N.P. Kronenberg, H.M. Levy, and W.D. Strecker. VAXcluster: a closely-coupled distributed system. ACM Trans. Comput. Syst., 4(2):130--146, 1986.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....of cross partition references, object level migration should be relatively straightforward in NFS 2. Migration and replication are two more areas requiring further research. 5 Related work There has been a significant amount of research and product development in the area of duster file systems [2,4,8]. Most are based on principles established in the VAXclusters [2] design. These systems use distributed lock management to control access to shared resources, which can restrict their scalability. NFS 2 partitions resources to eliminate DLM [5,7] Frangipani proposed one of the most scalable DLM ....

....straightforward in NFS 2. Migration and replication are two more areas requiring further research. 5 Related work There has been a significant amount of research and product development in the area of duster file systems [2,4,8] Most are based on principles established in the VAXclusters [2] design. These systems use distributed lock management to control access to shared resources, which can restrict their scalability. NFS 2 partitions resources to eliminate DLM [5,7] Frangipani proposed one of the most scalable DLM solutions in the literature [8] System resources are partitioned ....

Kronenberg, N., H. Levy, and W. Stecker, VAXClusters: A closely-coupled distributed system. ACM Tansactions on Computer Systems, 1986, 4(2): pp. 130-146.

....used in the management of these clusters. 1 Introduction The face of enterprise cluster computing is changing dramatically. Where as in the past clusters were dedicated resources for supporting particular styles of computing (OLTP, large batch processing, parallel computing, high availability)[7,12,13,14,21,22], modern Data Centers hold large collections of clusters where resources can be shared among the different clusters or at least easily re assigned to hotspots within the overall Data Center organization. A Data Center may see a wide variety of cluster types; cloned services for high performance ....

....and supports for multiple styles as cluster computing in the way that Galaxy does. Most of the cluster management systems provide only support for very specific styles of management as they are built to support the operation of specific commercial clusters. Of these commercial clusters VAXClusters [13] and Parallel Sysplex [14] deserve special mention as they did groundbreaking work in building general distribution support for clusters. Cluster management systems supporting compute clusters are ubiquitous, most of them built at that the major research labs to support custom clusters ....

Kronenberg, N., Levy, H., and Strecker, W.,"VAXclusters: A Closely Coupled Distributed System," ACM Transactions on Computer Systems, vol. 4, no. 2 (May 1986)

....writing a file. The GFS research includes development of a device based locking primitive accessed through a new SCSI command called DLOCK [7] The abstracted locking layer in GFS allows other host based global locking mechanisms like a DLM (Distributed Lock Manager) or IP Dlock server. VAXCluster [8] and Frangipani [9] are symmetric shared disk file systems which use host based locking. Other shared disk file systems are asymmetric. In this approach, one host on the network acts as a master to synchronize access to the shared data. The master can manage and dispense locks to other hosts or ....

N. Kronenberg, H. Levy, and W. Strecker. VAXClusters: A Closely-coupled Distributed System. ACM Transactions on Computer Systems, 4(3):130--146, May 1986.

....14 http: firewire.org 3 Our own Netstation research is using the ATOMIC 15 high speed switched local area network[22] originating in an interconnect technology for massively parallel computers and being commercialized as Myrinet 16 . The Digital VAXcluster CI and star coupler [39] and the UltraNet are currently in only limited use, due to the aging of the technology. Solflower Computer s Storage Crossbar[56] is a newer technology with some similarities to the VAXcluster. These all suffer from the drawback of being non standard interconnects. SSA and P1394 arguably do not ....

....Using a custom file system, also known as Shareable File System (SFS) that links into the kernel at the vnode, access to the disks is coordinated to prevent metadata corruption. Although details of the implementation are proprietary, in principle it seems to have some similarity to VAXClusters[39], providing buffering and multiple device control, and optionally acting as a processor to processor communications path. The Storage Crossbar may have some of the concurrency control and internal security problems common to NAPs, but resource discovery problems and external threats should not be ....

N. P. Kronenberg, H. M. Levy, and W. D. Strecker. Vaxclusters: A closely-coupled distributed system. ACM Trans. Comput. Syst., 4(2):130--146, May 1986.

....basis for a 64 bit internal ID (Fig. 7) 16 of these 48 bits are further hashed to identify the I O node that serves as a master node for this le. This is more direct than the twostage mapping used in VAXclusters, where object names are hashed to a directory node that may not be the master node [27]. The master node is the locus of the le object, but not necessarily of any part of the le data. Another 16 bit eld of the original 48 bits is further hashed to nd the le in the object table on the master node. Each entry in the table contains information such as the 64 bit system wide ....

