Workshop: Understanding I/O Performance Behavior (UIOP)
Understanding I/O performance behavior is crucial to optimize I/O-intensive applications but also the infrastructure of data centers. However, with the dawn of new technologies such as NVRAM, burst-buffers, active storage/function shipping, and network attached memory, the complexity of storage infrastructure increases significantly and the boundary between memory and storage blurs. During the procurement of new systems, data centers have to ensure that the application's needs are met. Therefore, they need to define the proper requirements for storage and provide I/O benchmarks that represent application workloads to quantify and verify I/O performance.
The main goal of the workshop is the discussion of tools to identify (in-)efficient usage of I/O resources on modern storage subsystems from the perspective of users and data centers.
The workshop covers:
- a discussion of design alternatives of storage architectures and their implications on user workflows;
- telemetry and monitoring information necessary to enable efficient performance optimization of system and applications;
- the development of representative benchmarks resembling the applications' needs.
The discussion of alternative storage architectures lays the foundation for the requirements of the monitoring and benchmarking efforts. Speakers involved in storage and file system research will present experience in alternative storage architectures, application workflows, monitoring tools to identify bottlenecks in I/O, and (benchmarking) tools to quantify I/O performance. Scientists involved in various application domains can give an introduction to their workflows and I/O requirements.
By bringing together application developers/users and I/O experts, we support the development of tools to identify and quantify I/O inefficiencies that support users and data centers.
|Date||March 22nd (starting at 14:00), 23rd (full-day), 2017|
|Venue||DKRZ, Hamburg, Germany Room 034|
|Registration||Deadline Feb. 25th|
It is embedded into a joint cooperation workshop between DKRZ and JAMSTEC.
The workshop is organized by
To participate, please register online (the deadline is now over). Alternatively, you may send an email to firstname.lastname@example.org. The deadline for registration is Feb. 25th. Note that our seats are limited and we have to confirm your attendance (you may also kindly ask about the status).
Within walking distance, we recommend the: Mercure Hotel Hamburg Mitte
Address: Schroederstiftstr. 3, 20146 Hamburg - Rotherbaum, Deutschland
DKRZ has a customer ID that we can provide to you upon request; this should give you a condition of 79€ per night.
The workshop is scheduled for a duration of one day, with discussion time between talks – directly followed by another workshop (to be announced). It is intended to use the workshop to foster the international collaboration for the IO-500 list, i.e., the definition of a meaningful I/O benchmark for tracking storage performance for current and future systems.
The Agenda is currently in preparation.
- 14:00 – Welcome
Thomas Ludwig (DKRZ), Makoto Tsukakoshi (JAMSTEC), Julian Kunkel (DKRZ) – Slides
- 14:10 – Goals of the workshop
Julian Kunkel (DKRZ) – Slides
- 14:30 – NVMe and NVMe over Fabrics - Architecture and HPC Usecases
Idan Burstein (Mellanox) – Slides
- 15:00 – Exascale Storage Architecture in the SAGE Project
Giuseppe Congiu (Seagate) – Slides
- 15:30 – Advanced Data Placement via Ad-hoc File Systems at Extreme Scale
Sebastian Oeste (TU Dresden) – Slides
- 16:00 – Coffee break
- 16:30 – In-memory analytics workflows on large scale climate datasets
Alessandro D'Anca (CMCC Foundation) – Slides
- 17:00 – I/O locality and I/O Workload Characterization
Jean-Thomas Acquaviva (DDN) – Slides
- 17:30 – Discussion
Jean-Thomas Acquaviva (DDN)
- 18:00 End
- 09:00 – NEXTGenIO Performance Tools for In-Memory I/O
Holger Brunst (ZIH / TU Dresden) – Slides
- 09:30 – Leveraging Continuous Monitoring & Periodic Benchmarking for Understanding Production HPC I/O Behavior
Shane Snyder (Argonne National Laboratory) – Slides
- 10:00 – Online Monitoring of I/O
Eugen Betke (DKRZ) – Slides
- 10:30 – Coffee
- 11:00 – LLview: Job-based I/O-monitoring with GPFS mmpmon data
Wolfgang Frings (FZ Jülich) – Slides
- 11:30 – Understanding and Improving Storage Accesses at the K computer
Yuichi Tsujita (RIKEN) – Slides
- 12:00 – Discussion
Wolfgang Frings (FZ Jülich)
- 12:30 – Lunch
- 13:30 – Vendor Perspective on HPC Storage Benchmarking
Denis Gutfreund (ATOS) – Slides
- 14:00 – Improving data storage system structure and performance over various applications at JAMSTEC
Tsuyoshi Nakagawa (JAMSTEC) – Slides
- 14:30 – Coffee
- 15:00 – Improving NoSQL Database Benchmarking: Lessons Learned
Steffen Friedrich (Universität Hamburg) – Slides
- 15:30 – Analyzing I/O Profiles
Keeran Brabazon (ARM) – Slides
- 16:00 – Coffee
- 16:30 – Benchmarking for NWP/climate
Hisashi Yashiro (RIKEN) – Slides
- 17:00 – Metadata/Small file benchmarking / Status of IO-500
Julian Kunkel (DKRZ) – Slides
- 17:30 – Discussion
Julian Kunkel (DKRZ)
- 17:55 – Farewell
Thomas Ludwig (DKRZ), Makoto Tsukakoshi (JAMSTEC), Julian Kunkel
- 18:00 – End of the workshop