Customer Focused Testing: Ten steps to improved SQL Server 2005 replication (5697-7434, March 2008)

In testing, the SQL Server behaved in a predictable fashion and used all but 500 MB of available system
memory (23.5 GB). If other programs or applications are installed on the server, consider placing an
upper limit on the SQL Server’s maximum server memory setting.
In an asynchronous SQL Server database mirroring configuration, after a local cluster node failover,
limit client access to the Principal database in order to enable the databases to synchronize once the
Principal is back online.
Storage administrators
When impleme
nting Continuous Access, place all LUNs associated with a database in one DR group
residing on one owner-controller. Because this might unbalance your EVA controller load, consider
assigning the majority of the remaining LUNs hosted by the EVA to the controller with the least load.
Consider defining the write history log size and destination DR group when configuring Continuous
Access to guarantee that space will be available if the ISL is dropped for a long period of time.
In a synchronous Continuous Access configuration, after there has been an outage on the ISL and the ISL
link has been reinitiated, limit client activity when conducting replication. It is likely that transactional
performance will be affected once the replication has been reestablished and is in a state of merging.
In an asynchronous Continuous Access configuration, loss of the source storage system when replicating
in asynchr
onous mode is likely to result in data loss. Manual intervention will be required to bring the
database
back online on the destination site because automated failover is not supported.
In an asynchronous Continuous Access configuration, a majority node set based cluster can be configured
tocontinuetorunintheeventofastoragesystemfailurebynotsettingthequorumresourcetobe
reliant upon shared storage.
Network administrator
SQL Server database mirroring seems to be more sensitive to latency on the ISL than the Continuous
Access solution, especially under a heavy workload. If an ISL has latencies above 4 ms, consider using
array-based replication instead of database mirroring. Alternatively, experiment with adding more
server memory to the Principal.
If using array-based, block-level replication for a heavy SQL Server workload, an OC-3 bandwidth
may not be sufficientwheninsynchronousmode.
From a performance standpoint, there seems to be no difference whether the SQL Server database
mirroring solution uses an OC-3 or an OC-12 bandwidth.
If an ISL outage occurs, closely monitor the resynchronization mechanisms of all the solutions because this
can have a significant impact on both the performance and the resilience of your database environment
after the link is reestablished.
If the ISL latency is below 6 ms, HP recommends synchronous replication for Continuous Access because
asynchronous mode may show some performance degradation due to the double caching mechanism
being exposed at low latencies.
Continuous access may use more ISL bandwidth in asynchronous mode than in synchronous mode,
especially in low-latency scenarios.
In a Continuous Access solution environment, after an ISL outage, the replication relationship enters a
merging state. The impact on SQL Server transactional performance is less in asynchronous mode than in
synchronous mode during the merging period.
If there is an ISL outage and latency on the link is high (>15 ms), monitor the write history log growth
rate both during and after the outage.
NOTE:
A full site copy will occur if the write history log fills, so management of the incoming requests may be
necessary until the write history log is depleted.
Customer Focused Testing: Ten steps to improved SQL Server 2005 replication
7