The Next Wave of Storage Technology and Devices, a new approach to storage technology, is currently the subject of extensive research and development in the storage industry. This work includes standards efforts being conducted by the Storage Network Industry Association (SNIA), advanced research and development by Intel Corporation, and academic research by universities, such as Carnegie Mellon s Parallel Data Lab. This paper describes the main types of storage technology in use today, compares the relative benefits and shortcomings of each, introduces and Devices (OSD), and presents an overall OSD technology vision. The paper also explains iscsi s role in relationship to OSD, and describes a working project named Lustre. In conclusion, the document suggests ways in which OSD might be integrated into datacenters of the future.
Contents Introduction 2 Current Storage Architectures 3 Direct Attached Storage 3 Fabric Attached Storage 3 Network Attached Storage 4 Storage Area Networks 5 Need for a New Approach 5 A New Solution 6 The World of Objects in Storage 6 The OSD Difference 7 Data Sharing and Security 7 Security 8 iscsi Something Old, Something New 9 Storage Objects at Work: Lustre 10 OSD Standardization and Development 11 OSD and the Near Future 11 The Intelligent OSD 12 Answering the Challenge 12 Glossary 13 For More Information 15 Introduction Existing enterprise storage infrastructures are feeling the strain as the volume of data generated by e-mail, e-commerce, data warehousing, and other network-based applications continues to escalate. Companies are attempting to leverage this rapidly accumulating mass of data in order to be more competitive, work more efficiently with their partners and suppliers, boost employee productivity, and better serve their customers. Storage, which had been considered a dull but necessary adjunct to networking and computing technologies, has now moved up to center stage. In fact, storage has taken on strategic proportions as the cost of managing data has become central to an organization s ability to support critical business processes. But there are problems. Most organizations have heterogeneous IT environments. As a result, they find it difficult to manage and exploit information residing in difficult to access storage silos scattered across multiple locations. Backup, disaster recovery and security are major concerns. And finding an affordable storage solution isn t easy proprietary networks and dedicated storage devices are expensive, particularly when the equipment and support systems must be deployed and maintained around the world. Storage Technology Today Compounding the problem is the confusing array of storage technologies now available, each with inherent trade-offs. Direct attached storage (DAS) is the old standby, but because of its high cost of management, distance limitations, limited scalability and connectivity, it appears to be a technology either on its way out or consigned to limited, local applications that require high-speed data transfer. DAS is being displaced by Fabric Attached Storage (FAS). FAS moves the storage off the local machine, making it a shared resource that is connected to a fabric such as Ethernet or Fibre Channel. Storage Area Networks (SAN) and Network Attached Storage (NAS) are two of the leading FAS implementations. Both have significant trade-offs with security, scalability and cross-platform file sharing. A technology is being developed to alleviate these problems. Rather than focus on device hardware or applications, it focuses on an entirely new way of abstracting storage the use of objects. 2
A New Solution Devices (OSD) enable the creation of self-managed, heterogeneous, shared storage by moving low-level storage functions into the storage device itself, and by accessing the device through a standard object interface. This is different than traditional block-based interfaces such as SCSI (Small Computer System Interface) or IDE (Integrated Drive Electronics), a hard-disk drive standard for PCs. This paper presents the background and technology of OSD, the benefits when teamed with SAN and NAS architectures, and the emerging iscsi (Internet SCSI) standard. In fact, the SNIA standards work is proposing that iscsi be utilized as a transport for OSD commands. A description of OSD s potential benefits (data sharing, storage security, and intelligent storage) is included. Finally, a technology vision describes how may evolve in the near future. Current Storage Architectures Direct Attached Storage With Direct Attached Storage (DAS) each server has dedicated storage (see Figure 1). In other words, each host is directly connected to one or more storage devices over a physical bus, such as parallel SCSI. As additional storage and servers are added to meet demand, a DAS environment can cause a huge management burden for administrators and an inefficient use of resources. The storage system becomes a heterogeneous mix of various vendor storage and management interfaces. Data sharing in these environments is severely limited because DAS is a block-based technology. This means that the host OS must maintain platform-dependent metadata that describes the blocks. Reading a file, for example, requires knowledge of this metadata. In addition, only a limited number of hosts can physically connect to the bus. Even if more hosts could connect, they must all agree on the platformdependent metadata this restricts data sharing to tightly coupled, clustered machines. (Metadata or data about data describe the content, quality, condition, and