J. J. Kistler. Disconnected Operation in a Distributed File System. PhD thesis, Carnegie Mellon University, 1996.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....very weak, connectivities to the servers. In these cases, updates are applied immediately to locally cached objects, and are also logged in a clientmodify log (CML) The logging mechanism allows updates to be eventually propagated to the servers that maintain the primary replica of the objects [11, 24]. This eventually propagation is called reintegration. A CML consists of records called CML entries, each is recording the effect of a mutating filesystem operation. For example, a chmod operation is logged as a CHMOD record, and a mkdir operation is logged as a MKDIR record. Figure 2 lists some ....

....record type STORE is special. First, it is recording the effect of a sequence of operations: a mutating open, interspersed possibly with several write s, and a final close operation on a file. Coda maps the whole sequence of operations to a single STORE record because it uses a session semantics [11]. Second, a STORE s associated data includes the content of the file being stored. In contrast, other record types do not include the content but only some directory attributes such as the owner or the name of an object. Note that file contents can be as large as many kilobytes or Coda supports ....

[Article contains additional citation context not shown here]

J. J. Kistler. Disconnected Operation in a Distributed File System. PhD thesis, Carnegie Mellon University, School of Computer Science, 1993.

....FR assumes that semi disconnected operation will be more common than full disconnected operation, and primarily stores encryption keys on such nodes which facilitate use of untrusted computing resources in its environment. 5.2. 1 Coda It might seem as if FR has striking similarities with Coda [78, 77]. However, the design philosophies behind the two systems are almost antithetical, and this becomes visible when parts of the system fail. The most important observation underpinning the design of Coda, is that conflicting updates on files are rare. That is, if a file is taken offline for a day ....

....rare. That is, if a file is taken offline for a day or two, chances are overwhelmingly large that a modified version can be copied back in place without any conflict. In fact, the authors report that within a week, the probability that two different users modify the same file is less than 0. 2 [77]. The smooth working of Coda relies on this being true. Coda is designed after a model where the laptop is in focus. Users have copies of their files on the laptop and bring the machine home in the afternoon. It is to be expected that files in a user s home directory remain unaltered. If, ....

KISTLER, J. J. Disconnected operations in a distributed file system, vol. 1002 of Lecture Notes in Computer Science. Springer Verlag, 1996.

....le systems, and or for le systems that provide built in support for concurrency [Stonebraker81, Stonebraker86] 3. Mobility: replicated and distributed le systems with disconnected and caching operations gure heavily in an environment where network latency and reliability is highly variable [Satyanarayanan90, Kistler91, Tait91, Tait92, Kistler93, Zadok93a, Kuenning94, Marsh94, Mummert95]. 4. Security: more secure le systems are sought, especially ones that securely export les over the network [Steiner88, Haynes92, Glover93, Takahashi95] An easy way to use encryption in le systems [Blaze93, Gutmann96, Boneh96] and the ability to provide special semantics via facilities such ....

J. J. Kistler. Disconnected Operation in a Distributed File System (Report CMU-CS-93156) . Technical report. Carnegie Mellon University, Pittsburgh, U.S., 1993.

....We conclude the paper with a brief review of related work and a summary of the results. 2B ACKGROUND 2.1 Coda File System Coda is an experimental distributed Unix file system providing a single location transparent name space to a large collection of clients. Support for disconnected operation [6], 7] 16] is primarily the responsibility of the user level cache manager, Venus, at each client. Venus allows applications to continue accessing cached data during periods of disconnection; it also records updates performed while disconnected and replays them upon reconnection to servers ....

....RVM log traffic. This strategy is used in two ways in the current IOT implementation. First, we use RVM in the asynchronous flush mode. In this mode, a separate daemon thread is dedicated to performing asynchronous disk flushes for buffered RVM updates, with a maximum delay bound of 30 seconds [6]. The second technique applies the same principle at a higher level: Deferring updates to the persistent data structures of SG in such a manner that a clean SG can still be reconstructed upon restart after a crash. File Access Monitoring. Minimizing the performance overhead of file access ....

[Article contains additional citation context not shown here]

J.J. Kistler, "Disconnected Operation in a Distributed File System, " PhD thesis, Carnegie Mellon Univ., Pittsburgh, May 1993.