N. P. Kronenberg, H. M. Levy, and W. D. Strecker, \VAXclusters: a closely-coupled distributed system". ACM Trans. Computer Systems 4(2), pp. 130-146, May 1986.

....tuned for the network file system protocols such as NFS. An alternative and increasing popular approach is clustering, which physically integrates stand alone computers using fast networks, shared disk storage, and a single system image to create scalable compute and data servers [Pfi86] [KrL86], Byte97] An important component in several cluster designs is a shared file system that allows cluster machines to directly access shared disk devices across a network [KrL86] Mat95] Gib97] DeM95] Pfi95] Llo92] instead of through a server, increasing cluster performance and ....

.... fast networks, shared disk storage, and a single system image to create scalable compute and data servers [Pfi86] KrL86] Byte97] An important component in several cluster designs is a shared file system that allows cluster machines to directly access shared disk devices across a network [KrL86], Mat95] Gib97] DeM95] Pfi95] Llo92] instead of through a server, increasing cluster performance and availability. In addition, we will show that such a shared file system also makes each node in a storage area network (SAN) more effective. A SAN consists of a local area network that ....

[Article contains additional citation context not shown here]

N. Kronenberg, H. Levy, W. Strecker, "VAXClusters: A Closely-coupled Distributed System," ACM Transactions on Computer Systems, vol. 4, no. 3, pp. 130-146, May 1986.

....To prevent partitioning, VAXcluster uses a quorum voting scheme. Initial vote and quorum values are set for each node by a system administrator. Miscellaneous Status and Contact: VAXcluster is a product still supported by several engineering groups at Digital Equipment Corporation. References: [337] 2.52 XDFS Main Goal XDFS (Xerox Distributed File System) from Xerox PARC was a research project to design a very robust multipleserver system. It was intended to provide a basis for data base research. Besides, XDFS is location, access, and concurrency transparent. 49 Advantages XDFS was an ....

N.P. Kronenberg, H.M. Levy, and W.D. Strecker, "VAXclusters: A Closely--Coupled Distributed System", ACM Transactions on Computer Systems, 4(2):130--146, May 1986.

....to allow time for a redesign. Early time to market. Sendmail, Inc. like most software companies, is competing on Internet time. We wanted to be able to get our system to market as quickly as possible. Avoid proprietary, non portable hardware and software. Platforms such as VAXcluster [KLS86] or products from IBM, HP, and other vendors could achieve many of these goals using o the shelf proprietary solutions. However, this system must be straightforward to port to a variety of UNIX operating systems and perhaps even to Windows NT. scott sendmail.com , npc sendmail.com , and ....

N. Kronenberg, H. Levy, and W. Strecker. VAXclusters: A Closely-Coupled Distributed System. ACM Transactions on Computer Systems, 4(2):130-146, May 1986.

....control is specialized to SCSI. This trend toward generalization in SAN systems was also seen more than 15 years ago in Digital Equipment Corp. s VAXclusters, which first used a proprietary physical interconnect (VAX CI) then moved to the more standard and more general purpose Ethernet [6]. Four Architectures It may be too early to know how the changing levels of specialization in storage systems will affect how computers are built and sold. We can, however, identify the range of possible outcomes, particularly four architectures under active or recent research and development: ....

Kronenberg, N., et al. VAXclusters: A closely coupled distributed system. ACM Transact. Comput. Syst. (TOCS) 4, 2 (May 1986), 130--146.

....described in [Cri90] achieve atomic broadcast even when faulty channels are allowed to deliver messages to arbitrary subsets of correct processors. All other processor group membership protocols that we know assume an asynchronous communication network in which partition failures can occur [BJ87, Car85, CM84, ASC85, KLW86], and [Wal82] These acknowledgement based, asynchronous protocols do not guarantee any bound on processor failure detection and processor join delays even when run in an environment where all delays are bounded. Progress in reaching agreement on new processor group memberships and learning about ....

N. Kronenberg, H. Levy, and Strecker W. Vax clusters: a closely coupled distributed system. ACM Tr. on Computer Systems, 4(2):130--146, 1986.

