Garth A. Gibson, David P. Nagle, Khalil Amiri, Fay W. Chang, Eugene Feinberg, Howard Gobioff Chen Lee, Berend Ozceri, Erik Riedel, and David Rochberg. A case for network-attached secure disks. Technical Report CMU--CS-96-142, CarnegieMellon University, June 1996.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....positioning time for random access workloads. 4. 4 Network Attached Storage Devices Analytic models and simulation studies have shown that network attached storage devices have the potential to improve scalability and bandwidth by providing direct data transfers between storage and clients [78, 79]. Experiments on prototype systems have con rmed the model and simulation results [80] Currently, most of the major storage players participate in the National Storage Industry Consortium, to explore, validate, and document the technological advancements that are prerequisite to the deployment ....

G. A. Gibson, D. F. Nagle, K. Amiri, F. W. Chang, E. Feinberg, H. Gobio, C. Lee, B. Ozceri, E. Riedel, and D. Rochberg, \A case for network-attached secure disks," Tech. Rep. CMU{CS-96-142, Carnegie-Mellon University, June 1996.

....the intelligent storage control modules to be embedded in a small chip at a cheap price and deal with direct access from clients, the location of disk drives need not be close to hosts any longer. This type of intelligent disk drives is 101 categorized into network attached secure disks (NASDs)[1 4]. In NASDs, upon receiving a data request from a client, the central file manager locates the data and provides the client with necessary information to access the data such as the address of a NASD, keys for access verification, etc. Then the client provides a NASD with proof of permission ....

G A Gibson et al., "A Case for Network-Attached Secure Disks," CMU Technical Report CMU-CS-96-142, 1996

....NAS term broadly to describe systems that include individual disk drives, robotic tape libraries, and dedicated server computers. To clarify terminology, Gibson et al. present a taxonomy of network attached storage architectures with regard to device roles in distributed file system architectures [1]. Four cases categorize these devices: Case 0 Server Attached Disks (SAD) Case 1 Server Integrated Disks (SID) Case 2 Network SCSI (NetSCSI) Case 3 Network Attached Secure Disks (NASD) The taxonomy base case is server attached disks (SAD) In this case disks locally attach to general purpose ....

....mirroring and data replication techniques which imply only 50 per cent utilization of storage devices. We expect that the evolution towards greater client functionality and private or merged file managers will continue, as will the use of direct network attached storage like NAS 2 and NAS 3 [1]. When these techniques are combined as in Cray s SFS [16] DEC s VAXcluster systems [12] and the Global File System we propose, greater functionality, performance, and availability can be achieved that with either approach employed alone. In fact, more specialization in data sharing systems ....

G. A. Gibson et al., "A Case for Network-Attached Secure Disks," tech. rep., Carnegie Mellon University, June

.... storage servers have employed the server attached disk architecture, in which disks are locally attached to servers and clients access data on disks via the server (see Figure 1(a) Recently, a new storage architecture has evolved that envisions a separation of storage devices from servers [1], 2] 9] 12] This architecture consists of a storage area network (SAN) to which storage devices such as disks are attached; servers access these devices via the storage area network (see Figure 1(b) Such a network attached disk architecture is markedly different from the traditional ....

....policies and mechanisms employed by existing file systems. As explained in Section I, in a storage area network, storage devices such as disks and disk arrays are attached to a network; servers access these devices via the storage area network resulting in an inherently distributed architecture [1], 2] 9] 12] This architectural change has several implications. First, since disks are attached to a network, the architecture allows clients to directly communicate with disks without the server in the critical path; the file server is involved only in infrequent operations such as ....

G A. Gibson et. al. A Case for Network-Attached Secure Disks. Technical Report CMU-CS-96-142, School of Computer Science, Carnegie Mellon University, 1996.

....has a limited number of licenses to run some proprietary software. In such a case, gangmatching could match licenses, machines and jobs and thereby ensure that licensing agreements are not violated. A variety of research ventures are exploring storage devices under various names, such as NASD [12], Active Disks [27] Flash [22] IBP [23] and buffer servers [4] Some commercial vendors such as NetApp [17] and EMC [11] also offer storage servers as a hardware package [14] We are making use of NeST [7] because it is an easily deployable software only appliance that speaks protocols suitable ....

