Dynamic Analysis and Profiling of Multithreaded Systems - Advanced Operating Systems and Kernel Applications - page 174

Figure 6. Pin performance test results (Luk et al., 2005)

.NET, which provide common services, such as
location transparency and concurrency manage-
ment. Distributed computing middleware often
provides a number of “hook points,” such as
interceptors and smart proxies that are accessible
to users. These hooks provide placeholders for
adding probe functionality that can be used to
measure events typically hidden deeper within
the middleware.
This section first discusses techniques that can
be used to place probes into operating system ser-
vices and how this information can be combined
with data generated from operating system-level
performance counters. We then discuss two ap-
proaches to profiling distribute middleware ap-
plications deployed on the CORBA platform.

provides detailed access to memory and thus ac-
cess to thread and process-control blocks useful
in profiling multithreaded applications.

operating SyStem and
middleware profiling
techniqueS

All applications rely upon services provided by
the underlying operating system. These services
are primarily used to coordinate access to shared
resources within the system. To measure service
“requests” probes can be placed directly within the
operating system code that can record individual
application access to provided services. Many
COTS operating systems also provide a number
of performance counters that explicitly track
usage of shared resources. Data generated from
these counters—along with data from embedded
probes—can be combined to form a more complete
picture of application behavior.
Another common form of shared processing
infrastructure is distributed computing middle-
ware, such as OMG's CORBA and Microsoft's

Profiling System Call Interception

A typical process contains one or more threads
and a shared memory space. Application code
that is executed by threads within a process is
free to access various operating system resources
and services, such as virtual memory, files, and
network devices. Access to these resources and
services is facilitated through APIs that are