Failover takes to long...

[Ebzed]

Hi All...,
I installed Windows 2000 Advanced Server and Microsoft SQL 2000 Enterprise over
a Compaq CL380 Cluster... everything looks nice... clean install, no problems.

But... (There is always a but...)
I am doing a small test, using a 2nd pc which is pinging the SQL virtual server,
pings is good..., the problem is when i turn off one of the nodes, the ping
stops for 30 secs, and then continue (i assume this is the time it took to
the cluster to understand and do failover).
my little question is if i can reduce this 30sec time to less, maybe 10 or
even less.

Any ideas?

Thanks,
Ebzed.

 

[YO123]

What are your settings for the resources in the cluster group.
Are they all set correctly ??

 

[Ebzed]

Basically, its the default settings...,
So if i look at the Cluster IP properties in the Advanced settings
the Cluster IP is set on restart, the affect the group thershold is
set to 3, period is on 900.
Look alive pool is 5000 and is alive pool is 60000
pending timeout is set to 180....
this is the same fro all the items on the cluster and sql group
beside the msdtc in the cluster group where the looks alive is 60000
and the is alive is 120000.

This is my first cluster... at the begining i tought its not normal, but
some people told me that the 30sec is quite normal... i dont really think
its suppose to be 30sec... i bet it can be much lower... any help will be
very appricated...

Thanks In Advance...

 

[ShackDaddy]

My cluster with SQL 2000 on a CL380 takes up to 42 seconds to fail-over. Some of this is due to the complexity of our SQL environment. I don't think there's anything wrong with 30 seconds.

ShackDaddy

 

[wjense]

Cluster services always take time to fail over...initial delay is caused by your heart-beat connection...the second node waits for several non-responses before assuming the cluster...then services have to be brought gracefully online. if 30 seconds is all it's taking, feel lucky. An Exchange failover takes up to 20 minutes.

You've got to remember that cluster services is high availalbility, not Fault Tolerant. In a Fault Tolerant system (such as Novells SFT), there is no delay whatsoever to the user...pickup is virtually instantaneous. But Cluster services must wait, for the heartbeat, then the resumption of service.