other characteristics of data.) Fabric Attached Storage Fabric Attached Storage (FAS) solutions (see Figure 2) consolidate storage over networks. They are rapidly replacing DAS devices. DAS is becoming a niche player, primarily used for applications that do not need to connect to a large number of servers and storage devices, and don t need to share storage with other hosts. FAS allows an IT organization to consolidate its data on a central system that can serve data to all connected hosts. The fabric can be one of several technologies that allow multiple hosts and devices to be interconnected through a switch Ethernet and Fibre Channel, for example. LAN WINDOWS* UNIX* WINDOWS* BACKUP SERVER RAID RAID SERVER RAID RAID SERVER CLIENTS DATA TAPE Figure 1. A traditional Direct Attached Storage model with dedicated resources for each server, with backup over the LAN Figure 2. Fabric Attached Storage 3
Today s leading FAS networked storage implementations are Network Attached Storage (NAS) and Storage Area Network (SAN). NAS and SAN offer unique benefits when compared to DAS. NAS solves the data-sharing problem by using a dedicated file server to provide file-based access to storage. A SAN provides a high-speed storage bus, such as Fibre Channel, that solves the limited connectivity problem associated with DAS. Network Attached Storage NAS is a file-based storage architecture with resources attached directly to hosts connected to an IP network or a LAN. NAS was introduced to enable data sharing across platforms. With NAS, the metadata describing how the files are stored on the devices is managed completely on the file server. This level of indirection enables cross-platform data sharing, but comes at the cost of directing all I/O through a single file server. In NAS architecture, multiple hosts can share files by using standard protocols such as CIFS (Common Internet File System) and NFS (Network File System), which enable file sharing across different platforms. Because the NAS file server handles all the metadata, the clients of the file server do not have to agree on either the metadata format or where the files are physically located on the disk. Because metadata formats and the way files are written to the disk are very platform and file system dependent, this approach makes the solution cross-platform. Note: In this paper, servers on the network are referred to as clients. The term client is not meant to suggest traditional PCs connected to a network. The NAS emphasis on providing shared access to files across a network can lead to problems. As the number of requests to the NAS device grows, queuing delays begin From RAMAC* to OSD The year 2006 will mark a major milestone in the history of storage. It will be 50 years since IBM introduced RAMAC,* the first magnetic hard disk for data storage. The IBM 305 RAMAC (Random Access Method for Accounting and Control), whose footprint took up the better part of a room, had a capacity of 5MB on 50 disks, each one 24 inches in diameter. Data was stored at 2000 bits/square inch with transfer rates of 8800 bytes/sec. The cost of leasing one of these monsters was $35,000 a year about $7,000 per megabyte per year. Times have changed. In 1998, IBM introduced a 26GB hard drive with 5,000 times the capacity of that first drive. There have been many other notable advances in the history of storage during the past 50 years, including the introduction of floppy disks (1967), Winchester 5.25 inch drives (1980), CD-ROMs (1983), Zip* drives (1994) and 1Gb Microdrives (2000). However, most of the storage research and development efforts over the decades have concentrated on the hardware, not the storage interface. In fact, it was the stability of the storage interfaces (SCSI and ATA/IDE) that made possible continued advances in storage devices and applications. IBM 305 RAMAC* circa 1956 But it s time for a change. As today s storage device technology reaches new levels of capacity and complexity, systems designers have begun to run up against a fundamental barrier the block interface. Storage technology has progressed to the point where the aging device interfaces need a major face-lift. Object-based storage is the new storage device interface that designers have been looking for. Image held by IBM Corporate Archives 4
to pile up at the file server. Also, the overhead associated with TCP leads to poor performance when it comes to moving massive amounts of data. NAS devices are not typically aggregated for performance and capacity. NAS virtualization is emerging, but does not provide aggregation for enhanced performance. Storage Area Networks SANs replace bus-based architecture with a switched fabric. Both hosts and devices are connected to the fabric, providing scalable performance, high capacity, and the ability to share devices among several hosts. SAN architecture achieves high performance by providing direct access to storage over a high-speed, low-latency fabric. Also, SANs are highly scalable to meet growing storage requirements, IT simply installs more storage and network resources. And, because the technology provides centralized storage, administrators are better able to cope with the problems of redundant file copies and multiple file versions that can clog high-use servers. Fibre Channel continues to be the connectivity protocol used by most SANs. This technology, which allows SCSI commands to be transmitted over serial, rather than parallel connections, provides high throughput, transmitting data at roughly 1400Mbps to 1600Mbps in the second-generation products now