Slow Response on one node of an active/passive SQL Cluster

Problem with response time depending on which node of an active/passive
cluster is active.

If checked everything I can think of or Google suggested.

What have I missed?


Win2k3 SP1 SQL 2000 SP3a
Hardware is the same on both nodes including memory (5 GB)
pagefile.sys is the same size on both nodes
Settings for Properties | Advanced | Performance are the same on both nodes
Large System Cache is 0 on both nodes
“fast” node has /3GB switch
I’ve tried the “slow” node both with and without the /3GB switch – same
results – slow
Have node set /PAE switch and AWE since I have one node in the cluster with
only 3 GB of memory.


The only thing I see different is the value of the System cache on the Task
Manager Performance tab. The “fast” node has .5 GB when it owns the
resources. The “slow” node has < 200 MB when it owns the resources.

You said, "Hardware is the same on both nodes including memory (5 GB)."

But then, you say, "Have node set /PAE switch and AWE since I have one node
in the cluster with
only 3 GB of memory."

Can you explain this a little better?

As far a performance between two nodes, if they are identical hardware, then
you need to go through the complete configuration, OS patches, hardware
driver versions, firmware revisions, and configuration parameters in order
to make sure they are the same.

Do not fool yourself. This is not an easy task. You've already noted a
difference in System Cache sizes on each node, which would indicate that
each platform is not the same configuration.

Also keep in mind that SQL Server does not use the System Cache for the
Buffer Pool; however, Disk and Network I/O will. Both systems should be set
to Size = 2
(HKLM\SYSTEM\CurrentControlSet\Services\LANMANServer\Parameters).

This is the Network Performance Options equivalent to Maximize Throughput
for Network Services. It will create a larger System Cache for I/O.

This is NOT the same as the File Server optimization LargeSystemCache
(HKLM\SYSTEM\CurrentControlSet\Control\Session Manager\Memory Management),
which should be set to 0.

Be careful of using the /3GB switch, especially when also using the /PAE
switch. You can get an additional 1 GB of active Buffer Pool/MEMTOLEAVE
space for SQL Server, but at the expense of System (kernel) memory. Pay
particular attention to the Memory:Free System Page Table Entries
performance counters. The system should maintain between 24,000 and 26,000
free System PTEs on average.

If you find that your system is dropping lower than these values, take a
look at using the /USERVA switch in conjunction with the /3GB switch. This
new switch available to Win2K3 allows you finer control over the User Mode
VAS between 2 GB and 3 GB than the /3GB switch alone.

Finally, investigate and consider running RAM in quantities equivalent to
powers of 2 increments: 2, 4, 8, 16, 32, 64, etc, given how memory
allocations are managed, allocated, and mapped.

Best of luck.

Sincerely,


Anthony Thomas


--

"Pat Hall" wrote in message

> Problem with response time depending on which node of an active/passive
> cluster is active.
>
> If checked everything I can think of or Google suggested.
>
> What have I missed?
>
>
> Win2k3 SP1 SQL 2000 SP3a
> Hardware is the same on both nodes including memory (5 GB)
> pagefile.sys is the same size on both nodes
> Settings for Properties | Advanced | Performance are the same on both
nodes
> Large System Cache is 0 on both nodes
> "fast" node has /3GB switch
> I've tried the "slow" node both with and without the /3GB switch - same
> results - slow
> Have node set /PAE switch and AWE since I have one node in the cluster
with
> only 3 GB of memory.
>
>
> The only thing I see different is the value of the System cache on the
Task
> Manager Performance tab. The "fast" node has .5 GB when it owns the
> resources. The "slow" node has < 200 MB when it owns the resources.
>