....disks and, therefore, to the entire database. This increases availability and also flexibility in the distribution of the workload. 1 CHAPTER 1. INTRODUCTION 2 The shared disk approach is widely accepted in commercial database systems. One of the first platforms supporting SD was the VAXCluster [KLS86] Oracle s Parallel Server [Ora96] and IBM s DB2 V3 [MPTW94] are examples of current SD based database management systems. Since the introduction of Oracle s Parallel Server, an increasing number of commercial hardware suppliers have provided SD platforms that support SD database systems ....

....bandwidth, since it is not only used for communication between nodes but also for page transfers from and to the disks. Efficient network protocols are necessary because disk I O requires the sending and receiving of large messages. For example, the VAXCluster and digital s TruCluster Solution [KLS86, dig96] follow this design. In the second approach, processing nodes are connected via the network to form a single cluster while access to the shared disks is routed via special disk buses and interfaces. For example, this approach has been taken by Bull s Escala Powercluster, IBM s High ....

N. P. Kronenberg, H. M. Levy, and W. D. Strecker. VAXclusters: A closely-coupled distributed system. ACM Transactions on Computer Systems, 4(2):130--146, May 1986.

....are called shared file systems [3] 4] 5] 6] 7] Shared file systems provide a server less alternative to traditional distributed file systems where the server is the focus of all data sharing. As shown in Figure 1, machines attach directly to devices across a storage area network [8] [9], 10] A shared file system approach based upon a shared network between storage devices and machines offers several advantages: 1. Availability is increased because if a single client fails, another client may continue to process its workload because it can access the failed client s files on ....

....including small file workloads. In GFS 4, write caching is write back, not write through. GFS uses Global Locks (glocks) which may or may not be dlocks. GFS uses interchangeable locking modules, some of which map glocks to Dlocks. Other locking methods, such as a distributed lock manager [9] or a centralized lock server, can also be used. Our group has developed a centralized lock server known as the GLM (Global Locking Module) 18] GFS sees the Glocks as being in one of three states: 1. Not Held This machine doesn t hold the Glock. It may or may not be held by another machine. ....

N. Kronenberg, H. Levy, and W. Strecker. VAXClusters: A closely-coupled distributed system. ACM Transactions on Computer Systems, 4(3):130--146, May 1986.

....of a particular LCB. Rather than 7 attempting to repair only the missing portion of this LCB, it would be much easier reconstruct the entire LCB based on the log records on all surviving nodes. Some issues related to ensuring the failure atomicity of database management structures are covered in [23, 25, 11], where crash recovery issues for an SD lock manager implementation on a VAXcluster are discussed. When a node crash is detected, all locking activity in the database system is stopped. Then, any updates performed by failed transactions are undone. This is accomplished by the installation of the ....

N. Kronenberg, H. Levy, and W. Streker. Vaxclusters: A Closely-Coupled Distributed System. ACM Transactions on Computer Systems, 4(2):130--146, May 1986.

....to avoid copying data to and from an application bu#er, the OS can pin the bu#er and store its address translations in the descriptor table or list. System calls are required to build the descriptors inside the OS. Autonet [41] for example, uses a chained list of descriptors and VAXClusters [28] uses a descriptor table. Page remapping is a method to avoid copying [30, 10, 15, 27, 6] When transfers are properly aligned and the right length (i.e. a multiple of the physical page size) the OS swaps the virtual to physical mappings between the pages of the kernel bu#er with those of the ....

N.P. Kronenberg, H.M. Levy, and W.D. Strecker. Vaxclusters: A closely-coupled distributed system. ACM Transactions on Computer Systems, 4(2):130--146, May 1986.

....to the virtualization of storage space, their model moves the actual backing store allocation to the disk, though the interface between Petal and Frangipani is a block level one. Various system vendors have developed their own networks on which distributed device sharing takes place. VAXclusters [KLS86] and ServerNet [HG97] are two examples which use message passing between devices and hosts; the new SGI Origin series uses custom hardware to implement a shared address space on a switched network which includes many processors and I O nodes [LL97] 5 VISA Architecture VISA, Netstation s ....

Nancy P. Kronenberg, Henry M. Levy, and William D. Strecker. Vaxclusters: A closelycoupled distributed system. ACM Transactions on Computer Systems, 4(2):130--146, May 1986.

....of primary business functions. In addition, growing organizations face the need to increase the amount of computing power available to them over an extended period of time. VAXcluster systems provide solutions to these data availability and growth problems that modern organizations face. [1] This paper begins with an overview of VAXcluster systems and application design in such systems and proceeds with a detailed discussion of VAXcluster design and implementation. The paper then focuses on how this information affects the design of applications that take advantage of the ....

