| ACLs Tutorial | Sam Falkner and Lisa Week, Sun Microsystems |
| NFS version 4 specifies a new model for Access Control Lists (ACLs). We will show how the new ACLs may be used, common problems in dealing with them, coexistence with UNIX permissions, idiosyncrasies of many implementations, and directions in future implementations. | |
| NFSv4 Co-existence with CIFS in a Multi-protocol Environment | John Hayden, EMC² |
| NFSv4 brings many elements to NFS that have long existed in the Windows CIFS protocol. These include items like delegation (oplocks in CIFS), a rich ACL model, location tags (DFS referrals in CIFS) among others. The similarities between the protocols will be discussed, and a specific approach on how to integrate both protocols on a common filesystem will be described. This includes managing the similarities as well as the differences between locking, access, and permission semantics between the two protocols. | |
| Filebench Tutorial | Richard McDougall, Sun Microsystems |
| Undocumented CIFS | Jeremy Allison |
| The SNIA spec does a good job of starting to document how CIFS really works, but there are many implementation details missing. Many of these details are used by Windows clients for specific Microsoft applications. This talk will cover some of the currently undocumented parts of the CIFS protocol, and explain some of the more obscure parts of the protocol that implementers really need to know in order to support Windows clients correctly. | |
| EMC² Keynote: Information Security Management | Pam Hynes |
| While the innovative pace of Networked Storage continues its trajectory with the spread of new technologies like iSCSI and NFSv5, an ever growing concern is surfacing around the secure protection of this data. This problem transcends a protocol level security discussion or even a platform security discussion - it becomes a discussion of how to manage information security in a networked enterprise. Information security is an end-to-end challenge that can only be effective when achieved through comprehensive integration with the enterprise's information infrastructure. This talk will describe the challenges of making information security management pervasive throughout an infrastructure and look at the technologies EMC is utilizing to meet these challenges. | |
| Using Filebench to Evaluate Solaris NFSv4 | Eric Kustarz, Sun Microsystems |
|
The NFSv4 protocol provides certain possibilities and challenges to its
implementation when compared to NFSv3. This talk will discuss how the
open sourced framework Filebench was used to evaluate the current
Solaris NFSv4 implementation as well to evaluate upcoming proposed
changes to the Solaris NFSv4 implementation.
This talk will also start the discussion to see if Filebench should be the replacement of SpecSFS. | |
| Hummingbird Keynote: PC NFS in Windows-Powered NAS | Robert Wong and Dan Trufasiu |
| The NAS market is large enough to support different vendors who are providing solutions to satisfy different needs. Windows-Powered NAS solution is quicker, cheaper and easier to use than the traditional Windows File Server. NFS is one of the protocols used by NAS vendors to ensure file serving to foreign hosts is successful. This presentation will address some of the requirements for a successful PC NFS and NAS integration in the Windows environment. | |
| Defending Against Broken NFS Clients | John Corbin, EP Network Storage Performance Lab |
| Some end-users install high powered NFS server farms, perhaps clustered NFS servers, while using inexpensive NFS clients only to have a misbehaving client cause the NFS Server to crash. I will take a look at some examples where the clients can crash NFS servers and discuss methods that system administrators and developers can use to minimize the impact of broken NFS clients on the NFS server and improve perceived NFS server quality. | |
| Sun Microsystems Keynote | Fidelma Russo, Senior Vice President, Data Management Group |
| NFS/RDMA Implementation | Tom Talpey, Network Appliance |
| Significant development has been happening over the past year on NFS/RDMA implementations, as well as related areas such as standards, the Linux OpenIB effort and the emergence of iWARP. This talk will present an update of the NFS/RDMA efforts, and also the many areas which are contributing to NFS/RDMA. | |
| NFS Observability: The Undiscovered Country | Bill Baker, Sun Microsystems |
| The observability of the common NFS implementations has historically been rather weak. This presentation will examine what is available right now and demonstrate some new tips and tricks at rolling your own observability scripts using the dynamic tracing facility in Solaris 10. The presentation will also explore the undiscovered country (with apologies to William Shakespeare), future improvements to NFS observability currently underway at Sun. | |
| Glamour: An NFSv4-based File System Federation | Jonathon Haswell, IBM Almaden Research Center |
|
We describe the design and implementation of "Glamour" - a
federated file system layer that enables clients to seamlessly
navigate data spread across multiple, heterogeneous and widely
distributed file servers. Glamour is not a globally distributed
file system. Instead, it enables a set of loosely coupled file
servers to function as one. Glamour uses the standard NFSv4
protocol and clients, and can operate on any underlying file
system.
