HP-UX Reference (11i v3 07/02) - 5 Miscellaneous Topics (vol 9)

a
Aries(5)
Itanium(R)-based Systems Only
Aries(5)
How Aries Sets PA-RISC Application Stack Size
At process startup time, Aries sets the size of the PA-RISC application stack based on following criteria:
1. Aries first checks if Aries option
-ssz is present in the Aries resource configuration file. If so, it sets
the value of the PA-RISC application stack with the value of
-ssz.
2. If the Aries option
-ssz is not present, Aries checks if the stack size limits are specified by the parent
process. If specified, Aries sets the size of the PA-RISC application stack as specified by parent pro-
cess. The parent process can modify the stack size limits by the system call
setrlimit
(RLIMIT_STACK, ...)
.
3. If the Aries option
-ssz is not present and the parent process has not specified the stack size limits,
Aries sets the value of the PA-RISC application stack size to the value obtained by the system call
getrlimit (RLIMIT_STACK, ...)
. If the application was invoked from a shell, this value will
be same as the value set by the user in the shell as
ulimit -s.
For details on the maximum possible stack size that Aries can allocate, see the section Relation between
pa_maxssiz_32|64bit, and Aries Parameters.
Note: Stack sizes can be inherited among an Itanium-based system’s native processes (32-bit and 64-bit)
and emulated PA-RISC processes (32-bit and 64-bit). This is subject to the limitation on available stack
space. For example, a 64-bit process can set a stack size which is not possible to allocate in 32-bit process’s
address space. In that case the 32-bit process will use maximum stack space.
ARIES PERFORMANCE
For the purpose of performance comparison we assume similarly priced and fully configured PA-RISC and
Itanium2-based systems. Because of different processor caches, memory slots, bus speeds it is not possible
to find PA-RISC and Itanium-2-based identical systems as regards to machine resources.
Relative performance of PA-RISC applications under Aries running on a Itanium-2-based system running at
1600 MHz is shown in table below.
Application Class Compared to Compared to Compared to
Itanium-2 1600 MHz PA8700 750 MHz PA8800 900 MHz
based system based system based system
General (mix of I/O, 60-70% 90-100% 80-90%
memory, system
intensive ops)
Integer intensive 40-45% 90-100% 65-70%
Floating point 15-20% 40-50% 30-35%
intensive
Java based 40-65% 70-80% 60-75%
applications
Multithreaded 40-50% 60-70% 50-60%
applications
Performance numbers shown in above table are only indicative. The actual performance of your application
may vary depending on application execution profile. It is recommended that you benchmark performance
of your applications under Aries on Itanium-2-based systems before actual deployment.
Aries performance can be characterized by following points:
1. General applications are those which have average mix of I/O intensive, memory intensive and system
intensive operations. Such applications run with good performance under Aries.
2. Integer intensive applications are those which spend significant time in algorithmic computations.
Such applications run with good performance under Aries.
3. Floating point intensive applications are generally scientific simulation, modeling applications. Such
applications suffer performance under Aries due to architectural differences between PA-RISC and
Itanium-2 processor feature and resources for floating point operations.
4. Unless your Java based application uses JNI code it is recommended that you run your application
with Itanium-2 native JVM. Depending on execution profile performance of Java based applications
74 Hewlett-Packard Company − 11 − HP-UX 11i Version 3: February 2007