I have been involved with a few escalations recently (I don't even work in support!), the cases have been mostly due to Jetstress tests failing – typically then the relationship between customer and storage vendor becomes more hostile and eventually someone points the finger at Jetstress or Exchange… this is often where I get involved
A recent trend has been to blame Jetstress for not taking passive workload into account i.e Jetstress has no concept of passive databases and treats all databases as active. The theory goes that since a passive database generate much less IOPS workload than an active database, testing them as if they were all active is overkill.
I generally respond to this by reminding them that we need to validate that every database copy that could become active has sufficient storage performance should that event arise, we validate this by testing it in its active state.
Having said all of that I decided to test to see what the difference actually is between active and passive database workload requirements. To do this I used a 3 node DAG in my test lab with a single database configured. I made MBX2 the active copy of the database and MBX1 and MBX3 replica copies. I then configured a loadgen test to generate some workload and observed the results…
You can clearly see here that the IOPS workload on all of the DAG nodes is similar – certainly it would not be a viable proposition to suggest that replica database copies always require significantly lower IOPS than active?
Food for thought…? but the moral of the story is simple… if you are going to go to the expense of configuring a database copy, make sure that it has sufficient storage performance
One point on the passive database IOPS, do we have a number on how much is the IOPS requirement of a passive db in comparison of Active DB. I believe with Exchange 2007 this use to be 40% of the active database copy.
If we have 60 databases on a member server of a DAG with 4 copies, there will never be a situation that a server will run with 60 active copies. This DAG has 6 servers, 3 local 3 remote and 2 copies are in each site.
In the above scenario surely we shouldn't be running Jetstrees against 60 databases?