# Tuesday, October 19, 2010

Some time ago I was working with performance data collected by a customer to look for I/O-patterns and make recommendations for the I/O-subsystem for a new SQL Server 2008 machine that would replace a couple of SQL Server 2000 servers. Unfortunately I ran into issues with the values for Avg. Disk sec/Read and Avg. Disk sec/Write, which were ridiculous with average values around 100-million seconds per read or write. Those values occurred on two virtualized Windows 2000 Servers.

Today, I was working for another customer. Physical Windows Server 2003 boxes (both nodes in the same cluster). I imported the .blg files into a database and was looking at very high values (up to 100 million) for Avg. Disk sec/Read and Avg. Disk sec/Write. I consulted the staff-DBA, who referred me to one off the Windows administrators. So the Windows administrator logged in, opened performance monitor and… got reasonable values for those counters (hovering at about 0.006, I’d even say those values are good). To prove I wasn’t making a fool of myself, we walked over to my desk and I showed the results I had… same servers, same counters, very different values. Next we tried using the account of the staff-DBA, which showed the “normal” values.

Simply eliminating, the Win-admin said the only difference is the account, so he went and checked whether the account I used had full administrator permissions. It had. Next he turned his attention to the user profile for the account I used (which happened to be quite big). The user profile belonging to the account I used was deleted and I logged in again, this time to be shown the correct values. RESOLVED!!!

Unfortunately the collected data still was useless. I started the counter log again, let it run for about 10 minutes and examined it… sky high values!!! In disbelief I switched to current activity, to be shown normal values!!! Next I created a new counter log definition, let it run for a couple of minutes and this time I got normal values.

Bottom line, to resolve insanely high values for Avg. Disk sec/Read and Avg. Disk sec/Write (it might apply for other counters of the type PERF_AVERAGE_TIMER too), use a fresh user profile and a fresh counter log definition.

Thanks Ruud.

Comments are closed.