....digital assistants, and notebook computers. An implicit assumption underlying mobile computing is that the mobile object is always at precisely one location. Replication is less an issue, except when dealing with fault tolerant issues as, for example, in the case of disconnected file operations [21]. Location management in mobile computing generally follows a home based approach. This means that the system assumes that there is always a home location that keeps track of the object s current location. Once the present location has been found through the home location, messages can be ....

J. Kistler. Disconnected Operations in a Distributed File System, volume 1002 of Lect. Notes Cornput. Sc. SpringerVerlag, Berlin, 1996.

....at a certain locality) with varying and multiple network connectivity and with different characteristics and costs [16] True prescience, of course, requires knowledge of the future. Two possible techniques, LRU caching and hoarding, are for example present in the Coda cache manager [30]. In order to effectively support mobile computers, system designers must view the network as a first class resource, expending CPU and possibly disk resources to reduce the use of network resources during periods of poor network connectivity. Modern high performance network protocols require ....

Kistler J.J.: "Disconnected operation in a distributed file system", PhD thesis, Carnegie Mellon University, School of Computer Science, 1993.

....in the clients of a distributed system to periodically send trace records to a collection unit. Such a system has the effect of passing the problems of trace data volume to the collector, a machine that would not necessarily be among those being traced. Kistler and Satyanarayanan [48] and Kistler [49] have used this technique to aid research into a new file system design. Kernel instrumentation has been combined with other techniques to provide additional information. Floyd and Ellis [37] combined kernel instrumentation with the use of benchmarks in a study on file reference patterns. Floyd ....

Kistler, J. J. Disconnected Operation in a Distributed File System. PhD thesis, School of Computer Science, Carnegie Mellon University, May 1993. Also available as CMU technical report CMU-CS-93-156.

....initiated concurrently at different sites, there is a need for reconciliation when write write conflicts occur. Likewise, Coda supports disconnected operation allowing a client to continue updating its local copy of a file, without any guarantees that another client may also be performing updates [11]. More important is that in these examples replication of files is again hard coded into the system. At best, they allow a callback to user provided component for conflict resolution. 2.2 Content Delivery Networks More recently, content distribution networks (CDNs) are providing services that ....

J. Kistler. Disconnected Operation in a Distributed File System, volume 1002 of Lect. Notes Comp. Sc. Springer-Verlag, Berlin, 1996.

....example, consider the create of a file, followed by a store. If they are followed by an unlink, all three CML records and the data associated with the store can be eliminated. Both trace driven simulations and measurements of Coda in actual use confirm the effectiveness of log optimizations [4, 8] Venus combines implicit and explicit sources of information into a priority based cache management algorithm. The implicit information consists of recent reference history, as in least recently used (LRU) caching algorithms. Explicit information takes the form of a perclient hoard database ....

....opportunity to be eliminated at the client. A good design must balance two factors. On one hand, records should spend enough time in the CML for optimizations to be effective. On the other hand, updates should be propagated to servers with reasonable promptness. Our solution, illustrated in Fig. 4, uses a simple technique based on aging. A record is not eligible for reintegration until it has spent a minimal amount of time in the CML. This amount of time, called the aging window (A) establishes a limit on the effectiveness of log optimizations. Based on the results of trace driven ....

[Article contains additional citation context not shown here]

J. J. Kistler, "Disconnected Operation in a Distributed File System," Ph.D. thesis, Dept. of Comp. Sci., Carnegie Mellon Univ., May 1993.

....at a certain locality) with varying and multiple network connectivity and with different characteristics and costs [11] True prescience, of course, requires knowledge of the future. Two possible techniques, LRU caching and hoarding, are for example present in the Coda cache manager [30]. In order to support mobile computers effectively, system designers must view the network as a first class resource, expending CPU and possibly disk resources to reduce the use of network resources during periods of poor network connectivity. Decomposition In a mobile multimedia system many ....

Kistler J.J.: "Disconnected operation in a distributed file system", PhD thesis, Carnegie Mellon University, School of Computer Science, 1993.

