Chapter 2. Positioning
37
The parallelism is there, but it is not as obvious to the processor and more
work has to be done by the hardware before it can be utilized.
Itanium 2 supports parallelism on multiple levels. Instruction-level parallelism
(ILP) is the ability to execute multiple bundles (three instructions in a bundle)
at the same time. The Itanium 2 micro-architecture can deliver faster
performance by executing multiple bundles per clock cycle. Parallelism, both
at the instruction level and at the SMP system level, permits more efficient
use of virtually all system resources to enable improved scalability.
The Itanium processor's instruction-level parallelism helps ensure the
scalability necessary to manage large data warehouses.
Large memory addressability
Another key advantage is that 64-bit operating systems can support far more
physical memory than a 32-bit operating system. The theoretical limit for
directly addressable memory was 4 GB in 32-bit architectures and is now 2
64
or 16 Exabytes.
The Itanium 2 processor memory subsystem has a three-level cache
structure consisting of first-level instruction cache, first-level data cache,
second-level cache (L2), and third-level cache (L3). In addition, to ensure the
processors are optimally used, the x450 has a fourth-level (L4) system cache
dedicated to the CPUs (the 64 MB XceL4 Server Accelerator Cache).
The increased physical memory includes the following benefits for
applications:
– Each application can support more users.
For a comparison of the number of maximum connected users in SAP for
various hardware platforms (including Itanium 2), visit the following link:
– Each application has better performance. Increased physical memory
allows more applications to run simultaneously and remain completely
resident in the system's main memory. This reduces or eliminates the
performance penalty of swapping pages to and from disk.
– Each application has more memory for data storage and manipulation.
Databases can store more of their data in the physical memory of the
system. Data access is faster because disk reads are not necessary.
– Applications can manipulate large amounts of data easily and more
reliably. Video composition and modeling for scientific and financial
Note:
Performance should no longer be measured by just the speed in
MHz, but also by the degree of parallelism that the processor achieves.
Summary of Contents for 450 xSeries
Page 2: ......
Page 8: ...vi IBM xSeries 450 Planning and Installation Guide ...
Page 80: ...66 IBM xSeries 450 Planning and Installation Guide ...
Page 130: ...116 IBM xSeries 450 Planning and Installation Guide ...
Page 148: ...134 IBM xSeries 450 Planning and Installation Guide ...
Page 156: ...142 IBM xSeries 450 Planning and Installation Guide ...
Page 157: ... 0 2 spine 0 17 0 473 90 249 pages IBM xSeries 450 Planning and Installation Guide ...
Page 158: ......
Page 159: ......