Installation guide

for the changes to take effect.
n To remove a processor from SQL Server and reserve it for the OS, uncheck the check box next to
the processor. Remove the processor that the network card will use so that network
communication does not affect SQL Server. Most network cards use the first processor, but some
Intel network cards use the last processor.
n When hyperthreading is enabled, the processor list normally starts at 0 and lists the number of
physical cores, and then repeats to include the hyperthread-created processors. For example, to
unlink the first core from SQL in a four-CPU hyperthreaded system, which includes eight
processors according to the OS, clear the check boxes next to CPU 0 and CPU 4. This is the
preferred logical processor enumeration sequence recommended to BIOS vendors by Intel as part
of its Netburst architecture. A BIOS that uses this preferred sequence shows the two threads of the
first Hyper-threaded CPU as logical CPU 0 and 1. To confirm which algorithm is used, verify with
the BIOS vendor or compare the SQL Server processor affinity options with and without
hyperthreading enabled.
8. Click OK to save the settings.
SQL Server I/O Configuration
IT organizations do not analyze the technical drivers behind Disk I/O subsystems. SQL Server installations
can result in configurations that have RAID 5 arrays, which are not preferred for SQL Server because of a
compromise between write performance and data redundancy. The more redundant a system, the more
work it takes to write data.
Because SQL Server is extremely disk-write intensive, performance suffers when SQL is configured with
RAID 5. Understanding the RAID levels can help SQL database administrators configure the disk I/O
subsystem in the most efficient manner.
n RAID 0. Striping Without Parity. In this configuration, each block of data is written to each disk in the
array in a circular order, which means each disk in the array holds only a portion of the total data
written. Depending on the array configuration, this method drastically improves read performance,
because data can be read in small parallel chunks. This method also provides improved write
performance, because data can be written in parallel. However, time is required to break the data into
the “stripe” that will be written. Because no fault-tolerance exists in this model, when a drive fails in the
array, the entire array fails. A minimum of 2 drives is required for RAID 0 and the resulting size of the
array is calculated by adding the sizes of the drives together.
n RAID 1. Disk Mirroring or Disk Duplexing. This configuration uses mirroring on a single channel or
duplexing when multiple channels are used. In this configuration, each bit of data that is written to a
single disk is duplicated on the second disk in the array. RAID 1 is limited to two physical disks, which
means the array is capable of increasing the read performance. In a duplexed environment, the
performance is theoretically doubled while providing fault tolerance in case a drive fails. Write
performance is not affected by RAID 1. Only two drives can participate in a RAID 1 array, and the size
of the array is the same as a single disk.
n RAID 5. Disk Striping with Parity. As with RAID 1, data is written to each disk in the array in a “round
robin” fashion, but an additional block of data written as “parity” also exists. This parity information
can be used to rebuild the array in case of a disk failure. RAID 5 is the most popular RAID configuration
in data centers and represents an effective compromise between read performance and fault tolerance.
Because time is required to calculate the parity stripe, write performance is not as good as RAID 0. A
minimum of 3 disks is required for RAID 5. The size of the array is calculated by taking the added size
of the total disks and subtracting the size of one disk. For example, 80GB + 80GB + 80GB is equal to the
total array size of 160GB.
Configuring SQL Server for VCM
VMware, Inc.
123