The Planet Blog

“What counters are important when troubleshooting SQL Server with Windows’ Perfmon (Performance Monitor)?”

As a database administrator architect, I can’t tell you how many times I’ve gotten that question from Windows administrators and new database administrators, and it’s a pretty daunting question to answer. When a SQL instance is installed on a server, there may be more than a thousand new performance counters added to Perfmon and to the Performance Monitor section of the Reliability and Performance Monitor in Windows 2008, so where does one start?

When it comes to performance tuning and troubleshooting, SQL Server has not changed much over the years. In a nutshell, it is all about CPU, memory and I/O. To reduce the time it takes to determine the bottleneck in a given environment, we should isolate the counters to these specific areas.

Of the many performance counters that can be selected when troubleshooting a SQL Server, these key indicators can help to quickly isolate bottlenecks and direct your investigation to the appropriate resources for corrective action:

At the Operating System Level

CPU Counter

Server Work Queues > Queue Lengths
Description: Queue Length is the current length of the server work queue for this CPU.
Value: A sustained queue length greater than twice the number of CPU cores might indicate processor congestion.

Memory Counter

Memory > Pages/Sec
Description: A page fault occurs when the operating system (OS) cannot find the requested information in its physical memory, forcing the OS to seek the information at the disk level. A soft page fault is when a page is found elsewhere in the physical memory, and a hard fault requires disk access. Most processors can handle large numbers of soft faults without significant consequence. However, hard faults, which require disk access, can cause significant delays.
Value: This value should stay below 20-25 pages per second.

I/O Counters

Physical Disk > Average Read Queue Length
Physical Disk > Average Write Queue Length
Description: Avg. Disk Queue Length is the average number of read or write requests that were queued for the selected disk during the sample interval.
Value: The value for this counter should always be under 2. This is the most reliable counter to use when the SQL instance is using external SAN storage for its databases.

Determining Application Resource Usage

Process Counter:

From Perfmon, select “Process” then select the application in question – in this case, sqlservr.

This will add all of the operating system counters associated with the application.

For SQL Server, the counters usually reflect information about the internal operations of the SQL instance that may not show up as a problem at the operating system level. For example, a customer might call to report that the server hangs for several moments throughout the day, but there are no errors in the application or system logs, and the OS Performance counters show all counters within normal limits.

These SQL performance counters can provide deeper insight into how the database engine is working against the customer’s application to help in your investigation.

SQL Server

SQL Server: General Statistics > User Connections
Description: Number of users connected to the system.
Value: This number will vary between applications on a given server.

SQL Server

SQL Server: Locks > Lock Waits/Sec
Description: Number of lock requests that could not be satisfied immediately and required the caller to wait before being granted the lock.
Value: This number should be near 0. If the value is greater than 0 for a sustained period, the “system hanging” as described by the customer is most likely due to locking and blocking issues within the database.

SQL Server

SQL Server: Buffer Manager > Buffer cache hit ratio
Description: The percentage of database pages that were found in physical memory. SQL Server stores information inside of 8KB data pages.
Value: The buffer pool cache hit ratio should be 100%. Values below 90% indicate that SQL Server is experiencing memory pressure.

SQL Server

SQL Server: Buffer Manager > Page life expectancy
Description: The number of seconds that a database page remains in physical memory.
Value: Values under 300 seconds indicate that the SQL instance is experiencing memory pressure.

SQL Server

SQL Server: Buffer Manager > Page reads/sec
SQL Server: Buffer Manager > Page writes/sec
Description: These values show the number of database reads and writes requested.
Value: Values for these counters will vary between database applications, but this information is useful when determining if SQL Server is the primary application using the disk. If the Buffer Manager page read-writes are low but disk-queue lengths are high, there might be a disk bottleneck. If the Page read-writes are higher than normal, a memory shortage is likely to exist.

SQL Troubleshooting Summary Chart

The performance counter suggested may not help resolve complex situations that involve application-level performance issues, but in my experience, it is a great starting point.

-George

And boy do we know how to throw a party! Well, at least our customers do. New Year’s Eve Nation, a new startup company out of New York, plans for one event 364 days out of the year: a national New Year’s Eve party spanning four time zones and eight cities. That’s one hell of a party.

Ex Spin City producer-turned-entrepreneur Jann Yogman has been throwing New Year’s Eve bashes in Manhattan since 1997 and is the genius behind the event. Last year, he decided to go national with his soirée and did just that using live feeds via the internet. Yogman connected five cities and allowed party goers to interact with each other via text messaging and video — all of which is broadcast on plasma screens at each location.

Last year, there were five cities. This year, it was bumped to eight, including New York, Los Angeles, Denver, Dallas, Seattle, Philadelphia, Boston and Detroit. 8,000 people attended the party billed as “8 Cities. 4 Timezones. 1 Outrageous Party.”

So what hosting company with world-class datacenters and the most reliable network powered the event?

That’s right.

Not only did this party welcome in 2008 with a serious bang, it also gives back by donating 10% of the net profits to Micheal J. Fox’s Foundation for Parkinson’s Research.

You can read more about the event and the goals for the future on CNN.com or the New Year’s Eve Nation website.

Happy New Year, Y’all!

-Ali