Garth A. Gibson, David P. Nagle, Khalil Amiri, Fay W. Chang, Eugene Feinberg, Howard Gobioff Chen Lee, Berend Ozceri, Erik Riedel, and David Rochberg. A case for network-attached secure disks. Technical Report CMU-CS-96-142, Carnegie-Mellon University, 1996.

....but the failure of the server will affect all the clients at the same time without any graceful degradation. In a distributed server environment (e.g. Global File System (GFS) 16] Frangipani [18] CentraVision File System (CVFS) 2] File systems for network attached secure disks (NASD) [8,9]) a set of designated nodes share the load and functions of the server. Thus the load is distributed over the servers enabling scalability of the system. More servers can be added as the number of clients increases. There is no single point of failure and in case of failure of a server the system ....

G. A. Gibson, et al., "A Case for Network Attached Secure Disks," Tech. Report CMU-CS-96-142, Carnegie Mellon University, June 1996. Available at http://www.cs.cmu.edu/Web/Groups/NASD.

....load under read intensive workloads. However, there are no provisions for handling disks that perform at different rates. NASD: A recent trend in storage systems moves disks away from a single attachment to a server machine and into the network. Known as Network Attached Storage Devices (NASD) [53], this CHAPTER 3. THE CASE FOR PERFORMANCE AVAILABILITY 22 affords a greater level of flexibility in assembling large scale storage systems. With this flexibility comes a potential danger: it is highly likely that such systems will have components with differing performance capabilities. ....

Garth A. Gibson, David P. Nagle, Khalil Amiri, Fay W. Chang, Eugene Feinberg, Howard Gobioff Chen Lee, Berend Ozceri, Erik Riedel, and David Rochberg. A case for networkattached secure disks. Technical Report CMU--CS-96-142, Carnegie-Mellon University, June 1996.

....are currently under utilized on the drives. If the disks understand the types of data objects being stored, they can optimize the placement of data on the disks in ways not possible with current block mode access. Although many open questions remain regarding the programming and security models [6], this is a promising path for improving the performance and lowering costs of distributed file systems. In summary, native ServerNet attachment gives the following attributes: Gives the lowest cost and highest performance peripheral attachment through ServerNet. Requires high enough ....

G. Gibson, et al., "A Case for Network-Attached Secure Disks," CMU-CS-96-142, Carnegie Mellon University, June 17, 1996.

....machine via SCSI or Fibre Channel connections, just as they are in distributed memory machines. For further information on parallel I O architecture we refer readers to the excellent surveys in [15] and [27] A relatively new area of research is that of network attached storage devices (NASD) [19]. In NASD, storage devices are not directly connected to their host systems via a specialized I O bus, but instead communicate with their host systems through a high performance network such as Fibre Channel [16] This approach has the potential to improve performance and scalability by providing ....

Garth A. Gibson, David P. Nagle, Khalil Amiri, Fay W. Chang, Eugene Feinberg, Howard Gobioff Chen Lee, Berend Ozceri, Erik Riedel, and David Rochberg. A case for network-attached secure disks. Technical Report CMU--CS-96-142, Carnegie-Mellon University, June 1996.

....I O operations, which are transparent to higher level RADD operations. For simplicity, you can readily apply the RAID 1 architecture on local disks. RADD implements mirroring on neighboring disks, but there is no parity among the distributed local disks. In network attached secure disks (NASD) 7][8], the RAIDs are directly attached to the network as a stable storage to allow shared access by all cluster nodes. Each workstation node in the cluster may or may not have local disk attached. Even with locally attached disks, they serve to buffer the data retrieved from the NASD to local nodes. ....

