HiI work on an instance where there 30 databases and all get synchronized from one main database - provider db.An import web service imports data in the provider db every second. Databases are all small, about 4GB size, recovery model is simple for all. The application uses snapshot isolation level, and tempdb is used massively. There are 13k trns(inserts,updates,merges)/sec on the instance.Tempdb is configured according to the recommendations for a busy tempdb. However, only one disk (raid 10) is used for all databases and the OS. RAM is 128GB, and the cpu is 12 cores. Avg CPU usage is about 40%. We use standard edition. Writes are 1000/sec and reads are 100/sec. PLE is very high. There is a high number of latch waits, and XTS_MGR dominates regarding the collisions. Indexing strategy is very good, not to many not to less, and queries are all covered. I put attention on the implicit conversion for queries. Queries are mainly parametrized.All of the best practices for configuration of sql server instances and databases are applied.I wonder if we install two additional ssd disks (128GB each), how would the spinlocks decrease so I don't have too much of them and loose deadlocks? Deadlocks are not very high, but happen, let's say some tens per hour.I plan to use the ssd drives for the tempdb, busy indexes and log files. Should I expect SSDs will help significantly?I appreciate your experiences, suggestions and advises...Thanks,Igor
↧