....information management, communication protocol research, resource allocation and scheduling under real time constraints, and disconnected operation. Many research efforts in these areas are currently underway [5] 16] 23] 24] Disconnected operation, for example, is investigated in the CODA project [13][20] at Carnegie Mellon University. Some of the scheduling issues in crisis management systems are described in [4] This paper concentrates on the information management aspect, specifically distribution and maintenance of dynamic, shared information. 3 Data Classification for Cooperative ....

J. Kistler, "Disconnected Operation in a Distributed File System," Tech. Report #156, Carnegie Mellon University, 1993.

....data for read requests, and by logging the update operations for write requests; 3) after the disconnection, CHAPTER 1. INTRODUCTION 7 the client sends the logged updates to the file server. James Kistler has demonstrated that disconnection operation can indeed be used effectively in a DFS [18]. His work is ground breaking, and has triggered many other research projects using similar concepts, some in the context of DFSs, some in other contexts [15, 59] 1.3.2 Weakly connected Operation Disconnected operation is great. However, as Lily Mummert put it, it is not a panacea. 31, 30] A ....

....in other contexts [15, 59] 1.3.2 Weakly connected Operation Disconnected operation is great. However, as Lily Mummert put it, it is not a panacea. 31, 30] A mobile client operating in a disconnected mode suffers from many limitations. First, although there is already the concept of hoarding [18] or predictive caching cache misses still have non zero probability to happen. When they happen, they impede the users work. Second, updates made by a user on the disconnected client are delayed from propagating to the server. The second limitation manifests in several ways. First, users ....

[Article contains additional citation context not shown here]

J. J. Kistler. Disconnected Operation in a Distributed File System. PhD thesis, Carnegie Mellon University, School of Computer Science, 1993. BIBLIOGRAPHY 182

....background We have implemented our prototype by extending the Coda File System [3] Although our experience is based on Coda, the general principles should also be applicable to other mobile file systems. We briefly describe Coda in this section; more information is available in the literature [3, 8, 7, 14, 13]. The Coda model is that there are many clients and a few servers. On each client, a cache manager, called Venus, carefully manages and persistently stores cached file system objects. To support mobile computing, Coda clients can be used in disconnected and weakly connected mode, where Venus ....

....This measure is to ensure that the target file will not be destroyed accidentally by a futile execution. If a temporary file is created and subsequently deleted during the execution of a user operation, its modification records will be deleted by the existing identity cancellation procedure [7]. They will not appear in the twoCMLs and will not cause naive rejections of re execution. However, if a temporary file is created and subsequently renamed during the execution of a user operation, its modifications records will be present in the CMLs, and might cause our validation to reject the ....

[Article contains additional citation context not shown here]

J. J. Kistler. Disconnected Operation in a Distributed File System. PhD thesis, Carnegie Mellon University, School of Computer Science, 1993.

....at a certain locality) with varying and multiple network 5 connectivity and with different characteristics and costs [8] True prescience, of course, requires knowledge of the future. Two possible techniques, LRU caching and hoarding, are for example present in the Coda cache manager [14]. In order to effectively support mobile computers, system designers must view the network as a first class resource, expending CPU and possibly disk resources to reduce the use of network resources during periods of poor network connectivity. Modern high performance network protocols require that ....

Kistler J.J.: "Disconnected operation in a distributed file system", PhD thesis, Carnegie Mellon University, School of Computer Science, 1993.

....Section 3.7 provides an analysis of our approach, and Section 3.8 presents conclusions. 68 3. 1 Data Centric Database Sharing The growing trend towards mobile computing has resulted in the adaptation of many traditional applications, including replicated database views, to intermittently connected[95, 96, 117, 137] environments. Recall the example given in Chapter 1 and depicted in Figure 23 where a server database system shares portions of data with a set of intermittently connected clients. Since the connectivity is intermittent, the clients maintain a copy of the shared data in their local database. ....

J. J. Kistler. Disconnected Operation in a Distributed File System. PhD thesis, School of Computer Science, Carnegie Mellon University, May 1993. CMU-CS-93-156.

....increases the availability of data during partial network or server failure. If all servers become inaccessible, the second mechanism allows Venus, the client cache manager, to service file requests using cached data. Details of server replication and disconnected operation are provided in [5] and [3, 4] respectively. 3. Overview of ASR mechanism On every cache miss Venus verifies that all replicas of the object being fetched are identical. If the replicas diverge, Venus automatically forks an appropriate ASR to resolve them. The application for which the object is being fetched blocks until ....