As a first step towards building globally distributed file services, Glamour uses the NFSv4 protocol's referral mechanism to provide a common enterprise-wide namespace across multiple file servers. Beyond this basic infrastructure, Glamour provides data management services that include replication, non-disruptive migration and persistent server-side caching of data. Glamour supports flexible data management by defining "fileset" - a logical unit of data management that can range from a single directory to an entire file system. Filesets can be individually replicated at multiple server locations and migrated across the federation. Each server stores sufficient namespace and data location information to operate independently. For ease of administration, Glamour centrally manages namespace operations and triggers data management events. We envision that once this mobility of data is established, Glamour can go a step further to determine when and where data needs to be placed with respect to server and network conditions. This will further work to reduce the overall cost of ownership of the server infrastructure, as per-machine utilization rises and the cost of administration and tuning falls. The data management aspects of Glamour have been implemented as a research project on Linux and AIX within the protocol specification and serves standard NFSv4 clients. NFSv4 servers have been enhanced to query location information and detecting fileset boundaries. Migration and server-side caching features are ongoing activities. | |
| Accessing NFS from Linux Laptops | Gregory Touretsky, Intel |
| A significant part of the Intel design environment is Linux-based. Engineers use Windows-based laptops to access Linux compute servers at the backend. As laptops become more powerful these days, the Intel IT department started to pursue an opportunity to utilize them also as mobile design workstations to run Linux-based design flows. As of today, the ultimate data sharing mechanism within the Intel data center is NFSv3. Most of project data, CAD tools and other important parts of the design flow remain on the NFS file servers. Adding roaming mobile clients with dynamically assigned IPs may compromise the existing netgroup-based security model. In this presentation, I'll address the issues we identified switching to a mobile NFS client, several options examined and our chosen directions - both for the short and long term. Some implementation specific notes will also be covered. | |
| The Mirror File System, a Multiple-server File System | John Wong, CTO, Twin Peaks Software Inc. |
|
Conventional file systems, whether local (UFS) or distributed
(NFS), store files and directories on storage media managed by a
single server in a single physical location. This single server
file system model presents many limitations in terms of RAS,
scalability, performance, and disaster recovery.
The presentation will include:
| |
| NFSv4 Checksums | Alok Aggarwal, Sun Microsystems |
| NFS, in the absence of Kerberos, relies on Ethernet CRCs and TCP checksums for data integrity. This presentation will look into the NFS data path and explore some of the deficiencies in the NFS stack from a data integrity standpoint. The presentation will also provide a glimpse of how checksums might be done for NFSv4. | |
| AIX NFS Client Performance Improvements for Database on NAS | Diane Flemming of IBM and Sanjay Gulabani of Network Appliance |
|
We continue to see growth in deployment of Network
Attached Storage (NAS) and the Network File System (NFS)
in business-critical database environments. Reasons for this
adoption have included ease of storage deployment, decreased
complexity, and decreased total cost of ownership. Due to
continued improvements in NFS client implementations, concerns
over inadequate performance in this type of environment are
being alleviated.
This talk will discuss NFS client characteristics that have traditionally been performance bottlenecks for this type of application. We will describe a state-of-the-art NFS client implementation on AIX that eliminates some of the main performance inhibitors. We will also show the performance benefits measured on a typical database workload running in a NAS environment with this new functionality, comparing the results to those captured on previous AIX versions and in SAN environments. | |
| NAS Technologies | Milan Shetti, Sun Microsystems | Object-based Storage for Very Large Distributed File Systems | Brent Welch, Panasas |
| Object-based Storage Devices (OSD) are a new emerging standard in the storage industry, with recent demonstrations from Seagate and IBM of devices that conform to the SNIA T10 OSD standard, and file system products from Panasas and Lustre that use object-based technology. This talk explains OSD, and goes on to discuss the impact that OSD has on the design and implementation of a distributed file system layered over OSD. Object-based storage devices are designed to be secure building blocks that can be aggregated together in large numbers to provide large, high performance systems. When building a file system on object storage, there are issues of metadata management, scaling the file system to support petabytes of data and hundreds of GB/s of throughput, fault tolerance, and internal system management so that even a very large storage system remains manageable. The talk will compare some of the different applications of object-storage, including Lustre and Panasas. Finally, we will discuss the performance properties of the scalable, object-based Panasas file system. | |