SQL Server 7.0 Clustering

Tip: If you are going to implement replication from a SQL Server 7.0 or 2000 clustered server, and the SQL Server cluster will participate as a Publisher and a Distributor, use a file share located on the cluster disk resource as the snapshot folder. This way, replication will failover when SQL Server fails over.

Explanation: When you configure a clustered SQL Server as a Distributor, SQL Server needs to have access to a snapshot folder as a temporary holding place during the replication process. In order to ensure that replication still works when failover occurs, this folder must be located on a shared folder on the cluster’s shared disk resource. If it is not, then when failover occurs, replication may stop working.

You will have to manually create the required folder on the shared disk resource, and additionally create the necessary share with appropriate permissions. In addition, you will have to configure the shared folder using Cluster administrator as a clustered shared folder.

If you don’t want to do the above, one option is to not make the clustered SQL Server a Distributor, but only a Publisher, and locate the Distributor on a non-clustered SQL Server. This way, the snapshot folder will exit on a non-clustered SQL Server, negating the above advice.

Version: 7.0, 2000

Date Added: 5-31-2001

*****

Tip: You have heard this advice before, but it is so important I am repeating it here, be sure all of your clustering hardware is on the Windows 2000 Advanced Server Clustering Hardware Compatibility List (HCL).

Explanation: I can’t overemphasize the importance of ensuring your hardware is supported by Windows 2000 Advanced Server clustering. If it isn’t, not only will you have trouble getting support from Microsoft, the nature of the problems you probably will have will be excruciating painful and hard to troubleshoot. I know, I have been there.

What makes it difficult to ensure that all of your hardware is on the HCL is Microsoft’s poor organization of the HCL itself. It is often very difficult, and time consuming, to locate all the hardware on the list to see if it has been certified.

A further complication is that Microsoft certifies hardware two different ways. First, it certifies some hardware by specific part. Second, in other cases, hardware is certified as a total system (a specific combination of parts). And in some cases, hardware certified one way is not certified another way. I recommend that you try to find hardware that is certified as a total system, that way, you can ensure that the total system you are purchasing has been tested as a total system by Microsoft.

Other problems with the HCL is that it is often out of date, sometimes parts that are still currently available (but are older models) are not listed in the HCL, and sometimes one variation of a part is certified, but a slightly different variation of the same part is not certified. So be very careful when comparing part numbers to the HCL to be sure you and the HCL are talking about the exact same part, not some obscure variation.

Version: 7.0, 2000

Date Added: 6-4-2001

*****

Tip: To ensure failover will work when needed, it is important to schedule manual failovers periodically in order to test if failover is really working.

Explanation: Just because your cluster has not had any problems you are aware of doesn’t mean that there are no problems. Periodically, at least every 60 days, you need to manually failover your cluster, and fail it back, to ensure failover is working as expected. This helps to ensure failover is working should it ever be needed for a real server failure.

Version: 7.0, 2000

Date Added: 6-6-2001

*****

Tip: Don’t run the regrebld utility, which are used to rebuild SQL Server’s registry, on a clustered SQL Server.

Explanation: The SQL Server clustering wizard alters the registry entries of SQL Server when it is run. If you try to rebuild the registry while SQL Server is clustered, it will remove important registry entries, and most likely, SQL Server will stop functioning.

Version: 7.0, 2000

Date Added: 6-6-2001

*****

Tip: Maximizing the hardware performance of a SQL Server cluster is very similar to maximizing the hardware performance of a non-clustered SQL Server.

Explanation: When specing out new hardware for a SQL Server cluster, assuming you want maximum performance, you may want to consider purchasing the following “minimal” hardware configuration:

Minimum four CPUs, each with a 2MB L2 cache.
4GB RAM (ECC).
Fibre Channel/RAID controller with enough channels to maximize I/O.
The SQL Server executables (and operating system) should be placed on mirrored drives (RAID 1) and have its own controller channel.
Logs should be placed on mirrored drives (RAID 1) and have its own controller channel.
Tempdb should be placed on mirrored drives (RAID 1) and have its own controller channel.
Quorum disk should be placed on mirrored drives (RAID 1) and have its own controller channel.
SQL Server data files should be placed on drives configured for RAID 0+1 with multiple controller channels (as dictated by maximum of number of physical drives supported per channel).
Disk drives should be 10,000 RPM or higher. Consider a SAN option.
Two 100MBs network cards per node, minimum. Should consider gigabit network cards for public network connection.
Of course, you may want to purchase a system even bigger than the “minimum” one described above.

Version: 7.0, 2000

Date Added: 6-26-2001

Continues…



Related Articles :

  • No Related Articles Found

No comments yet... Be the first to leave a reply!

Software Reviews | Book Reviews | FAQs | Tips | Articles | Performance Tuning | Audit | BI | Clustering | Developer | Reporting | DBA | ASP.NET Ado | Views tips | | Developer FAQs | Replication Tips | OS Tips | Misc Tips | Index Tuning Tips | Hints Tips | High Availability Tips | Hardware Tips | ETL Tips | Components Tips | Configuration Tips | App Dev Tips | OLAP Tips | Admin Tips | Software Reviews | Error | Clustering FAQs | Performance Tuning FAQs | DBA FAQs |