Superscaling SQL Server Singleton Inserts

Chris Adkin

This session covers tuning what is superficially the simple exercise of getting the maximum throughput out of a concurrent insert workload, from key construction, logging to tuning spinlocks. The session will look at two of the key spinlocks which influence of OLTP workload performance, namely XDESMGR, LOGCACHE_ACCESS and how they influence OLTP workload performance and scalability in great depth, the ramifications of NUMA on spinlock behaviour and SQL OS scheduling. Insights gained from windows performance toolkit will be used to provide a level 4~500 view into how the database engine behaves under extreme OLTP pressure. Finally the session will wrap up with the in memory OLTP engine and the performance of natively compiled stored procedures versus transactions that span the conventional disk based database engine and the in memory OLTP engine.