. G. A. Gibson, D. F. Nagle, K. Amiri, F. W. Chang, E. M. Feinberg, H. Gobioff, C. Lee, B. Ozceri, E. Riedel and D. Rochberg, "A Case for Network-Attached Secure Disks", CMU SCS technical report CMU-CS-96-142, September 1996

..... LLNL s High Performance File System [Wat95] DEC Vaxcluster file system [KrL86] DEC87] 1984) High Performance File Server [ArB93] 1993) Cray s Shared File System [Mat95] 1994) IBM s Parallel Journaled File System [DeM95] 1995) NASD (Network Attached Secure Disk) File Systems [Gib96], Gib97] 1995) Global File System [SoE97a] SoE97b] SoR96] 1995) Veritas Cluster File System (CFS) 1998) Our definition of a shared file system is simple and therefore fairly broad: a shared file system allows direct data transfers between computers (clients) transferring data and ....

....Locking in shared file systems is performed either in the clients or in the devices. In either case this locking may be either centralized or distributed. Asymmetric shared file systems exemplify centralized, client based locking: all metadata locking is performed on the file manager [Wat95] [Gib96], Gib97] In contrast, shared file systems like the Vaxcluster [KrL86] and Oracle Parallel Server [Llo92] use a distributed, client based locking scheme called the distributed lock manager. A distributed lock manager is not as vulnerable to file manager failure and can balance the lock ....

[Article contains additional citation context not shown here]

G. Gibson et al., "A Case for Network-Attached Secure Disks," Technical Report CMU-CS-96-142, Carnegie Mellon University, June 1996.

....media. Among research systems, the most similar effort is the High Performance Storage System NAP at Lawrence Livermore [26] Digital s Petal [18] uses UDP IP over ATM as part of a large distributed virtual disk system implemented in general purpose hosts. CMU s Network Attached Secure Disk [13] does use IP for its higher level protocols which are derived from NFS, however, they provide a network object store rather than a block level NAP. The two projects most closely related to Netstation are MIT s ViewStation [1] and Cambridge s Desk Area Network [14] both of which use ATM to ....

G. Gibson et al. A case for network-attached secure disks. Technical Report CMUCS -96-142, CMU, June 1996.

....chained declustering[14] to avoid having the load increase 100 on a machine when its mirrored counterpart fails. Petal is a block server. The authors of Petal have also designed a distributed file system called Frangipani[15] to run on top of Petal. CMU s NASD (Network Attached Secure Disks)[16] is another example of network attached storage. xFS[17] is a combination of network striping and LFS (Log structured Filesystem) 18] Our system avoids the complexity of distributed filesystems by using HTTP redirects and IP masquerading to distribute user requests and mask failures. 2.3 ....

Garth A. Gibson, David F. Nagle et. al. A case for network-attached secure disks. Technical report, Carnegie-Mellon University, 1996.

....locate the server that stores a block. Another difference is that Petal s servers map data blocks to fixed locations on disk. For writes we expect that the log structured approach of the Swarm server will enable it to use the disk bandwidth more efficiently. Network Attached Secure Disks (NASD) [Gibson96] are network attached disk drives that have built in capability to allow clients to directly access them. The goal is to off load much of the file manager s work onto the disk drive. Data intensive operations like file reads and writes are sent directly to the disk. This approach has the advantage ....

Garth A. Gibson, David F. Nagle, Khalil Amiri, Fay W. Chang, Eugene Fienberg, Howard Gobioff, Chen Lee, Berend Ozceri, Erik Riedel, and David Rochberg. A case for network-attached secure disks. Technical Report CMU-CS-96-142, Carnegie Mellon University, June 1996.

....because it can be easily integrated into any existing computer system and can transparently support most existing file systems and databases. One alternative to Petal is to design distributed storage with a richer interface that is more like a file system as is being done in the CMU NASD project [11]. This could potentially result in a system that is more efficient overall; however, we currently believe that the simpler Petal interface is adequate and that higher level services can be efficiently built on top of it. Petal s framework is sufficiently general to incorporate other classes of ....

Garth A. Gibson, David F. Nagle, Khalil Amiri, Fay W. Chang,Eugene Feinberg,Howard Gobioff, Chen Lee, Berend Ozceri, Erik Riedel, and David Rochberg. A case for network-attached secure disks. Technical Report CMU-CS96 -142, Department of Electrical and Computer Engineering, Carnegie-Mellon University, June 1996.

