January 23, 2011

Monitoring SQL server resources

While investigating performance issues related to SQL server resources, first identify bottleneck and then resolve that bottleneck.

1. Memory bottleneck:
SQL server uses all the memory of hosted server. Data pages resided in disk, but disks are slow , so SQL server load all the data pages into memory so that they are provided to users as quickly as possible when requested.
There are 2 types of Memory pressure.
i) External memory pressure: If sql server unable to acquire sufficient memory.
If SQLServer: Buffer manager – Total pages and SQLServer: Buffer manager – Target pages are same, SQL server has sufficient memory else if target is greater than total , then there is external memory pressure.
ii) Internal memory pressure: If sql server buffer pool( the memory location which holds data pages) is reduced in response to other processes like linked server, extended SPs, etc.

• Memory counter: Available bytes < 4 MB indicates there is memory pressure.
• VAS ( virtual address space) is made up of physical memory and memory provided by page file. If memory pages/second > 1, indicates there is memory pressure.
• SQL server: Buffer Manager – Page life expectancy < 300, then there is memory pressure
• SQL server: Buffer Manager- Buffer cache hit ratio reports the no. of pages requested by query found in sql server buffer. If it is <90, requested pages are mostly read from disk to memory and it is memory bottleneck.
• SQL server: Buffer Manager – stoles pages if greater than total target pages, there is memory pressure.
• SQL server: Buffer Manager Memory grants pending if non-zero, there is memory pressure.
• SQL server: Buffer Manager- Lazy writes- This records no. of times per second sql server relocates dirty pages from buffer pool to disk. If it is 20sec or more, then there is memory pressure.


2. CPU bottleneck:
• If % Total processor time is consistently >80, , there is CPU pressure
• If avg physical queue length > 2, there is CPU pressure
• If SQL server: SQL statastics- SQL compilations or Recompilations are more, there is CPU pressure

3. Disk bottleneck:
• If %disk time is >20ms, there is disk pressure
• If SQL server: Access Methods- Full scans > 1 second or high Page splits, there is disk pressure.
• If % Total processor time is consistently >80, there is disk pressure.

No comments:

Creating DataFrames from CSV in Apache Spark

 from pyspark.sql import SparkSession spark = SparkSession.builder.appName("CSV Example").getOrCreate() sc = spark.sparkContext Sp...