The Definitive Guide to SQL Server Performance Optimization

Chapter 1There are two methods for scaling, or growing, a solution. The first method is the one we tend tothink of first—buy bigger hardware. This technique is called scaling up, because it simply meansupgrading your existing hardware to support more capacity. SQL Server works well in scale-upsituations, but scaling up has an upper limit. Most servers are built to accommodate only acertain number of processors, a certain amount of memory, and so forth. After you’ve scaled upas far as you can go, you’re stuck. Even if you’re willing to buy brand-new hardware fromscratch, scaling up will only take you so far because you can only buy so big of a server at anygiven time.The other method of scaling is scaling out (adding more servers to the solution). Each new serverperforms the same tasks as the original server. Web farms are the classic example of scale-outarchitecture, in which you simply keep adding Web servers (all hosting the same content) untilyou’ve got enough of them to handle the number of users attempting to reach your Web site.Contrary to popular belief, SQL Server can scale out, although doing so is definitely much moredifficult than scaling out a Web farm. In SQL Server, scaling out means using federateddatabases, replication, linked servers, and other techniques that allow multiple SQL Servercomputers to look like a single server, even if they’re not hosting identical copies of thedatabase. Chapter 6 focuses on scaling SQL Server and includes an explanation of federated databases,distributed databases, replication, linked servers, and much more.Scaling can sometimes be the easiest solution to performance problems. If your current SQLServer system can’t handle your workload, either upgrade it (scale up) or add another one (scaleout) to help handle the load. Scaling doesn’t require any fancy query tuning, complexperformance-troubleshooting procedures, or any of that stuff. Unfortunately, scaling—especiallywith SQL Server—is almost always expensive because it involves buying more server hardware.In the case of SQL Server, scaling out can also mean a lot of additional administrative overhead,as you’re adding more SQL Server computers that you’ll have to maintain and troubleshoot. Soalthough scaling is often the easiest solution to performance problems, it isn’t always the bestsolution. The best solution is often to make more efficient use of the hardware you’ve got, whichmeans digging in and fine-tuning everything. Fortunately, that’s what the rest of this book is allabout!Holistic PerformanceAlways remember to look at the big picture when it comes to performance. Many administratorsfall into what I call the performance trap—looking for the one little detail that, once fixed, willnoticeably improve performance every time. Other folks call it the magic bullet, which, whenfired, will kill your performance problems. Unfortunately, there is no such thing. Modern serversare systems, which means they consist of several interlocking subsystems that all work together.Changing one subsystem might simply shift a performance problem to another subsystem. SQLServer is a system, too, with memory caches, a query optimizer, database statistics, and muchmore. SQL Server interacts closely with both the Windows OS and your server hardware, sochanging any particular performance component is bound to have an effect somewhere else.19