Overview | Tutorial Descriptions

T1 Clustered and Parallel Storage System Technologies UPDATED!
Brent Welch, Panasas

This tutorial is oriented toward administrators and developers who manage and use HPC systems, and especially for those involved with storage systems in these environments. Storage is often a critical part of the HPC infrastructure. This tutorial will show you how to get the most out of your HPC storage environment, based on a solid understanding of the fundamentals and the use of cluster-based performance tools and programming techniques.

Cluster-based parallel storage technologies are used to manage millions of files, thousands of concurrent jobs, and performance that scales from 10s to 100s of GB/sec. This tutorial will examine current state-of-the-art high-performance file systems and the underlying technologies employed to deliver scalable performance across a range of scientific and industrial applications.

The tutorial has two main sections. The first section will describe the architecture of clustered, parallel storage systems, including the Parallel NFS (pNFS) and Object Storage Device (OSD) standards. We will compare several open-source and commercial parallel file systems, including Panasas, Lustre, GPFS, and PVFS2. We will also discuss the impact of solid-state disk technology on large-scale storage systems. The second half of the tutorial will cover performance, including what benchmarking tools are available, how to use them to evaluate a storage system correctly, and how to optimize application I/O patterns to exploit the strengths and weaknesses of clustered, parallel storage systems. New to the tutorial this year is an MPI-IO theme where we explore how to use the MPI middleware layer in benchmarking and applications.

Brent Welch is Director of Software Architecture at Panasas. Panasas has developed a scalable, high-performance, object-based distributed file system that is used in a variety of HPC environments, including many of the Top500 supercomputers. Brent previously worked at Xerox PARC and Sun Microsystems laboratories. He has experience building software systems from the device-driver level up through network servers, user applications, and graphical user interfaces. While getting his PhD at UC Berkeley, he designed and built the Sprite distributed file system. Brent participates in the IETF NFSv4 working group and is co-author of the pNFS Internet drafts that specify parallel I/O extensions for NFSv4.1.

T2 Building a Cloud Storage System NEW!
Jeff Darcy, Red Hat

The trend toward moving computation into the cloud has resulted in new expectations for storage in the cloud. This tutorial is aimed at people who wish to implement their own task-specific cloud storage systems, as well as those who wish to understand why existing cloud storage systems have been designed the way they are and what tradeoffs they have made to achieve their respective goals. It will provide information necessary to build your own cloud-appropriate storage system, including:

• New requirements: Application-level users of cloud storage have come to expect a variety of data and consistency/ordering models that go well beyond those provided by traditional file, block, and relational-database systems.
• New constraints: Systems deployed in the cloud are often characterized by low levels of trust (both user/user and user/provider) and lack of hardware access for configuration flexibility.
• Techniques: Implementing a system to meet these new requirements and constraints will require a thorough knowledge of cluster and distributed-system techniques such as vector clocks, Merkle trees, Bloom filters, and various kinds of append-only storage.
• Case studies: Existing systems representing successful use of these techniques will be examined.

Jeff Darcy has worked on network and distributed storage problems for twenty years, including playing an instrumental role in developing MPFS (a precursor of modern pNFS) while at EMC and, more recently, leading the HekaFS project. He is currently a member of the GlusterFS architecture team at Red Hat, coordinating the integration of HekaFS's features and leading the asynchronous-replication development effort.

T3 Storage Class Memory: Technologies, Systems, and Applications NEW!
Rich Freitas and Larry Chiu, IBM Almaden Research Center

Over the next few years, inexpensive solid-state storage based on flash SSDs or, eventually, storage class memory technology will have a profound impact on the design and use of storage systems. This tutorial is intended for those interested in the design of storage systems for the latter part of this decade. It will briefly examine the leading solid-state memory technologies and then focus on the impact the introduction of such technologies will have on storage systems. Our discussions will include:

• current leading storage class memory technologies
• solid-state storage system design
• applications that depend on low latency storage
• performance assessment of low latency storage systems

Rich Freitas is a Research Staff Member at the IBM Almaden Research Center. Dr. Freitas received his PhD in EECS from the University of California at Berkeley in 1976. He then joined IBM at the IBM T.J. Watson Research Lab. He has held various management and research positions in architecture and design for storage systems, servers, workstations, and speech recognition hardware at the IBM Almaden Research Center and the IBM T.J. Watson Research Center. His current interests lie in exploring the performance and use of emerging nonvolatile solid state memory technology in storage systems for commercial and scientific computing.

Larry Chiu is Storage Research Manager and a Distinguished Engineer at the IBM Almaden Research Center. He co-founded both the IBM EasyTier and the IBM SAN Volume Controller products. EasyTier is a highly successful product for automated placement of data among different storage tiers in order to achieve optimal performance; the SAN Volume Controller is a leading storage virtualization engine, which has held the SPC-1 benchmark record for several years. In 2008, he led a research team in the US and in the UK to demonstrate a one-million-IOPS storage system using solid state disks. He is currently working on expanding solid state disk use cases in enterprise system and software. He holds an MS in computer engineering from the University of Southern California and another MS in technology commercialization from the University of Texas at Austin.

T4 Understanding the I/O of Columnar and NoSQL Databases NEW!
Jiri Schindler, NetApp

Structured data management systems have always been great consumers of data storage. As their architecture evolved from row-oriented systems for record-level transactional processing to columnar databases optimized for execution of complex data analytics queries, their I/O behavior and demands on storage changed as well. The shift from a relational data model with well-defined schema and strong data consistency fueled the recent outbreak of semi-structured data stores with flexible schema (so-called NoSQL databases). These developed a different storage model, one which moves away from a centrally managed (but potentially distributed) storage system to a collection of independent nodes with directly attached storage (both disk drives and flash-memory-based SSDs).

The tutorial is aimed at those who are familiar with the basics of storage and database technologies and want to gain a better understanding of how columnar and NoSQL databases are organized and how to best use storage systems. It assumes some working knowledge in those areas, but an undergraduate database and/or OS course is not a prerequisite. It targets professionals (e.g., application and storage administrators), students, and veterans of relational databases who want a better understanding of how current systems such as SAP HANA, HBase, Cassandra, and MongoDB organize and use storage. It will examine some database systems internals but will not go into great depth about their APIs or programming models. It will, rather, cover the basics of using and programming these systems in order to connect their visible behavior to their internal structure and storage organization.

Jiri Schindler is a Member of Technical Staff at the NetApp Advanced Technology Group, where he works on storage architectures, integrating flash memory and disk drives in support of applications for management of (semi)structured data. Recently, he has been investigating the I/O profiles of columnar databases and designed a system for efficient de-staging of small updates to disk drives with the help of flash memory—work he presented at FAST '11. Jiri has systems experience ranging from device-level request scheduling, through file systems, data layouts, and whole-system performance analysis. Previously, Jiri worked at EMC on Centera, the shared-nothing clustered content-addressable storage system for fixed content. While getting his PhD at Carnegie Mellon University, he and his colleagues designed and built the Fates (Clotho, Atropos, and Lachesis) system for efficient execution of mixed database workloads with different I/O profiles. Jiri has been an adjunct professor at Northeastern University, where he taught storage systems classes. He actively works with graduate students and supervises PhD theses.