At the physical level, file sharing services should include integrated namespace support, file compression, configurable block size, mirroring, duplexing, striping with or without parity, removable device support, link tracking, and a means to automatically archive unused data while allowing it to remain available to the user. In terms of providing additional services, a good file system implementation provides integrated content indexing, user-definable file properties, and a tracking log to audit storage services usage.
From a management perspective, volume defragmentation, backup and restore, easy security administration tools, disk quotas, and the ability to configure file systems dynamically without downtime are also key features. Finally, support for the latest performance-enhancing hardware technologies is important in high usage environments.
Solaris 7 Implementation Details
Solaris 7 offers a comprehensive, but at times complex, file sharing service implementation. Out of the box, Solaris 7 provides support for two file systems – UNIX File System (UFS) and the industry standard Network File System (NFS).
In terms of value added file system features, Solaris 7 provides the following on the UFS and NFS file systems:
Block Sub-Allocation allows the operating system to use unused space within each physical block to store additional information, resulting in less wasted disk space – especially on systems with large block size.
Self-describing Disks allow for metadata that describes disk configuration to be stored on the device itself and to be replicated. Self-identification of managed disks ensures that disk controller ownership transfers are completely error free. Disk reconfigurations and cluster disk ownership transfers are also error-free.
Data Striping at the software level. With this, a volume can be equally spread between two or more physical devices, greatly enhancing read performance and reducing disk device wear.
Mount Points are tools provided in UFS for grafting storage name spaces together. It allows the mounting of a file system at the directory level on an existing volume – similar to junction points in Windows 2000. Mount points are transparent to applications unless an application is explicitly instructed to notice them. This means that users can use junction points to reroute applications or users accessing a local UFS directory to any other partition.
Disk Mirroring allows disks to be physically replicated to other disks, block-for-block, within a server. In the event of a failure, the mirror will automatically activate, allowing the server to continue operations despite having lost a physical disk. Solaris 7 supports 2-drive and 3-drive mirror sets.
Data Migration provides the necessary operating system hooks at the file system level to allow a Hierarchical Storage Management (HSM) implementation. With HSM, infrequently used data can be archived to removable storage such as tape, CD-ROM, or optical disk but still readily available to users. In the event a user requests an archived file, it will be automatically loaded and retrieved.
The Network File System (NFS) is a standard for network file sharing and provides support for any client with NFS software. Solaris 7 includes the Solstice NFS Client 3.2, which provides integrated client access for Windows 95, Windows 98, and Windows NT. Management of network shares is best accomplished from the command line, with no straightforward GUI tool available to create and maintain network shares.
Sun provides an enhanced version of NFS that is designed specifically for sharing files over the Web. This version, called WebNFS, extends the standard features of NFS to the Web and enables Web-based collaboration. With WebNFS, users can access data on the Web just as they access local data. WebNFS uses HTTP over TCP/IP to communicate and is designed to be more reliable and dynamic than FTP. Many Windows-based applications have similar functionality built-in. For example, with Office 2000 applications, you can access Web folders directly in the Open File dialog box. In both Windows NT and Windows 2000, you can also grant access to data over the Web through Web sharing, which is similar to file sharing. For true Web-based file handling, Windows users can rely on World Wide Web Distributed Authoring and Versioning (WebDAV), which is discussed later in this white paper.
Removable Storage Support
CD-ROM support is provided natively in Solaris 7. Support for other removable media (other than backup media) is fairly limited and becomes problematic if the media is not SCSI-connected.
As mentioned earlier, the DiskSuite 4.2 tool included in Solaris 7 is probably the best management tool in the package. From this tool, detailed control is available over the hard drives installed in the system and the manner in which data is laid out on the drives. Data Striping, data striping with parity, mirror sets, and other storage fault tolerance features are controlled from this same interface.
