Surfing the Multicore Wave: Processors, Parallelism, and Performance

Today's server hardware ships with a truly massive amount of CPU power. And while SQL Server is designed to automatically and seamlessly take advantage of available processing resources, there are a number of options, settings, and patterns that can be exploited to optimize parallel processing of your data. This full-day seminar starts with an overview of CPU scheduler internals from the perspective of both Windows and the SQL Server Operating System (SQLOS). Next, you will learn about intra-query parallelism, the method that SQL Server uses to break up large tasks into many smaller pieces, so that each piece can be processed simultaneously. You will come to understand how to read parallel query plans and how to interpret parallel wait statistics, as well as best practices for the various parallelism settings within SQL Server. Finally, you will be introduced to techniques for exploiting parallelism at the query level: patterns that can help the optimizer do a better job of parallelizing your query. After attending this seminar, you will be in full control of your CPUs, able to compel your server to use every clock tick to your end users' advantage. 

This session is intended for mid-level to advanced DBAs and DB developers. 

Session Outline:

Module 1: History and Theory

• CPU evolution over time
• Heat and dissipation
• Moore’s Law
• Amdahl’s Law
• Gustafson’s Law
• How this all fits together

Module 2: Windows Process/Thread Internals

• Processes
• Threads
• Context switching
• Quantums
• Clock interrupts
• Affinity

Module 3: SQL Server Scheduler Internals

• SQLOS workers and schedulers
• Waits and queues
• Important DMVs

Module 4: Query Processor Parallelism

• Parallel iterators
• Non-parallel iterators
• Row distribution strategies
• How to read and mine data from parallel query plans

Module 5: Controlling Parallelism at the Query Level

• Ideal parallel query patterns
• Parallelism inhibitors and workarounds
• The mathematics of "super-scaling" parallel queries

Module 6: Server Settings that Influence Parallelism

• Affinity masks
• Worker threads
• MAXDOP
• Cost threshold
• Resource Governor
• OLTP vs. OLAP vs. mixed workload considerations

Module 7: Monitoring Parallel Processes

• Parallel task architecture and the Tasks DMV
• Parallel waits and the Waiting Tasks DMV
  • CXPACKET
  • EXECSYNC
  • LATCH_EX
• Finding out how much work is being done
• Finding parallel and serial plans in the cache

Adam Machanic is a Boston-based SQL Server developer, writer, and speaker. He focuses on large-scale data warehouse performance and development, and is author of the award-winning SQL Server monitoring stored procedure, sp_WhoIsActive. Adam has written for numerous web sites and magazines, including SQLblog, Simple Talk, Search SQL Server, SQL Server Professional, CoDe, and VSJ. He has also contributed to several books on SQL Server, including "SQL Server 2008 Internals" (Microsoft Press, 2009) and "Expert SQL Server 2005 Development" (Apress, 2007). Adam regularly speaks at conferences and training events on a variety of SQL Server topics. He is a Microsoft Most Valuable Professional (MVP) for SQL Server, a Microsoft Certified IT Professional (MCITP), and an alumnus of the INETA North American Speakers Bureau.