Kistler, J. J. Disconnected Operation in a Distributed File System. PhD thesis, School of Computer Scient, Carnegie Mellon University, Pittsburgh, 1993.

....operations. G. Object Prefetching An essential component to accomplishing useful work while disconnected is having the necessary information locally available. This goal is usually accomplished during periods of network connectivity by filling the mobile host s cache with useful information [50]. Applications should decide which objects to prefetch. We believe the usability of applications will be critically dependent upon simple user interface metaphors for indicating collections of objects to be prefetched. Requiring users to directly list the names of objects that they wish to ....

....implemented as an ordinary UNIX file. Rover performs both a flush and a synchronize operation to force new QRPCs to the disk. Thus, the flush is on the critical path for message sending. The prototype implementation favors simplicity over performance: it does not perform any compression on the log [50] and it does not employ efficient techniques for implementing stable storage (e.g. Flash RAM [58] or group commit [59] 3. Execution cost. The time to process the QRPC at the server. Combining the transport and logging costs and examining the total costs, we measured the round trip time for a ....

J. J. Kistler, Disconnected Operation in a Distributed File System, Ph.D. thesis, School of Computer Science, Carnegie Mellon University, May 1993.

....are its main concern. Coda supports data reintegration, which is the process to propagate any updates during the disconnection period and integrate the data from clientside back to the server upon the communication channel reestablisment. Note that in Coda, it retains certain readfrom relations [9] between file system objects. Based on its transactional file system model, data reintegration is made possible only if all the updates satisfy the one copy view serializability condition [9] If the updates cannot satisfy the condition, the whole update transaction in the client side is rejected ....

....upon the communication channel reestablisment. Note that in Coda, it retains certain readfrom relations [9] between file system objects. Based on its transactional file system model, data reintegration is made possible only if all the updates satisfy the one copy view serializability condition [9]. If the updates cannot satisfy the condition, the whole update transaction in the client side is rejected to avoid data inconsistency. Ficus, on the other hand, uses reconciliation process to synchronize each volume replica. Conflicts are defined on a file by file basis and only individual file ....

J. J. Kistler, Disconnected Operation in a Distributed File System, PhD thesis, School of Computer Science, Carnegie Mellon University. May 1993

....as a file system. 106 5.3.7 Name Caching Name caching component deals with similar issues as the caching component described above. However, there are subtle differences between the two components. Directories and files differ in how they are modified and what semantics is expected of them [Kistler, 93] A file is normally treated as an indivisible object. When a file is modified, an old copy is considered useless. On the other hand, directories are slowly changing objects with more predictable patterns of change. They differ from files in three respects. First, directories change more slowly ....

J. Kistler, Disconnected Operation in a Distributed File System, Ph.D. Thesis, School of Computer Science, Carnegie Mellon University, PA.

....digital assistants, and notebook computers. An implicit assumption underlying mobile computing is that the mobile object is always at precisely one location. Replication is less an issue, except when dealing with fault tolerant issues as, for example, in the case of disconnected file operations [24]. Location management in mobile computing generally follows a home based approach. This means that the system assumes that there is always a home location that keeps track of the object s current location. Once the present location has been found through the home location, messages can be ....

J. Kistler. Disconnected Operations in a Distributed File System, volume 1002 of Lect. Notes Comput. Sc. Springer-Verlag, Berlin, 1996.

....of the ward set. Such replication flexibility, called selective replication [7, 8] provides improved efficiency and resource utilization: ward members locally store only those objects in the ward set that they actively require. Replication decisions can be made manually or with automated tools [3, 5]. Since the ward set varies dynamically, it follows that different wards might store different sets of data as well; that is, not all ward sets will be equivalent. In other words, the model provides selective replication between wards themselves. The reconciliation topologies and algorithms [7] ....

J. J. Kistler. Disconnected Operation in a Distributed File System. Ph.D. dissertation, Carnegie-Mellon University, May 1993.