N. Kronenberg, H. Levy, and W. Strecker, "VAXclusters: A Closely-coupled Distributed System," ACM Transactions on Computer Systems, vol. 4, no. 2 (May 1986): 130-146.

....clusters. services is to attach storage devices to a system area network (SAN) and let nodes share these devices directly (Figure 1. 4) 152, 52, 114, 122, 76] This architecture is often used in conjunction with a cluster aware operating system that arbitrates device ownership among nodes [85, 157, 148, 74, 75]. This architecture can be seen as a fault tolerant variant of statically partitioned clusters the data remain available as long as the disks remain operational, even when some of the nodes are down. It also automates some of the data assignment tasks a file system can grow ....

....27 3.3 Clustering Infrastructure 3.3.1 Clustering for Reliable Storage Management Numerous fault tolerant, cluster based computing products have been developed in the past, with supports for membership agreement, distributed locking, and atomic resource fail over. Examples include VMS clusters [85], Microsoft cluster service [157] Sun clusters [148] HP high availability clusters [74] and IBM HA CMP [75] Often coupled with these systems are reliable, networkattached storage systems [152] such as EMC Celerra [52] Network Appliance Filer [114] Compaq ServerNet [122] and IBM SSA [76] ....

[Article contains additional citation context not shown here]

Nancy P. Kronenberg, Henry M. Levy, and William D. Strecker. VAXclusters: A closelycoupled distributed system. ACM Trans. on Computer Systems (TOCS), 4(4):130--146, 1986. 1.3.3, 3.3.1, 3.5.1, 3.6.1

....number of node failures and continues to serve users after network partition by relaxing the data consistency guarantees. Another approach is to build an email system on top of a cluster based operating system that supports membership agreement, distributed locking, and resource fail over (e.g. [Kronenberg et al. 1986; Vogels et al. 1998; Sun Microsystems 1999; IBM 1998] While this solution simplifies the architecture of the software, it tends to cost more than previous solutions because these systems run only on proprietary hardware. They also have limited Porcupine: A Highly Scalable, Cluster based Mail ....

....file system [Christenson et al. 1997] or protocol redirectors [Deroest 1996] As we demonstrate in this paper, this static approach is difficult to manage and scale and has limited fault tolerance. Numerous fault tolerant, clustered computing products have been described in the past (e.g. [Kronenberg et al. 1986; Vogels et al. 1998; IBM 1998; Sun Microsystems 1999] These clusters are often designed specifically for database fail over, have limited scalability, and require proprietary hardware or software. Unlike these systems, Porcupine s goal is to scale to hundreds or thousands of nodes using ....

KRONENBERG, N. P., LEVY, H. M., AND STRECKER, W. D. 1986. VAXclusters: A closely-coupled distributed system. ACM Trans. on Computer Systems 2, 4, 130--146.

No context found.

N.P. Kronenberg, H.M. Levy, and W.D. Strecker. VAXcluster: a closely-coupled distributed system. ACM Trans. Comput. Syst., 4(2):130--146, 1986.

No context found.

Kronenberg, N., Levy, H., Strecker, W., "VaxCluster: A Closely-Coupled Distributed System", ACM TOCS, V4.2, 1986.

No context found.

Kronenberg, N. P., Levy, H. and Strecker, W. D. "VAXclusters: A Closely-Coupled Distributed System". Trans. Computer Systems 4, 2 (May 1986), pp.130-146.

No context found.

Nancy P. Kronenberg, Henry M. Levy, and William D. Strecker. VAXclusters: A closely-coupled distributed system. ACM Transactions on Computer Systems, 4(2):130--146, May 1986.

No context found.

Nancy P. Kronenberg, Henry M. Levy, and William D Strecker. VAXclusters: A closely-coupled distributed system. ACM Transactions on Computer Systems, 4(2):130--146, May 1986. 127.

No context found.

Nancy P. Kronenberg, Henry M. Levy, and William D. Strecker. VAXclusters: A Closely-Coupled Distributed System. ACM Transactions on Computer Systems, 4(2), 1986.

No context found.

Nancy P. Kronenberg, Henry M. Levy, and William D. Strecker. VAXclusters: A closely-coupled distributed system. ACM Transactions on Computer Systems, 4(2):130-146, May 1986.

First 50 documents

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