available. A Host Bus Adapter (HBA) connects the server to devices in the SAN. Typically the server will have Ethernet and Fibre Channel connections with the organization s Ethernet LAN and the Fibre Channel SAN. The HBA on the SAN side serves the same purpose as the LAN s Network Interface Card (NIC). While SAN technology addresses many of the limitations of NAS, it has a few of its own. For example: Although a SAN will allow a large number of hosts (including servers and storage devices) to interconnect over shared fabric, the ability of the servers to share data over the SAN is limited. The problem is that the servers must agree on the way metadata is represented in the storage array. Like DAS, this requirement limits data sharing in a SAN environment to tightly coupled, clustered machines. Security is another major SAN drawback. SAN security relies on techniques involving LUN masking and zoning, but these are rather coarse-grained approaches. (LUNs are logical unit numbers, identification numbers given to devices connected to a SCSI adapter.) NAS and SAN are complementary technologies: each solves a different set of problems. For example, NAS provides cross-platform file sharing, while SAN provides shared access to storage devices. In some cases, NAS file servers may use a SAN to achieve scalable access to storage. But because both these solutions suffer from the limitations described above, a new approach is needed that can retain the many advantages of these FAS technologies and, at the same time, overcome their restrictions. Need for a New Approach IT managers point out that their needs are very straightforward all they want is a storage solution that is affordable, scalable, provides cross-platform capabilities, is secure, and easy to manage. Unfortunately, there is no solution currently on the market that meets all these criteria. In addition, because stored information has taken on such strategic importance, backup and disaster recovery procedures are mandatory. However, they are hard to implement in complex IT environments. Enterprises are also concerned with the security of their valuable data, and recognize that current storage architectures may pose some security issues. Plus, the enterprise is about to encounter a tidal wave of information that has to be processed and stored what the Enterprise Storage Group (ESG) calls Reference Information. Reference information is defined by ESG as a digital asset retained for active reference and value. It includes, but is not limited to, electronic documents such as contracts, e-mail and e-mail attachments, presentations, CAD/CAM designs, source code and Web content, voice data and all manner of images from check images to bioinformatics, videos, historical documents, medical images, and photographs. ESG claims that the amount of reference information is growing faster than and will surpass all other types of information by the end of 2004. 5
In research conducted by ESG in 2002, enterprise IT organizations indicated that only 37 percent of their cumulative storage was reference information. However, by 2005 they expected that number to grow to 54 percent of their newly created information. That number will be even higher in the healthcare and life sciences industries. ESG flatly states that traditional disk storage, DAS, SAN and NAS, was not designed to meet the unique requirements of reference information. The optimized solution, they claim, will abstract the information from the medium on which it is stored. Reference information will need to be easily retrieved from storage systems that hold it. Security of this reference information will also be a major requirement. Future storage technologies must better address data sharing and security. A New Solution The use of storage objects and Devices (OSD) looks highly promising as a way of handling the enterprise s rapidly escalating storage requirements. OSD technology enables a scalable, high-performance, cross-platform and secure data sharing architecture. Products based on these architectures will combine the advantages of high-speed, direct-access SANs and the data sharing and secure capabilities of NAS. Research projects are investigating methods for achieving high performance, cross-platform capabilities and new storagefocused computing opportunities. This work is based on a considerable amount of continuing storage research that Terabytes 2001 Reference Information 2002 2003 2004 2005 Figure 3. The transition to reference information Source: Enterprise Storage Group, April 2002 Reference Information increasing exponentially Non-reference Information 92% CAGR 61% CAGR has been taking place over many years. Carnegie Mellon University s Parallel Data Lab has been one of the leading proponents of research into the model. Already underway is the Lustre Project (described later in this paper), an open source, high-performance file system that makes full use of object technology and leverages open standards such as Linux*, XML (Extensible Markup Language), LDAP (Lightweight Directory Access Protocol), open source libraries and existing file systems. It is available as open source software licensed under the GNU General Public License (GPL). More than a proof of concept, Lustre is a functioning system that underlines the fact that objects and storage appear to be made for each other. Based on progress being made, new applications and products may emerge into the marketplace in the near future. Over the long term, OSD technology may lead to intelligent processing and new types of storage applications. The World of Objects In Storage Objects, writes Mesnier, et. al., can be regarded as the convergence of two technologies: files and blocks. Files provide applications with a higher-level storage abstraction that enables secure data sharing across different operating system platforms, but often at the cost of limited performance due to file-server contention. Blocks offer fast, scalable access to the shared data; but without a file server to authorize the I/O and maintain the metadata, this direct access comes at the cost of limited security and data sharing. 