Utilizing NVRAM has long been recognized as an excellent method of improving performance while maintaining reliability in NAS appliances. It provides the ability to increase NFS ops in industry standard benchmarks such as SpecSFS while still complying with network protocol requirements. With the proliferation of journaling filesystems utilizing NVRAM in NAS appliances has become even easier. This discussion will focus on how implementing NVRAM in scalable, clustered NAS appliances can be accomplished in an open network with High Availability Failover.

Tim will be presenting on some of the information gathered during a recent voice of the customer exercise and sharing some of the insights gained.

The file model was so simple ten years ago: a bucket if bits and a handful of simple attributes. The modern file model is more complicated and there are problematic differences between various systems. The stage is being set for potentially unrestrained and divergent changes. What will the file model ten years from now look like? This talk reviews emerging file features, contemplates future evolution, and concludes with a discussion about fostering industry consistency.

The talk presents a new suite of file system benchmarks that support both component and system level benchmarking. The presentation will provide an update to last years presentation on NFS client benchmarking. Examples will be presented of using the benchmark to measure file system and NFS client performance. The release plans for the benchmark will be presented.

Some very interesting trends in the data center and in storage technologies in general are driving novel storage architectures and specific design features. For example, trends related to utility computing, data life-cycle management and remote service are driving solution architectures that not only include storage systems, networks and data hosts, but also the service infrastructure deployed by solution developers. Other trends, such as the continued press on data integrity, storage device capacity, component cost and high-availability are resulting in the deployment of enhanced system features such as advanced RAID across the range of products in the data center.

This talk will present one perspective on key storage trends and how they are influencing the design for robust NAS solutions.

Computer architectures seem to operate on a "steady state" theory - things go in and out of style as the underlying technology changes. Network attached storage - in the form of Netware, SMB and NFS - was born in the networking of the early 80s, but largely fell out of vogue in the 90s. Now, another ten years on, we are faced with ever increasing complexity in data, along with the usual advances in hardware and software. In this environment, NAS has been returning to vogue. But it's more than fashion: developments in NAS seem poised to substantially change the way data is accessed and especially managed.

pNFS is a protocol extension that provides direct, parallel I/O capabilities to NFSv4. Direct I/O means that pNFS clients can access storage devices directly, bypassing the NFS server during Read and Write operations. Parallel I/O means that a single file can be distributed over several storage devices so that a client can access a file using concurrent I/O operations. Direct I/O bypasses the NFS server so that it can support more clients. Parallel I/O means that clients can get more bandwidth to a file than a single storage device can deliver. Parallel I/O also allows large numbers of clients in compute clusters to access many storage devices at once. The pNFS extension allows for different storage protocols between clients and the storage devices. These include traditional block storage devices accessed on a SAN, object storage devices accessed via iSCSI, and file servers accessed via NFS. This inclusive approach supports a goal of having a single NFS client that is interoperable with a wide variety of high performance file system implementations.

Statelessness allowed NFS Versions 2 and 3 servers to export shared storage in parallel with ease and opened the door to parallel, high-bandwidth, industry standard NAS that scales in fine-grained increments.

NFSv4 servers don't have it so easy. They have their own state to manage -- like OPEN -- but the protocol does not support distributing it among multiple servers, making it difficult to export shared storage in parallel.

Yet the aggregate bandwidth demands of modern clustered clients surpase even the bandwidth availiable with these multiple parallel NFS service systems; as long as access to a single file system through the NFS protocol requires access to a single NFS server, bandwidth will be limited to the single NFS servers resources.

Fortunately, NFSv4 is extensible. Efforts are under way to understand the common use cases for parallel access, to shape appropriate minor version extensions that overcome the single NFS server hardship, to prototype, to measure, and to engineer these solutions.

In this talk, I will describe partnerships under way at CITI to solve the parallel NFS bugaboo.

Distributed Hierarchal Storage Management allows customers to create filesystems on multiple tiers of storage and automatically move file resources from more expensive to less expensive storage resources based on user defined policies. These policies demote resources to more cost effective storage based on access patterns or other business process logic. Whatever the policy, the net impact to the customer is the same - better, more cost effective, storage utilization without an impact on the typical user experience. Traditional NAS DHSM solutions have not been transparent to the end user, however new developments now allow fully transparent, multi-protocol DHSM for NAS Environments.

We discuss the integration issues surrounding NFS and Clustering, in particular, how the Sun Cluster product helps customers protect their key NFS infrastructure from hardware and software failures. We look at the speed of NFS failover in a clustering environment, as well examine issues peculiar to clustering, such as client-side fencing, security identity failover, and file range-lock failover. Finally, we address the question of scalable (or parallel) NFS in the context of distributed filesystems available on clustering platforms such as Sun Cluster.

NFS is an important storage infrastructure component that continues to grow well beyond the scale of a single box serving up files. This scaling creates demands in multiple dimensions, including performance, capacity, and manageability and requires new developments in NFS. This talk will describe the scaling challenges facing NFS and advanced technology that EMC is creating to meet these challenges.

NFSv4 introduced the concept of delegation, a technique by which the server may delegate certain responsibilities of file management to the client. This presentation will review the delegation related aspects of the protocol, illustrate the performance gains and discuss the future evolution of the delegation concept.

The Network Data Management Protocol defines a mechanism for controlling NAS device data operations such as backup, recovery, and other data transfers between primary and secondary storage. This presentation will provide an overview of the protocol and discuss the benefits derived from implementing an NDMP solution on a NAS device.

"My file system can do 1000 I/Os per second, but what does that mean?" This and related questions are often at the top of the list for system administrators, system architects and developers. This talk shares filesystem measurement methodologies which connect real application behaviour with file system performance. Using a series of application models, we are able to accurately report the performance of databases, webservers, file servers and other applications on a target file systems.

Most network admins only want to support one protocol to the desktop. They have to support CIFS, so why make them add NFS ? This talk will explore how the Samba server program and Linux CIFS client will work together to provide a full-features UNIX-to-UNIX remote filesystem to compete with NFS to the desktop.

Describes recent progress and applications of NFS as a general purpose storage architecture. Highlights competitive architectures. Describes near term and long term directions for NFS and NAS architectures in general, and the underlying market and technology drivers for NFS.

System administration of even small networks requires that the network filesystems be centrally administered. Autofs is a technology available on most flavours of UNIX that many administrators consider vital to network operation. Unfortunately, the current Linux automounter is not yet feature complete, which impedes multi-OS network setups. This talk will cover the work we are doing to create a new feature-complete automounting framework on Linux and our progress thus far.