The choice of hard drives is crucial to achieve good performance with SQL Server, all versions: 15000 rpm SAS drives will of course outperform the SATA II 7200 rpm. The first will offer a number of input-output per second and therefore higher rates of larger transfers, which is extremely important for use in a server database where the limiting factor is often the performance of RAID and not the processor.

There is a relatively little known to increase further the number of inputs / outputs per second (IOs / s) on Windows 2003 if you use a RAID partition alignment. Since Windows 2008 partition creation automatically manages this optimization on Windows 2003 but it must be done manually.

RAID 5

Continue reading "Aligning partitions for SQL Server"

On one of my SQL Servers SNMP service continuously took nearly 20% of the CPU. Rather than restart the service I took the opportunity to restart the server completely and apply the latest up-to-day. The launch of Windows I got an error message indicating a crash snmp.exe. I first thought about a port problem but errors in the event viewer did not match:

Event Type: Error
Event Source: Service Control Manager
Event Category: None
Event ID: 7034
Date: 20/10/2009
Time: 6:52:40 p.m.
User: N / A
Computer: SERVERSQL
Description:
The SNMP Service service ended unexpectedly for the first time.

Finally I discovered that the problem was an extension of the SNMP service installed by my Adaptec RAID controller card. To disable this extension simply run Regedit, go to HKEY_LOCAL_MACHINE \ SYSTEM \ CurrentControlSet \ Services \ SNMP \ Parameters \ ExtensionAgents and delete value aussnmp containing the given SOFTWARE \ Adaptec \ RAIDController \ AAC \ CurrentVersion.

This data is apparently he recreated each time you restart the machine, it will therefore be to think of removing the value manually, or simply create a startup script to delete this value automatically.

On one of our SQL servers I was using an Adaptec 2810SA RAID controller card with three SATA drives in RAID 5. The server has suddenly passed away: start it got stuck on the initialization of the card and eventually display a timeout.

Fortunately I had an Adaptec 2820SA, and as these cards store the RAID configuration on the disks themselves and not cached, the map change was possible with a good hope of not losing any data. So I removed the 2810 card noting the order of connection of the three hard drives, added the 2820 and reconnected the drives in the same order. At startup the card initializes properly and displays a message indicating the presence of a RAID 5 while abroad he must accept as configuration to use. RAID 5 will go into Recovery mode / Verifying and is immediately usable.

Problem, however: the version of the map is not quite the same, Windows 2003 Enterprise R2 rebootait launch. It was therefore necessary to re-install in repair mode and providing a floppy disk with the drivers for the new RAID controller card. Once completed, all data and programs installed and was operational the server could return to production.

Total period of downtime was three hours, which for a server "noname" is quite acceptable, especially without data loss.

I recently purchase a backup server with two 250GB drives and six disks of 1 TB My idea was to install the system on 250GB drives in RAID 1 and RAID 5 to create a nearly 5 to the other six discs. The server motherboard has two RAID controllers, a trademark of LSI and NVIDIA. Unfortunately, the LSI does not support RAID 5 and NVIDIA is limited to a maximum size of 2 TB

In Windows 2003 it is nevertheless possible to create a RAID 5 software directly into the Disk Management (Administrative Tools> Computer Management> Disk Management) if Windows is able to see these discs. Why it was necessary for me to add all six disks to a RAID mode interleaving in NVIDIA MediaShield. In the Disk Management tool must then convert each disk to dynamic and GPT disk, then right click on one disk and select New Name. A dialog appears in which you can select the RAID 5 and indicate which disks should be included.

Following the sudden and simultaneous death of two of the three hard drives connected in RAID 5 on my server SQLServer 2000 (like how this is possible), I had to urgently rebuild a new server. As a reminder you can lose a hard drive in a RAID 5 without losing data, you lose everything beyond ... I detail here the steps I followed to fully restore the data.

Continue reading "Restoring SQL Server"