1 The Best of Both Objects bring together the best characteristics of both files and blocks. Like blocks, objects are primitive, logical units of storage that can be directly accessed on a storage device no server needed. It is the metadata that provides the information needed to directly access the objects (along with other key information about the data such as its attributes, security keys and permissions). This direct, file-like access provides performance advantages similar to blocks and is a key contribution of. 1. Michael Mesnier, Greg Ganger, Erik Riedel,. Accepted for publication in IEEE Communications Magazine, August 2003. 6
An object, being of variable length, can be used to store any type of application data such as files, database records, medical images or multimedia. In fact, a single object can contain an entire file system or database (see Figure 4) it s really up to the application as to what gets stored in the object. Because objects can grow and shrink dynamically, the storage device the OSD is responsible for all space management. Objects bring out the benefits of both the NAS and SAN architectures and remove many of the trade-offs that are now stumbling blocks to the use of the two technologies. Objects make NAS high-level file abstraction possible. This enables cross-platform data sharing as well as policy-based security. Objects also make it possible to take full advantage of SAN high-speed direct access and the scalability provided by a switched fabric. The OSD Difference An Device can take many forms with a wide range of capabilities from a single disk drive to an array of drives. In addition to random access or even writable devices, tape drives and optical media can be used to store objects. The big difference between an OSD and traditional block-based devices is not the physical media it is the commands that the OSD understands. OSDs offload space management, and therefore the metadata, from storage applications such as files systems and databases. The metadata that describes the physical location of data is moved to the OSD and the device interface is changed from blocks to objects. By off-loading the metadata to the storage device, objects achieve a cross-platform interface similar to that provided by file systems today, a key capability of NAS systems. Even though hosts may use different applications to process data, objects allow the hosts to share the data. For example, a database application and a file server running on different hosts will be able to share their data with a backup application. Also, because applications can now store their structures as objects rather than a collection of blocks, security policies can be established at the object level. These policies can range in coverage from a collection of objects to a single object, or even specified bytes within an object. Data Sharing and Security Given these capabilities, the two most immediate benefits of OSD technology are improved, cross-platform data sharing and application-level security. Data Sharing The improved data sharing of objects is a result of a higher-level interface to the storage as well as the attributes that describe what is being stored. Traditional Storage Device MEMORY Traditional Block Based Storage Devices Traditional storage devices deal with blocks of data, with no comprehension of the meaning of the blocks of data. It's just raw data to the storage device, which has no knowledge of what the data represents. CPU DISK CACHE BLOCK OF DATA PHYSICAL INTERFACE (SCSI, IDE, SATA, FC,...) I/O INTERFACE DATA DATA 10110110111 11011111011 10110110111 01110110101 01101101101 01101101101 01101101101 01101101101 DATA COMMANDS Figure 4. Traditional block-based storage device 7
Objects use file system semantics. Among the OSD methods are: Create and remove Open and close Read, write and append Get and set attributes for both objects and groups The interface can be easily extended to provide applicationspecific methods for manipulating data within the object. In addition, the SNIA Technical Work Group (TWG) is looking at allowing applications to create sessions within the device in order to establish application-specific parameters such as quality of service (QoS) or security guarantees. A storage device based on objects has a new level of programmability it is able to serve data to the OSD server clients on an individual basis. Attributes Object attributes contain information about the object and its environment, such as access time, size, and group and user information. In addition, information such as backup or QoS requirements, object access patterns (sequential or random), or relationships to other objects, are stored as object attributes. Attributes provide greater interoperable access to objects and provides information to the OSD that describes the meaning of the objects. This information can facilitate more effective data management, such as on-disk layout and resource provisioning. Once the OSD has access to the attributes that describe the