Database Journal
MS SQL Oracle DB2 Access MySQL PostgreSQL Sybase PHP SQL Etc SQL Scripts & Samples Tips Database Forum

» Database Journal Home
» Database Articles
» Database Tutorials
MS Access
SQL Scripts & Samples
» Database Forum
» Slideshows
Free Newsletters:

News Via RSS Feed

Database Journal |DBA Support |SQLCourse |SQLCourse2

Featured Database Articles

SQL etc

Posted Dec 19, 1998

Microsoft SQL Server performance monitoring with Microsoft Windows NT Server Performance Monitor - Page 2

By Alexzander Nepomnjashiy

Choosing Performance Monitoring tool

Microsoft SQL Server, along with Microsoft Windows NT, represents an extensive toolkit for monitoring. The choice of which concrete tools/utilities to use is determined first of all by which tasks the administrator faces. Microsoft SQL Server 7.0 offers the following monitoring tools:

  • SQL Server Enterprise Manager
  • SQL Server Profiler
  • Stored procedures
  • Transact-SQL commands

In turn, the operating system Microsoft Windows NT features an additional set of monitoring tools:

  • Windows NT Performance Monitor
  • Windows NT Task Manager
  • Windows NT Event Viewer

As an alternative to the firmware monitoring utilities, you may use applications supplied by 'third-party' companies that support the SNMP protocol as well as SQL Server 7.0.

Server monitoring with Microsoft Windows NT Performance Monitor

Performance Monitor is used to observe the operation of the server system, as well as all events happening on that system. Both the computer on which Performance Monitor is run locally, and a remote computer (running Microsoft Windows NT) may be included in the role of the system being monitored. It is possible for a centralized administrator to monitor operation productivity of both the remote server and any server in the local network without leaving his workstation. Remember, at the execution of the monitoring, there will be an increased load on the network that may affect its capacity.

Performance Monitor is a Windows NT 4.0 Administrative Tool for monitoring the performance of Windows NT workstations and servers. It uses a series of counters to track data, such as the number of processes waiting for disk time, the number of network packets transmitted per second, and the percentage of processor utilization. You can watch this data in real time, log it for future study, use it in charts and reports, and set alerts to warn you when a threshold value is exceeded.

Processor:% Processor Time shows CPU loading. In an SMP environment, (having more then 1 system processor), the administrator can observe how busy every processor is. In addition, the same parameter can be used for the definition of processor utilization by each Windows NT thread. If Processor:% Processor Time counter shows an average value in a range from 80 up to 100% (and hard drive and network related counter values are low), consider either a faster processor or an additional processor.

Note that short-term levels of the counter Processor:% Processor Time of 80%, or even peaks (up to 100%) are not necessarily a sign of productivity problems with a processor subsystem. Since operational resourses of the processor are used not only by the database server, you should define whether SQL Server is the cause of high utilization of CPU time. Use SQLServer:CPUtime to define share of SQL server in Processor:% Processor Time. Having determined that the cause of increased load is SQL server, and also which process provokes it, you should analyze the project of query executed at this time. Make sure that indexes are used optimally in queries.

The query may be cached well, yet at the same time overload the I/O subsystem, causing exessive CPU cycles. It may mean that when designing the table the index was not optimally planned. If the query plan is optimal, it is possible to troubleshoot the problem by scaling; for example, add processors or install a more productive processor. On the other hand, if a value of Processor:% Processor Time counter is constantly low--it is not nesessarily good. For example, if the CPU graph is consistently low (less than 50 percent), and the %Disk Time is consistently high, this can indicate an I/O bound state.

Processor:% Privileged Time is convenient for determining excessive I/O loading. If the average value exceeds 20%, and Processor:% Processor Time is much lower than 80%, it means that SQL Server is putting a strain on a subsystem of input-output. You need to analyze the database design, workload on RAID--controller and/or the network card.

System:Processor Queue - The number of tasks waiting for processor time. If this value is consistently two or more, it means that the central processing unit is overloaded. Additional processor power is necessary to solve this problem.

System:Context Switches/sec - is the rate at which the processor is switched from one thread to another, and is an indication of which threads are getting processor time.

Process:Thread Count - number of active threads. The value for this counter along with System:Context Switches/sec can be used for optimal configuration of SQL server to reduce excessive CPU utilization.

Process:Virtual Bytes - determines what memory volume uses SQL server and which applications do not use it effectively.

Process:Working Set - memory size used by the process. The modification of SQL server configuration after the analysis of these counters will allow optimizing a memory allocation between a database server, an operating system and other applications of a database server.

SQLServer:Cache Hit Ratio - for well balanced applications, the number of hits in the cache should be "near" 100%. Often, reaching a high level of hits in the cache is achieved by a simple extension of the RAM,(adding additional memory modules). Also, you can use a technique, such as SQL Server trace flags, for example trace flag 1081. Trace flag 1081 allows the index pages to make a "second trip" through the data cache. When SQL Server needs to flush an index page out of cache to bring a new page in, it chooses a different page unless this particular index page has already been bypassed once. Therefore, index pages are allowed to stay in the data cache longer.

And in conclusion some tips to memorize:

To optimize the allocation of SQL files of the server try to follow the following rules:
1) Always use disk arrays RAID. Their usage ensures reliability and productivity of a disk subsystem.

2) The more disks that are in the array, the higher the general productivity of the disk array. The more channels on the RAID-controller, the better the bus is loaded. The more RAID-controllers there are, the more effective the usage of the disk array. The more buses on the motherboard, the more of the above-mentioned components may be installed on the server thereby increasing productivity as a whole (if processors "cope"). Further, there will already be clusters. By the way, even in the cluster-less system try to use RAID-controllers, as they can work in a cluster (For example, do not initialize SCSI bus when power is turned on).

3) Locate files of sequential access and files of a failed access on different disk arrays (not logical disks created on one RAID the array). As a rule, logs of transactions represent files of sequential access, and files of databases - a failed access. For example, you can place LOG-files on RAID 1, and the database on RAID 5 or 10 (= RAID 1 + 0). Further, if you do not want to mix your data with system data, you may create two data files in two different file groups. The first file group will contain system objects, and the other will contain your objects. It will not cause any noticeable scoring in productivity, but will allow dividing essentially different data types. The principle of diversity of the data in file groups, according to access type to the information, can be also applied to tables. Significant management complications will be paid back with interest by the growth of productivity of the application.

4) Place files of sequential access on different disk arrays. Often the server should access several tables and/or indexes simultaneously. Optimal allocation of such files with the help of their inclusion in different file groups, (located on different disk arrays), may also positively affect the general productivity of the application.

SQL etc Archives

Latest Forum Threads
SQL etc Forum
Topic By Replies Updated
MySQL rollback UAL225 0 August 21st, 09:56 PM
Complex Search Query Galway 0 May 20th, 10:04 PM
change collation at once supercain 2 May 15th, 06:18 AM
SQL Features, tools and utilities question Neomite 1 April 10th, 09:13 AM