Storage and Perfminer Seminars

The RES (Spanish Supercomputing Network) coordinated by BSC organized the Technical Training Storage management in the RES and Function and application of Perfminer the last 14th and 15th September 2010 in Barcelona.

Date: 
Tuesday, 14 September, 2010 (All day) to Wednesday, 15 September, 2010 (All day)
Objectives: 

In respect of the first seminar: description of hardware storage in the RES, typical maintenance tasks, monitoring tools and optimization. GPFS description commands, fees and multicluster version. Presentation of alternative hardware and software.

Regarding the second seminar: descripction of function and applicaton of Perfminer in order to collect, store and display data on system performance.

Agenda: 

1st Seminar - Storage management in the RES
(Tuesday, 14th September 2010)
11.00h - Welcome
11.15h - Session 1. Description of the storage hardware in the RES (Javier
             Bartolomé, BSC-CNS)
             1a) Storage Manager (dsclient)
             1b) Warnings
             1c) Options,...etc.
11.45h - Session 2. GPFS: Description, commands, quotas, multicluster,...etc.
            (Javier Bartolomé & Xavier Abellán, BSC-CNS)
             2a) GPFS perfomance theoretical explanation
             2b) GPFS Multicluster
             2c) How to get GPFS load per blade
             2d) MPI I/O
13.15h - Lunch break
14.30h - Session 3. Typical maintenance tasks, monitoring tools, optimization...etc.
            (Javier Bartolomé, BSC-CNS)
            3a) Crash and hang recovery: mmfsadm dump tscomm / mmfsadm dump
                 waiters and restart.
            3b) Block-level and small files optimization
15.00h - Session 4. BIFI backup system and Lustre as another HPC filesystem
            (Guillermo Losilla, BIFI-UZ)
15.45h - Coffee break
16.00h - Session 5. Upgrade Storafe in BSC and the RES (Javier Bartolomé, 
             BSC-CNS)
17.00h - Session 6. Questions, comments and impressions.
18.00h - End of the seminar

2nd Seminar - Function and application of Perfminer
(Wednesday, 15th September 2010)
10.00h - Session 1. Introduction and system description (Xavier Abellán, BSC-CNS)
            1a) Perfminer contributions
            1b) Architecture and general structure of the system
            1c) Extraction of information
            1d) Integration with SLURM (plugin)
            1e) Collection of information
            1f) Storage. DB.
            1g) Visualization
                  - Classical web application
                  - Perfminter
11.00h - Session 2. Deployment en MN (Xavier Abellán, BSC-CNS)
            2a) Demonstration of the entire system in MareNostrum
            2b) Requirements
            2c) Next Steps
            2d) Typical problems
11.40h - Session 3. Conclusions and results obtained so far. (Xavier Abellán,
             BSC-CNS)
12.00h - Coffee break
12.15h - Session 4. Other methods of extracting performance measures. (Xavier
            Abellán & Jorge Naranjo, BSC-CNS)
13.30h - End of the seminar

Target group: 

The seminar is quite specialized and is aimed at all the RES nodes tecnhical teams.