File system backup and restore tools are provided by the Solaris Data Backup Utility. This is a licensed version of the Legato Networker enterprise backup tool. Unlike the backup utilities included with Windows NT 4.0 and Windows 2000, the Solaris Data Backup Utility is a full featured, fully functional enterprise backup tool that supports cross-platform clients other than UNIX. However, Sun decided to package Solaris Data Backup Utility separately from Solaris 7. This means you must acquire and install this product.
Storage Hardware Enhancements
Lastly, Solaris 7 also provides support for the I2O standard on the Intel platform. With I2O, CPU overhead associated with disk requests can be offloaded to intelligent microprocessors on I2O-compliant systems and disk controllers, lowering server CPU usage and improving performance.
Windows NT Server 4.0 Implementation Details
Windows NT Server 4.0 provides support for two file systems as part of the operating system: FAT16 and NTFS. For the purpose of this review, only NTFS will be evaluated as the FAT16 support has been provided merely for compatibility with prior versions of Windows. First introduced with Windows NT version 3.1, NTFS is Microsoft’s 64-bit, next-generation file system.
Name space support in NTFS is integrated into the file system. Hence, files of virtually any type can be created and saved on NTFS-based systems.
NTFS, as it exists in Windows NT Server 4.0, features many advanced storage features including the following:
Compression allows for the compressed storage of files and directories so that less physical space is required, reducing the amount of disk space actually utilized. Compression is configurable on a volume, directory, or file basis.
Spanning is available, allowing administrators to extend volumes across multiple physical disks.
Data Striping implements RAID Level 0 functionality at the software level. With this, a volume can be equally spread between two or more physical devices, greatly enhancing read performance and reducing disk device wear.
Data Striping with Parity implements RAID Level 5 functionality at the software level. With this, a volume can be spread between three or more physical disks with parity information distributed across all disks. In the event of a single disk device failure, the parity information can be used to keep the volume and system running until the system administrator can rectify the problem, greatly reducing the chance of disk-related server failure.
Data Mirroring/Duplexing allows for the physical duplication of disk devices in real-time, providing for instantaneous recovery in the event of a failure. With mirroring, you use one disk controller for both mirrored drives. With duplexing, you use two disk controllers, one for each mirror drive, which prevents a controller-related problem from causing a system crash.
Windows NT Server 4.0 provides no disk quota support, although the functionality is available from third-party independent software vendors (ISVs). As would be expected, full security is available for the restriction of access to files and directories. For Windows NT-based systems running in a workgroup, the local account database will be used. For servers participating in a domain environment, the domain’s account database will be used. The Windows Explorer utility provides a means to administer security using friendly extensions to Windows Explorer for managing access control.
Network File Sharing
Any NetBIOS-capable clients capable of connecting to Server Message Block (SMB) shares can connect to and access files from NTFS shares. Additionally, with the optional Services for NetWare and Services for UNIX add-ons, file shares can be made accessible to NCP-compatible NetWare clients and any UNIX system that supports the NFS protocol.
No removable storage or Hierarchical Storage Management features are present in Windows NT Server 4.0. Management of volumes is through the graphical Disk Administrator utility and the command-line CHKDSK is used for volume repair. Changes are not dynamic and require the server to be rebooted.
Finally, backup and restore support is provided through the Windows NT Backup Utility. Windows NT Backup provides a rudimentary feature set to backup and restore files and directories on any Windows NT-compatible ATAPI or SCSI tape device. Registry information, domain controller configuration, and file systems are all supported as targets for backup and restore by Windows NT Backup.
Windows 2000 Server Implementation Details
Windows 2000 Server features the latest implementation of the NTFS file system. It provides numerous additional features and enhancements over the prior versions of NTFS. Additionally, support for the FAT16 file system is carried over from Windows NT Server 4.0 and support for the FAT32 file system found in Windows 98 has been added. Again, for the purpose of this document, Windows 2000 Server shall be evaluated only on the NTFS file system.
The NTFS implementation in Windows 2000 Server contains all of the features found in the Windows NT Server 4.0 NTFS implementation and also offers the following improvements to the core file system:
Native Property Sets are now supported on any file or directory. These are indexed periodically by the integrated Indexing Services, providing for fast searches based on properties such as document author. Potential applications include flat file annotation, metadata caching, and content management.
