Table of Contents
Advertisement
more resources. However, today on x3950 M2 servers, you can scale-up to
systems that support 16 processors (96 cores) and 64 GB of memory on 32-bit
versions of operating systems and 1 TB of memory on operating systems with
64-bit extension EM64T.
NUMA architecture, when compared to other architectures, provides near linear
scalability and minimum overhead in resource management that limits the
scalability of a single large systems when you add processors, memory, and
storage.
The x3950 M2 server is a good example of an SMP system based on eX4 and
NUMA technologies. The server starts with a base 2-way configuration and, as
your requirements grow, you can add incremental capacity to a maximum of 16
processor sockets (64 cores). Likewise, memory can be expanded from 4 GB to
1 TB. This modular server architecture delivers investment protection without the
up front costs of expensive switch-based alternatives.
Advantages of scale-up include:
Easier to configure and administer
Good when most queries access small blocks of data
Best for applications that maintain state (OLTP)
Add CPU and memory as required (scheduled downtime especially for CPUs)
All tools and queries work as expected
Can be maintained by lesser skilled DBAs
Disadvantages of scale-up include:
Requires higher cost hardware
The database has finite capacity limits tied to hardware
Must balance CPU, memory, and I/O to achieve peak performance
Fail-over cluster server usually configured equal to primary
2.8.2 Scale-out
Scale-out means adding discrete servers to your server
processing power. Although many options exist for implementing a farm
comprised of small low-end servers, we consider the use of the IBM
BladeCenter, 1U rack servers or iDataPlex for large scale-out implementations
such as the System x3550 as the most viable alternative when discussing this
requirement.
farm
to gain more
Chapter 2. Product positioning
85
Advertisement
Table of Contents
