Monitoring the Universal Driver Linear Hash Service With Windows Performance Monitor (Performance)

Overview

This article is a step by step procedure to setup Windows Performance Monitor to monitor the Universal Driver's Linear Hash service. Performance Monitor can be a useful metric to determine the impact of the Linear Hash service on the server. Common uses include:

        Monitoring daily workload - The performance monitor can beset to silently gather performance data during the day for display at a later point in time. Use it to correlate user reports of slow performance to server activity when you can't bring up task manager fast enough.

        Comparing performance before and after - Gathering this information after installing your application on new hardware can be useful for troubleshooting performance problems later if the client adds other 3rd party software to the server.

        Display current activity - Task Manager is an incredible tool but it is hard to gather specific point in time data for active tasks. Performance monitor can be customized to display the key metrics that are of interest.

Setting Up Performance Monitor

1.     Open the Performance Monitor tool by clicking "All Programs → Administrative Tools → Performance". The window will open to the "System Monitor" node and display disk, memory, and cpu usage by default.

2.     Click plus sign on the button bar to add a counter (item to monitor).

3.     Choose the items as displayed in the screen shot below.Depending on the linear hash version your instance may be named differently.This example uses the Universal Driver 4.6 so the instance (process to monitor)is named lh46srvc. Your instance may be lhsrvc,lh30srvc,lh40srvc, etc.

4.     

5.     By default the information from the counter (i.e. linear hash service) is displayed in chart mode. For easier point in time viewing change to report view. Your screen should resemble the image below.

6.     

Analysis

The Linear Hash service memory usage is normally constant so most of the metrics associated with page file thrashing (page faults) and private working memory sets aren't useful. Drastic changes in these metrics could indicate an overloaded server.

Most of the server load is CPU and IO operations. Key indicators are:

        % Processor Time - Overall processor time consumed by the service (privileged + user)

        % Privileged Time - Processor usage fulfilling kernel mode requests for this process. Examples of kernel mode time are requests to network and disk drivers that run in kernel mode.

        % User Time - Processor usage specific to the linear hash service.

        IO (Data|Other|Read|Write)Bytes/sec - Throughput activity for the service. The "Data" is a total metric for the sum of other + read + write.

        IO (Data|Other|Read|Write)Operations/sec - Same as above except it measures individual operations instead of bytes throughput.

Please note, bytes and operations metrics cannot be translated into disk through put because of the Windows System Cache. If you convert the statistics from Bytes/sec to MB/sec and find it is over 50 MB/sec then chances are the Windows System Cache is fulfilling read/write requests. The inner workings of the Windows System Cache are beyond the scope of this document but will be covered in an upcoming article.