Sparse File Support allows an application to create huge files without actually committing disk space for every byte. For example, a user can create a 42-GB file that only has data written to the first and last 64K segments within the file. In this case, the file would only physically occupy 128K of space on the disk, but in all other respects it would act as if it were 42-GB in size. Sparse file support provides for many interesting applications such as sparse arrays and circular queues.
Reliable Change Journal tracks all changes to files and directories over long periods of time and across system reboots. With this feature, I/O can be analyzed and the creation, renaming, and deletion of files can be easily tracked. Potential applications include I/O performance analysis and application state recovery.
Mount Points are tools provided in NTFS for grafting storage name spaces together. Essentially, it allows the mounting of a file system at the directory level on an existing volume – similar to mount points in UNIX. Mount points are transparent to applications unless the application is explicitly instructed to notice them. You can reroute applications or users accessing a local NTFS directory to any other partition. Local file systems mounted on top of mount points can be accessed through them even if they do not have drive letters assigned – essentially removing the 26 drive letter limit.
Distributed Link Tracking provides a service to preserve shortcut integrity when users move or rename files. With this feature, client applications are allowed to track link sources that have been moved. Links can either be shortcuts for files or embedded OLE objects, such as a Microsoft Excel worksheet embedded in a Word document. Changes tracked include renaming the link source, moving the link source within the same volume, moving the link source between two volumes on the same computer, moving the link source between two computers in the same domain, moving a volume from one computer to another in the same domain, renaming a computer within a domain, changing a network share under which the link source is shared, or any combination of these scenarios.
Windows 2000 Server now provides full disk quota support. Quotas are tracked on a per-user, per-volume basis and users are charged only for files they own. Events are automatically logged when users exceed warning thresholds and quota limits. Policies are configurable to set up wide-scale remote management of disk quotas. Quota information can be saved along with other volume information during backup. Restoring a backup will always override quota limits, provided that the user performing the operation has the appropriate backup privileges. Windows 2000 also provides scripting extensions that you can use to build powerful quota reporting tools.
Bulk Access Control List (ACL) Checking
Also added to Windows 2000 Server is the Bulk ACL Checking feature. With it, administrators can perform accessibility tests against multiple files by specifying an arbitrary access mask. This allows administrators to perform such tasks as:
Determining what user X can do with N files.
Check multiple ACLs simultaneously for file access.
File System Content Indexing
Full content indexing capabilities have been added to Windows 2000 Server with the introduction of integrated Index Services. This completely indexes all files available on a file system according to content or attributes. Index Services is fully integrated with Windows Explorer. It can be used from the Find Files or Folders dialog.
Index Services can also be used to index content on the Internet or remote sites. The index can then be queried via Web-based forms. All operations of Index Services are automatic, including updates, index creation and optimization, and crash recovery. Manageability of Index Services is extremely powerful. It can be configured to operate on a per file or per directory basis. It takes full advantage of other enhancements to NTFS and also features full integration with the Windows 2000 Hierarchical Storage Management (HSM) feature, providing for the indexing of content archived through HSM.
A complete, dynamic volume management has been provided in Windows 2000 Server in the Disk Management tool. Authored by VERITAS Software Corporation, the Disk Management tool provides the following functionality:
Online Disk Management support has been added, allowing administrators to perform all common disk administrative tasks without having to shut down the system. A volume can be created, extended, or mirrored without requiring a reboot.
Self-describing Disks allows for metadata that describes disk configuration to be stored on the disk itself and then be replicated. Self-identification of managed disks ensures that disk controller and other disk reconfigurations or cluster disk ownership transfers are completely error-free.
Simplified tasks represent one of the most important features of the new Disk Management tool, which is based in the Microsoft Management Console. Consequently, the Disk Management MMC snap-in tool is considerably easier to use than the Disk Administrator tool it replaces. It provides shortcut menus to walk customers through tasks that can be performed on a selected object. Wizards guide users through creating partitions and volumes and initializing or upgrading disks.
