Database Reference
In-Depth Information
Lock Pages in Memory permission. Versions prior to SQL Server 2012 require the Lock
Pages in Memory right as well as Trace Flag 834 to be enabled. More information on
how to configure these settings can be found in Chapter 7 . Note that large pages are
allocated at boot time by SQL Server, so a restart of the virtual machine is required
after configuration of these settings.
Note
Make sure to set the Lock Pages in Memory privilege (SQL 2012) and also
turn on large pages for SQL (version prior to SQL 2012). See this
Microsoft KB article for more information:
http://support.microsoft.com/kb/920093 .
Tip
When large page support has been properly configured, SQL Server will attempt
to allocate contiguous pages in memory at boot time. This can cause longer boot
times for the SQL Server virtual machine. Refer to this blog post for more
information: http://blogs.msdn.com/b/psssql/archive/2009/06/05/sql-server-and-
large-pages-explained.aspx .
NUMA
NUMA stands for non-uniform memory architecture. When looking at your server, you
will notice sockets, cores, and memory residing on these servers. Memory is associated
with the sockets and core. Cores will preferably access memory local to them versus
memory located on another “stack.” This is all about data locality: The better the
locality, the better the performance. The goal here is to have the cores access memory
local to them versus having to travel to another socket and core stack to access memory.
This is known as “remote memory access” or “cross-socket communication” and has
performance implications because the request has to travel across the front-side bus of
the motherboard to access the remote memory region and then back to the originating
core to return the information. Figure 5.6 details out a NUMA configuration. NUMA
cross-talk occurs when the CPU on a NUMA node must traverse interconnects to access
memory on another NUMA node.
 
Search WWH ::




Custom Search