objects that it contains, the OSD can use the attributes to create improvements in the device. For example, the OSD might make sure that the most frequently accessed data is available more quickly, or that previous versions of objects are automatically saved when the OSD detects that they are being changed. When the OSD understands the meaning of the data that it contains, it can use that information to facilitate the creation of all kinds of new storage functions and applications that take advantage of processing power in the OSD. Security Security is another differentiator between object-based and block-based storage. Block-based SANs do provide a coarse-grained layer of security at the device and fabric level either a device gets access to another device, or it doesn t. Security is not implemented at the user or group level. On the other hand, object-based storage brings a new level of flexibility and sophistication to the SAN security architecture. Objects make it possible to partition large devices or fabrics into individual security domains whose access policies are determined by the storage applications. Every access is authorized, but without the delays associated with central authority communications. Device CPU MEMORY DISK CACHE Devices Devices deal with objects which have meaning to the OSD. OSDs could store database records and process those records in anticipation of future requests. OBJECTS 1 2 PHYSICAL INTERFACE (SCSI, IDE, SATA, FC,...) 3 4 Account No. I/O INTERFACE DATA DATA Balance Balance Due n DATA Date Due COMMANDS Figure 5. In this example each Object represents a record in a database. The OSD can organize the records so that queries are sent to the OSD to retrieve objects by balance amount, due date or balance due. For example, a command from an application for all accounts due today could be sent to the OSD, which could handle this request. In traditional storage systems the application would have to retrieve al the records and search them to get this information 8
The SNIA OSD working group s approach to the Security Architecture for is shown in Figure 6. This approach is based on a cryptographically strong capability that contains a tamper-proof description of the rights of a client. Object Managers are responsible for administering the security policy and granting access to storage for clients by creating capabilities. As shown in Figure 6, OSDs and the Object Manager share a key, but the clients do not have the key. When a server client makes a successful storage request to the Object Manager using cryptographic techniques, the Object Manager creates a capability based on the key for the OSD. This capability describes the authorized storage operation. Also, the cryptography insures that the capability is only useful for the authorized client. The server client then uses this capability to make the request to the Device. The capability describes the client s security policy and is examined by the OSD to determine the access rights of the client to an object. Since the OSD also has the security key, it can examine the capability received by the client for authenticity and accept or reject the request. The OSD itself does not provide the security policy, only the mechanism for enforcing security, a separation that is the key to building a scalable security architecture. Not having to maintain client-specific authentication information on the device allows the storage device to scale independently without regard to the number of clients in the system. iscsi Something Old, Something New There is another technology now making its debut that, together with, promises to have a major, beneficial impact on the world of storage not just architecture and infrastructure, but total cost of ownership (TCO) as well. Internet SCSI (iscsi) is an emerging IP-based technology that can provide a high-speed, low-cost storage solution for Web sites, service providers, enterprises, and even small- to medium-sized businesses. iscsi has the potential to make SANs affordable by quickly and efficiently moving large blocks of storage over the Internet. The new protocol takes a well-understood, high functioning older technology SCSI and uses IP to transport the SCSI commands over the Internet. International Data Corporation projects that the fortuitous result of blending something old and something new could create an iscsi market segment of approximately $2.48 billion by 2005. SERVER CLIENTS Metadata Capability Attribute OBJECT MANAGERS NETWORK Data Capability Attribute Management OBJECT-BASED STORAGE Figure 6. security 9
SNIA notes, The ubiquitous nature of Internet Protocol (IP) networks, their inherent intelligence and industry momentum, have prompted the industry to develop IP-based storage access technologies such as iscsi (Internet SCSI) and rapidly bring products to market. iscsi s big advantage is that it brings together storage and networking. Ahmad Zamar, a senior product line marketing engineer at the LAN Access Division of Intel, and the iscsi subgroup chairman for the SNIA IP Storage Forum, points out that With iscsi, Ethernet can transport network (messaging) data and storage (I/O Block) data over the same wire. The ability to move messaging and I/O block data together simplifies SAN configurations and enhances their functionality It is very easy to set up an iscsi SAN since all you need is to connect your iscsi devices to the network. Since iscsi uses Ethernet, IT staffs already know how to plug [in] that familiar Ethernet cable. 