Hierarchical Storage Management (HSM)
Windows 2000 Server provides a complete Hierarchical Storage Management (HSM) solution with the inclusion of Remote Storage Services (RSS) – a data archiving solution based on technology provided by Seagate Technology, Inc. RSS makes it easy for customers to increase disk space on a server without adding additional disk capacity. RSS automatically monitors the amount of space available on the local hard disk. When the free space on a managed primary hard disk drops below a preset threshold, RSS will automatically remove local data that has been copied to remote storage devices, while still keeping the directory and property information active. Since removable optical disks and tape are less expensive on a cost per megabyte basis than hard disks, this represents an economical way to provide both maximum storage and optimal local performance.
RSS uses an administrator-defined rule set to move infrequently accessed files to long-term storage. Reparse points are used to store information about the file in the file system. This information itself is physically stored in a stub file containing the reparse point whose data points to the device where the actual file is now archived. Windows 2000 can use this information to retrieve the file in the event a user submits a request for it.
Removable Storage Management
Windows 2000 Server provides a comprehensive removable storage solution in the form of the Removable Storage Manager (RSM). With RSM, management tasks such as mounting and dismounting media are now performed automatically. RSM presents a common interface to robotic media changers and media libraries. It enables multiple applications to share local libraries and tape or disk drives and controls removable media within a single system.
Seagate Software has provided an update to the Windows NT Backup software as the backup and restore solution for Windows 2000 Server. It is based entirely on RSM technologies and sports a new user interface. It has wizards to ease common backup and restore tasks. Windows NT Backup in Windows 2000 Server represents a significant step forward over the version in Windows NT Server 4.0. Most importantly, Windows NT Backup now supports a variety of optical and magnetic storage devices not supported in Windows NT Server 4.0. The Windows NT Backup in Windows 2000 Server can backup and restore all file systems, the registry, and its Active Directory service.
Disk defragmentation support has been added to Windows 2000 Server based on technologies from Executive Software. This reorganizes clusters on a disk volume so that files, directories, and free space are more contiguous. Depending on the extent of fragmentation, overall system performance can be improved dramatically. The defragmentation utility is implemented as a Microsoft Management Console (MMC) snap-in. It can defragment FAT, FAT32, or NTFS file systems.
Network File Sharing
File sharing client support remains essentially the same as that in Windows NT Server 4.0. NetBIOS clients running over TCP/IP, IPX/SPX, and NetBEUI are supported by default. NetWare-compatible (NCP) and UNIX (NFS) clients are supported with optional add-ons.
Storage Hardware Enhancements
Storage hardware support also has been enhanced in Windows 2000 Server to provide support for the following new, storage technologies:
Fiber Channel is a technology for 1-gigabit-per-second data transfer that maps common transport protocols such as SCSI and IP. It is an open standard, defined by ANSI and OSI, which operates over copper and fiber-optic cabling at distances of up to 10 kilometers. Fiber Channel storage support is implemented under Windows 2000 Server by layering into the SCSI stack.
IEEE 1394 is a standard for high-speed peripheral interconnects. The most compelling attributes of IEEE 1394 are simple connectivity combined with bandwidth for multimedia. The benefits include a single-connection for A/V data and control.
I2O is fully supported in Windows 2000. Consequently, performance can be enhanced on machines utilizing I2O-capable storage cards as processing can be offloaded to the storage card’s microprocessor rather than burdening the server’s CPU.
Distributed File System
The most significant enhancement to file services in Windows 2000 Server is the addition of the Distributed File System (DFS). The DFS provides a single name space for disparate file system resources at the enterprise level. So, using a single name, users can access network resources spread out over many different servers within an organization.
Technically DFS is organized as a logical structure, called a tree, which is totally independent of the physical resources. Logical volumes can be added to a DFS tree and are made available to end-users. Logical volumes can be directories on disk, entire Windows NT volumes, or another DFS tree, allowing for sub-trees.
Security is fully integrated with Windows 2000 Server. Administrators can impose access restrictions based on users, groups, and quotas. Permissions are physically associated with the shared resources themselves and are not propagated as part of the DFS tree structure.
DFS tree structure is published to Active Directory, which serves as a central coordinator for all of the topologies for all DFS trees. DFS configuration information is automatically replicated to all DFS trees within the Active Directory service. Consequently, in the event of a server failure, the topology and all resources that were not physically stored on the failed system remain available throughout the outage and configuration is restored properly when the failed system returns to online status.
Resources also can be replicated via DFS. By creating alternate, replicated resources and assigning them to a DFS root or volume, administrators can ensure that users have absolute, uninterrupted access to their files. Synchronization is totally automatic and does not require administrator intervention once configured. In addition to providing fault tolerance, replication also serves to enhance performance by balancing load between servers and by reducing network traffic via redirection so that users only connect to the server that is physically closest to them over their LAN/WAN.
Consequently, in comparison to standard file sharing services, DFS provides many additional benefits that directly benefit the customer including the following:
Fault Tolerance - When multiple servers are used as the basis for a DFS share, users will be automatically redirected to available servers. Consequently, if one of three systems is unavailable, users are transparently redirected to the remaining online servers, providing better availability and reliability than non-DFS file sharing solutions. Additionally, all configuration information is replicated in Active Directory so that it is available across all domain controllers on the network.
Load Balancing – When multiple servers are used to host a DFS share, load is automatically balanced between available servers. The benefit to customers is enhanced performance and better availability under high usage scenarios.
File Replication – A DFS share can also be made up of multiple shares across multiple services with automatically replicated file content. Synchronization is totally automated and requires no user intervention. With file replication, absolute availability is guaranteed. If any one of the servers within DFS fails, users will be able to continue working and be transparently redirected to one of the other online replicas without any interruptions in service.
Site Proximity – DFS automatically redirects users to the DFS share that is physically closest to them. This process is completely transparent and provides for greatly reduced network utilization by keeping users working locally (assuming a share is available) rather than having to access information over a wide area link.
File Sharing and Storage Services Summary
Windows 2000 Server and the NTFS file system represent the best solution for the enterprise customer. At the physical level, NTFS supports junction points, sparse files, distributed link tracking, and a volume change log. Windows 2000 Server also provides complete content indexing of the file system, hierarchical storage management. A major new feature in Windows 2000 Server is dynamic volume management. This allows for live configuration changes without requiring a server reboot. Windows 2000 Server also provides a powerful and full-featured suite of tools for backup, restore, and disaster recovery. It also supports all three of the latest developments in next-generation storage technologies – I2O, IEEE 1394, and Fiber Channel. Finally, Windows 2000 Server Distributed File System (DFS) provides for integrated load balancing, fault tolerance, and replication services – features that are unmatched by either Solaris 7 or Windows NT Server 4.0.
In terms of functionality, the Windows NT Server 4.0 implementation falls behind that of Windows 2000 Server and Solaris 7. When compared with Windows 2000 Server, Windows NT Server 4.0 falls short by not offering Fiber Channel, IEEE 1394, I2O, dynamic volume management, hierarchical storage management, removable storage management, disk quotas, junction points, sparse files, distributed link tracking, complete content indexing, and a volume change log. As Solaris 7 almost matches the feature set of Windows 2000 in this regard (though lacking such features as built-in file compression and disk defragmentation support), it provides a superior environment to Windows NT 4.0. The DiskSuite 4.2 tool used to manage and configure drive configuration and utilization is one of the few complete tools in Solaris 7 that provides a high level of integrated usability in terms of management and configuration. Unfortunately, however, DiskSuite is no longer bundled with the server. Also, although IEEE support is provided on the Solaris for Intel platform, there are no actual 1394 devices supported at this time.