....to replicate information storage introduces a number of issues to address. Below, we survey some of those issues, and briefly consider approaches to solution. 3. 1 Optimistic Update It is a conventional wisdom that, at least for many files and applications, concurrent update is quite rare [17] [21], 23] 25, 29] If this pattern is generally true in practice, and those rare problems which do occur can be corrected by reasonable means, then this entire approach has promise. Fortunately, data collected in the studies at UCLA and Locus indicate that conflicts are generally rare both in ....

....the user will always know better what his or her intent is likely to be. Note however that the user is not best able to specify the content of a task s information cluster, as users are often not even aware of what files, utilities, linked information, and the like are in use on his or her behalf [21] Fortunately, the system can be of considerable aid in this respect [27] Therefore, we expect that an effective predictive caching system can be developed, which together with the user can effectively decide what files should be replicated for the mobile user. 3.3 Network Naming Navigation Once ....

Kistler J, Disconnected Operation in a Distributed File System. Ph.D. dissertation, Carnegie-Mellon University, May 1993.

....JetFile takes an optimistic approach to file updates. JetFile promises to detect and report update conflicts, but only after the fact that they have occurred. The optimistic approach assumes that write sharing will be rare and uses this fact to hide network latencies. Experience from Coda [23, 19] and Ficus [29] tell us that in their respective environments write sharing and update conflicts were rare and could usually be handled automatically and transparently. JetFile is designed to support large caches (on the order of gigabytes) to improve availability and to avoid the effects of ....

....to two files per second, which is almost unusable. It is arguable that these kind of optimistic approaches are dangerous and should be avoided because of the potential update conflicts that may arise. There is however empirical evidence indicating that sequential write sharing is rare, Kistler [19] and Spasojevic et. al [33] instrumented AFS to record and compare the identities of users updating files and directories. Kistler found that over 99 of all file and directory updates were by the previous writer. Spasojevic found that over 99 of all directory modifications were by the previous ....

J. J. Kistler, Disconnected Operation in a Distributed File System, PhD thesis, Carnegie Mellon University, May. 1993.

....will sustain a number of clients that is significantly larger. A prototype has been developed and performance studies are currently in progress. 1 Introduction The growing trend towards mobile computing has resulted in the adaptation of many traditional applications to intermittently connected [1, 2, 3, 4] environments, including replicated database views. Consider a database environment where a server database Supported by the Georgia Research Alliance under a Technology Development Investment Grant. y Currently working at Informix Software Inc. Portland, Oregon 97205. system shares ....

J. J. Kistler, Disconnected Operation in a Distributed File System. PhD thesis, School of Computer Science, Carnegie Mellon University, May 1993. CMU-CS-93-156.

....Figure 3: Daily High Water Marks of Resolution Log Sizes that over 99 of all operation logs grow less than 240KB per day. 4.3 Disconnected Operation Mutations made during disconnected operation at a Coda client are recorded in a per volume replay log. Coda employs many cancellation optimizations[8] to reduce the amount of log space used. Upon reconnection, the client transparently invokes reintegration of each modified volume. If a volume s log is successfully replayed by the servers, they proceed to backfetch the contents of modified files. If replay fails, the log and associated files are ....

KISTLER, J. J. Disconnected Operation in a Distributed File System. PhD thesis, Carnegie Mellon University, School of Computer Science, 1993.

....digital assistants, and notebook computers. An implicit assumption underlying mobile computing is that the mobile object is always at precisely one location. Replication is less an issue, except when dealing with fault tolerant issues as, for example, in the case of disconnected file operations [21]. Location management in mobile computing generally follows a home based approach. This means that the system assumes that there is always a home location that keeps track of the object s current location. Once the present location has been found through the home location, messages can be ....

J. Kistler. Disconnected Operations in a Distributed File System, volume 1002 of Lect. Notes Comput. Sc. SpringerVerlag, Berlin, 1996.

....was borrowed from the Bayou room scheduling tool [12] This informs users that the appointment might be canceled due to a conflict. E. Object Replication and Consistency An essential component to accomplishing useful work while disconnected is having the necessary information locally available [33]. RDO replica caching is the chief technique available in Rover to achieve high availability, concurrency, and reliability. In this section, we discuss strategies for selecting objects to replicate and for reducing consistency related costs. E.1 Replication RDO replication is accomplished during ....