2 Intel R&D is conducting pioneering research projects that explore how iscsi can speed up the adoption of new, low-cost SANs. They believe that iscsi may reduce storage TCO, overcome current distance limitations, improve interoperability, and offer cross-platform scalability. Summary of OSD Key Benefits Better data sharing Using objects means less metadata to keep coherent, which makes it possible to share the data across different platforms. Better security Unlike blocks, objects can protect themselves and authorize each I/O. More intelligence Object attributes help the storage devices learn about its users, the applications and the workloads. This leads to a variety of improvements, such as better data management through caching. Active disks can be implemented on OSDs to implement database filters. An intelligent OSD can also continuously reorganize the data, manage its own backups and deal with failures. iscsi is not required to implement OSD. However, because OSD can be implemented as a new set of SCSI commands, it holds promise as the basis for new storage solutions. These solutions combine the best of NAS and SAN technology, enhanced interoperability and high levels of security all at a dramatically reduced TCO. iscsi and OSD allow OSDs to be connected to any physical network that IP is running on top of. Storage Objects at Work: Lustre An object-based approach to storage has been underway since 1999 in the form of the Lustre File System. Lustre is an open source, high-performance file system conceived by the National Labs (Los Alamos, Livermore and Sandia National Laboratories) and being developed to the Lab s specifications by Cluster File Systems, Inc., Hewlett Packard Network Storage Systems Operation and Intel R&D. Lustre (short for Linux Cluster) runs on commodity hardware using object-based disks for storage and metadata servers for storing file system metadata. Lustre depends on several layers of abstraction. At the file system level, Lustre treats files as objects that are located by requests to the metadata servers. Metadata servers support all file system namespace operations and direct actual file I/O requests to Object Storage Targets (OST), which manage the storage that is physically located on underlying Object Based Disks (OBD). This division of actual storage and allocation into OSTs and underlying OBDs allows hardware developers to introduce performance improvements in the form of new disk drives that provide object-oriented allocation and data management facilities in hardware. Lustre is being used to develop powerful storage systems at the Pacific Northwest National Laboratory in Richland, WA and at Lawrence Livermore National Laboratory. The two labs are using Lustre in conjunction with clustered supercomputers to more efficiently retrieve huge volumes of information by eliminating I/O bottlenecks. The resulting system is expected to be 10 times less expensive than comparable existing supercomputer clusters. 2. In an 2002 interview on TechStorage.com. 10
Lustre is one of several emerging applications that benefit from the use of storage objects. Others include the Panasas Storage System, EMC s Centera* product, and the IBM Storage Tank* project. All use object-based storage technology, but these are early implementations and are not based on the emerging SNIA OSD Technical Work Group efforts. OSD Standardization and Development Much of the key work on OSD is being handled by SNIA s OSD Technical Work Group (TWG). Co-chaired by Mike Mesnier, Intel Research & Development (R&D), and Julian Satran, IBM, the OSD TWG has delivered draft standards to the ANSI T10/OSD committee and iscsi requirements documents to the IETF (Internet Engineering Task Force) to facilitate interoperable solutions. Intel R&D is also conducting research projects to determine how OSD can enable the creation of self-managed, heterogeneous and shared storage. The work of Intel R&D and OSD TWG builds on concepts from the Network Attached Storage Devices (NASD) Project of the National Storage Industry Consortium (NSIC), and the Network Attached Secure Disk (NASD) project at Carnegie Mellon University. OSD and the Near Future OSD has the potential to significantly impact the networked storage industry. OSD will drive the development of architectures that address cross platform and performance requirements of storage customers. The biggest immediate benefits of OSDs will be the creation of scalable, secure storage solutions that are more easily administered and managed. The OSD architecture will have a lower TCO, which will make it very appealing to organizations as they strive to keep storage administration costs under control. The storage data that is stored as objects in OSDs also contains metadata that describes the objects. This metadata makes it desirable for the storage device to have additional processing capability so that it can execute software that uses the metadata to provide new storage functions. OSDs also make storage systems more robust by moving some storage management chores to the OSD. The OSD can use metadata that describes the objects to optimize the location of raw data on the media to improve storage performance. In addition, by understanding the Data Metadata Metadata Management METADATA CONTROL SYSTEMS NETWORK SERVER CLIENTS Data Management OBJECT STORAGE TARGETS Figure 7. A Lustre-based cluster consisting of clients, metadata control systems and object storage targets 11