....system. Substantial additional processing power will be available not only in end user processors but also in server machines, storage devices and subsystems, and network devices. Putting this additional processing power to work may be a powerful tool for meeting the strategic goals of I O systems [21, 27]. The computing power is local to the devices and can take advantage of specialized and real time knowledge. Subsystem processing can be used for on the fly modeling of alternative storage configurations and dynamic reconfiguration. It can also be used to execute more complex, application provided ....

GIBSON, G. A., NAGLE, D. P., AMIRI, K., CHANG, F. W., FEINBERG, E., LEE, H. G. C., OZCERI, B., RIEDEL, E., AND ROCHBERG, D. A case for network-attached secure disks. Tech. Rep. CMU-CS96 -142, Carnegie-Mellon University, October 1996.

No context found.

Gibson, G., Nagle, D., Amiri, K., Chang, F., Feinberg, E., Gobioff, H., Lee, C., Ozceri, B., Riedel, E., and Rochberg, D., A Case for Network-Attached Secure Disks, School of Computer Science, Carnegie Mellon University, Technical Report CMU-CS-96-142, 1996.

....system. Substantial additional processing power will be available not only in end user processors but also in server machines, storage devices and subsystems, and network devices. Putting this additional processing power to work may be a powerful tool for meeting the strategic goals of I O systems [21, 27]. The computing power is local to the devices and can take advantage of specialized and real time knowledge. Subsystem processing can be used for on the fly modeling of alternative storage configurations and dynamic reconfiguration. It can also be used to execute more complex, application provided ....

Gibson, G. A., Nagle, D. P., Amiri, K., Chang, F. W., Feinberg, E., Lee, H. G. C., Ozceri, B., Riedel, E., and Rochberg, D. A case for network-attached secure disks. Tech. Rep. CMU--CS-- 96--142, Carnegie-Mellon University, October 1996.

....system. Substantial additional processing power will be available not only in end user processors but also in server machines, storage devices and subsystems, and network devices. Putting this additional processing power to work may be a powerful tool for meeting the strategic goals of I O systems [21, 27]. The computing power is local to the devices and can take advantage of specialized and real time knowledge. Subsystem processing can be used for on the fly modeling of alternative storage configurations and dynamic reconfiguration. It can also be used to execute more complex, application provided ....

Gibson, G. A., Nagle, D. P., Amiri, K., Chang, F. W., Feinberg, E., Lee, H. G. C., Ozceri, B., Riedel, E., and Rochberg, D. A case for network-attached secure disks. Tech. Rep. CMU--CS-- 96--142, Carnegie-Mellon University, October 1996.

No context found.

Garth A. Gibson, David P. Nagle, Khalil Amiri, Fay W. Chang, Eugene Feinberg, Howard Gobioff Chen Lee, Berend Ozceri, Erik Riedel, and David Rochberg. A case for network-attached secure disks. Technical Report CMU--CS-96-142, CarnegieMellon University, June 1996.

No context found.

G. A. Gibson, D. P. Nagle, K. Amiri, F. W. Chang, E.Feinberg,H.G.C.Lee,B.Ozceri,E.Riedel,and D. Rochberg. A case for network-attached secure disks. Technical Report CMU--CS-96-142, 1996.

No context found.

G. A. Gibson, D. P. Nagle, K. Amiri, F. W. Chang, E. Feinberg, H. G. C. Lee, B. Ozceri, E. Riedel, and D. Rochberg. A case for network-attached secure disks. Technical Report CMU--CS-96-142, 1996.

No context found.

Garth Gibson, David F. Nagle, Khalil Amiri, Fay W. Chang, Eugene Feinberg, Howard Gobioff, Chen Lee, Berend Ozceri, Erik Riedel, and David Rochberg. A case for network-attached secure disks. Technical Report CMU-CS-96-142, Carnegie Mellon University, 1996.

No context found.

Garth Gibson et al. A case for network-attached secure disks. Technical Report CMU-CS-96142, CMU, June 1996.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC