Description:

The goal of this project is to bring together experts in different aspects of multicore performance analysis and adjacent areas from the European academia and industry. We want to investigate automation of performance analysis on multicore and heterogeneous architectures, to help the users cope with the growing complexity and size of the performance data, especially traces detailing thread communication data.

The cluster participants share between them expertise in multicore workloads, performance tools, and operating system. We would like to leverage this knowledge to analyze multicore workloads from different areas (e.g., video and image processing, biology, etc.) and identify opportunities for automatically detecting performance patterns, and anomalies, trends and stages. Descending the execution stack from the application to the OS level, we would like to combine application-level performance data with the OS-level information, such as scheduling. We would like to evaluate how this knowledge helps identify variations in application performance over time and associate them with specific causes within the application or in the system at large. These techniques would be combined and applied to a heterogeneous system.

Technical tasks:
- Explore automated performance analyses with different families of workloads targeting heterogeneous architectures.
- Collect the most suitable information to extract patterns and programs phases that more accurately represent the program execution
- Explore the possibilities of combining OS-level and application-level profiling to improve identification of sources of performance problems. Pay special attention to memory data transfers and heterogeneous processor schedulers.
- Taking into account to the expertise of the partners, we plan to develop the first prototypes on the Cell BE platform.

We will use existing analysis tools and visualization tools as much as they can be applied to heterogeneous cores. Some extension to cover multiple binaries and address spaces may be needed.

Nature:
Research cluster

Total funding:
Requested: € 17600
Granted: € 17600

Including fellowship funding:
Requested: € 8000
Granted: € 8000 (Lluc Alvarez)

Description of how the funding will be used:

4 people traveling x 4 trips/year x 600 euros/trip = 9600 euros
Fellowship at IBM Haifa: 8000 euros

Duration of the funding:
Requested: 15 month(s)
Granted: 0 month(s), starting on: Thu, September 27, 2007

Participating members:
MENDELSON Bilha (IBM) (--member--)
BIBERSTEIN Marina (IBM) (--colleague--)
MARTORELL Xavier (UPC) (--member--)
GIL Marisa (UPC) (--member--)
NAVARRO Nacho (UPC) (--member--)
MARAZAKIS Manolis (FORTH) (--colleague--)
BILAS Angelos (FORTH) (--member--)

Other people collaborating:

Esther Salami (UPC)
Orit Edelstein (IBM)
Victor Leikehman (IBM)
Dimitris Nikolopoulos (Virginia Tech)
Lluc Alvarez (PhD student)