characteristics of the files they contain, OSDs will be better able to automatically back up and recover these files. Future advances in OSD include enabling the development of scalable NAS that extends all the way across wide-area networks (WAN), something not possible today. In addition, NAS file servers (CIFS and NFS) could be implemented to take advantage of OSDs to offer improved security, scalability and manageability. The Intelligent OSD But perhaps some of the most interesting OSD near term developments will come in answer to the question, Can storage devices actually learn? Mesnier, et. al. think so. They state, With the emergence of object-based storage comes the potential for storage devices to actively learn important characteristics of the environments in which they operate. Storage devices today are largely unaware of the users and applications actually using the storage, because block-based storage devices manage opaque data blocks. With objects, storage devices can understand some of the relationships between the blocks on the device, and can use this information to better organize the data and anticipate needs. Objects provide attribute-based learning environments in which storage devices can become aware of the environments in which they operate, allowing them to better allocate and provision resources. Because of the more expressive object interface, storage devices can perform application-specific functions. As Mesnier points out, a SAN is, after all, a computer with processors, network connections and memory. Why not let it function like one? Answering the Challenge OSD technology has potential to answer one of the biggest challenges facing the enterprise: providing data sharing that is both secure and scalable across an organization s entire networked value chain its employees, partners, suppliers and customers. The object interface offers storage that is secure, easy to share across platforms, and no longer burdened with the trade-offs between files and blocks that plague conventional storage solutions. Because object-based storage paves the way for the evolution of a new generation of intelligent storage devices that are self-configuring, self-protecting, self-optimizing, self-healing and self-organizing, history may show that the breakthrough represented by OSD is as significant as IBM s RAMAC* was upon its introduction in 1956. Objects, says Mesnier, can have a variety of rich attributes such as timestamps, accounting information, QoS parameters, group and user information and client-specific usage patterns. These attributes provide a base for policy-based storage management, and also for information gain in which an object s actions can be correlated with its attributes. (For example, any object written in the last 24 hours will be read at 4:00 a.m. the following day for backup.) They can also furnish information about the object s behavior, such as the expected read/write ratio, most likely patterns of access, and even the object s expected lifecycle. 12
Glossary Some of the most common storage terms are included here for convenience. Some entries have been used, with permission, from an online dictionary maintained by the Storage Networking Industry Association (SNIA). We appreciate their contribution. The complete SNIA dictionary may be found at www.snia.org/dictionary. direct attached storage (DAS) A storage device directly attached to the server either internally or externally. disk block The unit in which data is stored and retrieved on a fixed block architecture disk. Disk blocks are of fixed usable size (with the most common being 512 bytes), and are usually numbered consecutively. Disk blocks are also the unit of on-disk protection against errors; whatever mechanism a disk employs to protect against data errors (e.g., ECC) protects individual blocks of data. fabric attached storage (FAS) Fabric attached storage is any storage solution that has storage devices connected to a common interconnect fabric. NAS and SAN are both fabric attached storage solutions. The fabric can be either Fibre Channel, Ethernet or another physical network. The point is that the hosts and storage devices are connected over a fabric. file An abstract data object made up of (a.) an ordered sequence of data bytes stored on a disk or tape, (b.) a symbolic name by which the object can be uniquely identified, and (c.) a set of properties, such as ownership and access permissions that allow the object to be managed by a file system or backup manager. Unlike the permanent address spaces of storage media, files may be created and deleted, and in most file systems, may expand or contract in size during their lifetimes. LUN masking A technique used by storage appliances and RAID systems to limit the set of clients to which Logical Unit Numbers (LUNs) are presented. LUN masking provides some security in SANs by controlling which clients see LUNs. This type of security is much more limited than what can be provided with object-based storage. metadata Data that describes data. In disk arrays, metadata consists of items such as array membership, member extent sizes and locations, descriptions of logical disks and partitions, and array state information. In file systems, metadata includes file names, file properties and security information, and lists of block addresses at which each file s data is stored. network attached storage 1. Storage elements that connect to a network and provide file access services to computer systems. Abbreviated NAS. A NAS storage element consists of an engine, which implements the file services, and one or more devices, on which data is stored. NAS elements may be attached to any type of network. When attached to SANs, NAS elements may be considered to be members of the SAS class of storage elements. 