....semantics directly to servers is an important functional advantage, especially in the presence of intermittent connectivity. However, it may lead to an operation log that grows in size at a rate exceeding that of a simple write ahead log. The traditional approach is log compaction [33]. Rover takes a different approach by directly involving applications in log compaction. Applications can download procedures into the access manager to manipulate their log records. For example, an application can filter out duplicate requests (e.g. duplicate QRPCs to verify that an object is ....

J. J. Kistler, Disconnected Operation in a Distributed File System, Ph.D. thesis, School of Computer Science, Carnegie Mellon University, May 1993.

....is cheap. True prescience, of course, requires knowledge of the future. In practice, we approximate this knowledge in Coda using the following three strategies. Two of these techniques, LRU caching and hoarding, are already present in the Coda cache manager and have been described elsewhere [5, 4]. The first of these, LRU caching, automatically keeps objects that have been recently accessed in the cache. The second of these, hoard ing, allows the user to supply a prioritized list of files that should be present in the cache in case of a disconnection. Coda transparentlycombines these ....

....new data available at the server suggests that updates be propagated eagerly, doing so would render ineffective the log optimizations used to coalesce updates. There is substantial evidence from our experience with disconnected operation indicating that these optimizations are highly effective [4]. To keep network communication to a minimum, we monitor the compressibility of the replay log over time, and defer propagation of updates until the log is quiescent. Each of these techniques requires us to expend CPU and disk resources in order to reduce network demands. Minimizing data ....

KISTLER, J. J. DisconnectedOperation in a Distributed File System. PhD thesis, Carnegie Mellon University, School of Computer Science, 1993.

....granularity of subtrees called volumes. At each client, a cache manager, Venus, dynamically obtains and caches data as well as volume mappings. 4.1. Disconnected Operation Disconnected operation, a concept first conceived and demonstrated in Coda, is an important initial step in mobile computing [6, 7, 17]. In this mode of operation, a client continues to have read and write access to data in its cache during temporary network outages. Transparency is preserved from the viewpoint of applications because the system bears the responsibilities of propagating modifications and detecting update ....

Kistler, J.J. Disconnected Operation in a Distributed File System. PhD thesis, Department of Computer Science, Carnegie Mellon University, May, 1993.

....three running beta test codeand storing 2GB of data replicated three ways, and four machines used only for alpha testing. Our current release policy promotes alpha to beta and beta to production every month. Coda remains a very active platform for research; two Ph.D. theses have been published[10, 12], and three other students are currently finishing their doctoral research within Coda. Needless to say, the system is undergoing active development. 3.4 Why Coda Was Measured When first proposed, Coda was met with some skepticism by the academic community, which raised several pointed questions. ....

KISTLER, J. J. Disconnected Operation in a Distributed File System. PhD thesis, Carnegie Mellon University, School of Computer Science, 1993.

....determine when application components are removed from the cache. Another important direction of our future research are prefetching strategies. ACHILLES currently allows users to manually prefetch applications, but does not use any automatic prefetching mechanism as for example employed in Coda [5,6]. In the future we hope to use the information maintained in the resource dependency graph for developing adaptable prefetching strategy as part of ACHILLES. Finally, we hope to make resource qualification a dynamic process [2,4] In it s current form, resource qualification is performed once ....

J.J. Kistler. Disconnected Operation in a Distributed File System. PhD. Thesis, Carnegie Mellon University, May 1993.

No context found.

J. J. Kistler. Disconnected Operation in a Distributed File System. PhD thesis, Carnegie Mellon University, 1996.

No context found.

J.J. Kistler, "Disconnected Operation in a Distributed File System," PhD thesis, Carnegie Mellon Univ., School of Computer Science, 1993.

No context found.

J.J. Kistler, Disconnected Operation in a Distributed File System. Technical Report CMU-CS93 -156. School of Computer Science, Carnegie Mellon University, 1993. http://www.cs.cmu.edu/ afs/cs.cmu.edu/project/coda/Web/docs-coda.html

No context found.

J. J. Kistler. Disconnected Operation in a Distributed File System. PhD thesis, Carnegie Mellon University, 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