CPU time. Theninspect the dynamic thread priority of the compute-bound thread withrespect to that of other slowly-responding threads. You will usually seethe compute-bound thread is equal or higher in priority than the others.
Schedulers of this type work remarkably well over a wide variety ofsituations, are well understood, and have low overhead. However, especiallywith compute-bound processes, they have limitations that are difficult toovercome. In these situations, the compute-bound process tends to eitherget all the available CPU time, or little to none. It is very hard tothrottle a compute-bound process so that it runs at a decent rate yet doesnot dominate the system.
An item that controls thread priority is the "tasking" option of theControl Panel System applet. This allows you to control the degree offoreground priority boost that normal priority class applications receive.A foreground application is one that is made active by selecting it on theNT desktop, thus, bringing it to the foreground. All other applicationsrunning are then termed background applications with respect to theforeground.
The Windows NT default is "Best Foreground Application Response Time,"which results in a foreground application's priority being increased twolevels.
Neither foreground boost nor the heuristic priority adjustmentwill cause an application to change classes from normal to realtime, orvice versa. The priority adjustment from these sources is bounded by theprocess class.
The intermediate Control Panel tasking option is "foreground applicationmore responsive than background," which equates to a foreground boost ofone level.
The last Control Panel tasking option is "foreground and backgroundapplications equally responsive," which deactivates any foreground boost.The SQL NT setup program selects this setting during installation. This ismainly to prevent the database server from being CPU-starved by foregroundapplications and is the best setting for a dedicated SQL Server.
The SQL Server setup program allows two options that affect the priority atwhich SQL runs, hence, system behavior. These two options are listedunder the "Set server options" option of setup and are called "DedicatedMultiprocessor Performance" and "Boost SQL Server Priority."
Selecting Dedicated Multiprocessor Performance on a symmetricmultiprocessor (SMP) machine increases the scalability improvement thatmultiple CPUs have on SQL Server's performance. Selecting this option isnot necessary for SQL Server to benefit from multiple CPUs, but it doesincrease the amount of improvement. A side effect of selecting the optionis that it causes SQL Server to run at a priority of 13. This helpsminimize scheduling overhead and obtain maximum benefit from SMP.
Selecting "Boost SQL Server Priority" is possible on either a uniprocessoror SMP machine. When selected on a uniprocessor, SQL Server runs atpriority 13, which is high but within the variable class of priorities thatrange from 1 through 15. When selected on a SMP machine, SQL Server runs atpriority 24, which is midway into the realtime class of priorities thatrange from 16 through 31.
This chart displays the priority at which SQL Server for Windows NT runsdepending on configuration:
Priority Multiprocessor (MP) Support Boost SMP Machine
7 n/a OFF No 13 n/a ON No 7 OFF OFF Yes 24 OFF ON Yes 13 ON OFF Yes 24 ON ON Yes
With the previous in mind, a number of behaviors concerning SQL Server andthe Windows NT thread scheduler are more understandable. One of theseconcerns when SQL Server for Windows NT is performing a CPU-intensiveoperation, such as some types of joins. Depending on the exact operationand the amount of memory available, the operation could be done entirely incache--effectively becoming CPU-bound. During this period SQL Server, evenwhen running at the default priority of 7, could degrade interactiveperformance. The conditions necessary for this behavior to occur arerelatively narrow and do not happen frequently.
Another situation where this might occur is if a large single transactionor a large number of smaller transactions needed processing during start uprecovery, as would be the case if the server was abruptly shut down whilethese transactions were in progress. Depending on the amount of SQL cachebuffer memory and the exact disk subsystem used, a lengthy recovery cantake place mostly or entirely in cache. Because of this it would be CPUbound and could degrade interactive performance during this interval.
On a machine dedicated mainly as a database server, interactive consoleperformance is usually not a priority. However, if necessary, as aworkaround SQL Server can be started from the command line in the IDLEpriority class with a command like:
start /low sqlservr -c -dc:\sql\data\master.dat
Another example is when SQL Server does a large disk init. In this case theoperation is extremely I/O bound, and the lack of any scheduler I/O quotascauses similar behavior to a CPU bound operation.
Using the Dedicated MultiProcessor Support option when on an SMP machinededicated to SQL Server can improve performance. This will also result inSQL Server running at priority 13. In some CPU-intensive operations, thiscan cause logins to be slower (or the Windows NT console responsiveness tobe sluggish) as SQL Server may allow less CPU time for other systemprocesses.
Using the Boost SQL Server Priority on a SMP machine dedicated to SQLServer can also improve performance, although often not as much as the MPsupport option. Boost Priority results in SQL Server running at priority24, which is in the realtime class. The impact on Windows NT consoleresponsiveness can be similar to that of the MP support option.
Additional Suggested Reading
"Inside Windows NT" by Helen Custer, ISBN 1-55615-481-X"Advanced Windows NT" by Jeffrey Richter, ISBN 1-55615-567-0"Optimizing Windows NT" by Russ Blake, ISBN 1-55615-619-7 (Vol. 3 of the Windows NT Resource Kit)"Windows NT SDK," Microsoft Developer Network CD-ROM"Operating Systems, Design and Implementation" by Andrew S. Tannenbaum, ISBN 0136374069"Modern Operating Systems" by Andrew S. Tannenbaum, ISBN 0135881870Microsoft Knowledge Base article Q96418