2. A class of systems that provide file services to host computers. A host system that uses network attached storage uses a file system device driver to access data using file access protocols such as NFS or CIFS. NAS systems interpret these commands and perform the internal file and device I/O operations necessary to execute them. object-based storage Object-based storage utilizes objects to provide direct file-like access to storage devices. An object-based storage system provides direct access to data on a storage device, without the need for a server to be in the path to storage. Another feature of object-based storage is that it allows heterogeneous applications and systems to access storage more easily and with better security. Objectbased storage will have comparable performance to block-based storage architectures. Object-based storage doesn t replace current storage systems; it complements them. SAN and NAS solutions can be built with objectbased storage technology which will improve their datasharing, performance and security characteristics. 13
object This definition of object is with respect to object-based storage. A logical collection of bytes on a storage device used to store application data (e.g., file, record). Objects use file-like methods of access (e.g., open, close, read, write), may contain attributes which describe the data (e.g., access times and sizes) and may enforce security policies to prevent unauthorized access. object-based storage device (OSD) Any device that stores objects is an object-based storage device (OSD). storage area network 1. A network whose primary purpose is the transfer of data between computer systems and storage elements and among storage elements. Abbreviated SAN. A SAN consists of a communication infrastructure, which provides physical connections, and a management layer, which organizes the connections, storage elements, and computer systems so that data transfer is secure and robust. The term SAN is usually (but not necessarily) identified with block I/O services rather than file access services. 2. A storage system consisting of storage elements, storage devices, computer systems, and/or appliances, plus all control software, communicating over a network. Note: The SNIA definition specifically does not identify the term SAN with Fibre Channel technology. When the term SAN is used in connection with Fibre Channel technology, use of a qualified phrase such as Fibre Channel SAN is encouraged. According to this definition an Ethernet-based network whose primary purpose is to provide access to storage elements would be considered a SAN. SANs are sometimes also used for system interconnection in clusters. zoning A method of subdividing a storage area network into disjoint zones, or subsets of nodes on the network. Storage area network nodes outside a zone are invisible to nodes within the zone. Moreover, with switched SANs, traffic within each zone may be physically isolated from traffic outside the zone. 14
For More Information Carnegie Mellon University Parallel Data Laboratory www.pdl.cmu.edu A leading center of academic research in storage and related technologies. Intel Research and Development www.intel.com/technology Accelerating the convergence of computing and communications. Intel Research and Development/ Devices www.intel.com/labs/storage/osd Intel work in storage is described via presentations, white papers, links to relevant sites and more. Intel Storage Web site www.intel.com/go/storage Information about Intel storage products and cooperative industry projects. iscsi Storage www.iscsistorage.com iscsi and storage over IP site. White papers, news, links, product reviews and more. Lustre www.lustre.org Lustre is a collaborative, open-source storage research project with the National Labs (Livermore, Los Alamos and Sandia), Cluster File Systems Inc., Hewlett-Packard s Network Storage Systems Operation, and Intel Research and Development. SCSI Object-Based Storage Device Command Set document ftp://ftp.t10.org/t10/drafts/osd/osd-r06.pdf Current draft specification describing the addition of Object-Based Storage commands to the SCSI specification. Storage Networking Industry Association (SNIA) www.snia.org Dedicated to ensuring that storage networks become complete and trusted solutions across the IT community. SNIA/OSD Technical Work Group www.snia.org/tech_activities/workgroups/osd Enables the creation of self-managed, heterogenous, shared storage for storage networks. Includes IBM, Intel, Seagate and over 75 other companies. T10 Technical Working Group www.t10.org The place to find more information about I/O interfaces. 15
THIS DOCUMENT IS PROVIDED AS IS WITH NO WARRANTIES WHATSOEVER, INCLUDING ANY WARRANTY OF MERCHANTABILITY, NONINFRINGEMENT, FITNESS FOR ANY PARTICULAR PURPOSE, OR ANY WARRANTY OTHERWISE ARISING OUT OF ANY PROPOSAL, SPECIFICATION OR SAMPLE. Intel disclaims all liability, including liability for infringement of any proprietary rights, relating to use of information in this specification. No license, express or implied, by estoppel or otherwise, to any intellectual property rights is granted herein, except that a license is hereby granted to copy and reproduce this document for internal use only. Copyright 2003, Intel Corporation. All rights reserved. Intel and the Intel logo are trademarks or registered trademarks of Intel Corporation or its subsidiaries in the United States and other countries. *Other names and brands may be claimed as the property of others. 0603/GO/HB/PDF 253204-PDF