D. E. Lowell and P. M. Chen. Free transactions with Rio Vista. In Proceedings of the 16th ACM Symposium on Operating Systems Principles, October 1997.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....allows even widespread failure to be quickly eradicated. The benefits of restarting quickly after failures have been recognized by many system designers, as they employed techniques ranging from the use of non volatile memory (e.g. Sprite s recovery box [10] derivatives of the Rio system [26, 67]) to non overwriting storage combined with clever metadata update techniques (e.g. the Postgres DBMS [96] Network Appliance s filers [55] A common theme, which we have identified in section 4.3, is that of segregating and protecting state that needs to be persistent, while treating the rest ....

D. E. Lowell and P. M. Chen. Free transactions with Rio Vista. In Proc. 16th ACM Symposium on Operating Systems Principles, Saint-Malo, France, 1997.

....single round trip instead of two, a factor of two in reduced latency is easily achieved. We also evaluate the utility of the partial write optimization in a more realistic benchmark setting. In addition to the PostMark benchmark, we also investigate the traffic savings on a debit credit benchmark [28], a TPC B like benchmark that is intended to model the workload on a database server that manages bank transactions [43] Measurements reveal that the partial write script reduces bandwidth considerably for both the PostMark and debit credit benchmarks. For PostMark, total message traffic (in ....

D. E. Lowell and P. M. Chen. Free Transactions With Rio Vista. In Proceedings of the 16th ACM Symposium on Operating Systems Principles (SOSP '97), pages 92--101, SaintMalo, France, October 1997.

....been extensively used in database applications; these are rather heavy weight transactions that typically involve writing to permanent storage. The concept of transactions has also been applied at the operating system level, where changes of virtual memories and even system calls can be undone [13, 17, 23]. Transactions used for the sake of error containment and recovery can be very fine grained, on the order of tens and hundreds of instructions. As such, we must reduce the overheads involved in implementing these transactions. By keeping transaction state in the speculative buffers of a TLS ....

....a loop) while we discuss only procedures here. Loop speculation is not relevant to the instrumentation code and transactions we have discussed here. Transactional memory implementations have been proposed using a variety of implementations: hardware[13, 26] software[24] and combinations thereof[17]. Rollback and recovery in large scale distributed systems have been studied extensively[8] Our goal is to support fine grained transactions with little or no overhead so they can become prevalent in everyday code. Speculative thread level parallelism is being considered by many microprocessor ....

D. E. Lowell and P. M. Chen. Free transactions with Rio Vista. In Proceedings of the 16th ACM Symposium on Operating Systems Principles, October 1997.

....too easy to modify data structures, potentially causing file system inconsistency. Wild references in the file system (or elsewhere in the operating system) could overwrite valid metadata in MRAM, corrupting the file system. HeRMES will avoid this problems using techniques similar to those in Rio [12]. By keeping file system MRAM protected except when explicitly necessary, HeRMES will ensure that only desired changes are made to MRAM. The process of switching a page from read only to read write in the page table is fast, and will not significantly slow down HeRMES MRAM operations, particularly ....

D. Lowell and P. Chen, "Free transactions with Rio Vista," 16th ACM Symposium on Operating Systems Principles, pp. 92--101, 1997.

....it is to issue bogus writes to disk blocks. If an uninterruptible power supply protects against the most common hardware failures, and the virtual memory system protects against software failures, then Rio can treat main memory as persistent storage, and provide very high performance transactions [LC97] Frigate is a user extensible filesystem based on persistent CORBA objects [KP97a] ACID stands for atomicity (a transaction appears either complete or never initiated) consistency (logical invariants over the database can be maintained) isolation (operations of transactions are not ....

D.E. Lowell and P.M. Chen. Free transactions with Rio Vista. In Proceedings of the Sixteenth ACM Symposium on Operating Systems Principles, pages 92--101, October 1997.

.... small synchronous disk writes impacts the performance of important applications such as recoverable virtual memory [27] persistent object stores [2, 18] and database applications [33, 34] These systems have become more complex in order to deal with the increasing relative cost of small writes [19]. Similarly, most existing file systems are carefully structured to avoid small synchronous disk writes. UFS by default delays data writes to disk. It is also possible to delay metadata writes if they are carefully ordered [9] The Log structured File System (LFS) 25] batches small writes. While ....

....to avoid having the small updates to the log conflict with reads. Our interest is in the limits to small write performance for a single disk. Because of the limitations imposed by disks, non volatile RAM (NVRAM) or an uninterruptable power supply (UPS) is often used to provide fast stable writes [3, 14, 15, 19]. However, when write locality exceeds bu#er capacity, performance degrades. There are also applications that demand stricter guarantees of reliability and integrity than that of either NVRAM or UPS. Fast small disk writes can provide a cost e#ective complement to NVRAM. Our basic approach is to ....

Lowell, D. E., and Chen, P. M. Free Transactions with Rio Vista. In Proc. of the 16th ACM Symposium on Operating Systems Principles (October 1997), pp. 92-- 101.

....to support standard POSIX system call semantics; the designers of QuickSilver found that the transaction model at times complicated this task. 2.4. 2 Rio Vista and Non Volatile RAM Lowell and Chen have developed what they term free transactions in the Rio Vista recoverable virtual memory system [Lowell97]. Rio Vista transactions provide the uncommon combination of durability and extremely high performance (5 10 s) by using non volatile RAM (NVRAM) for stable storage. NVRAM has the performance characteristics of RAM, but retains its contents in the face of system failure and power loss. Unlike ....

Lowell, D., Chen, P., "Free Transactions With Rio Vista," Proceedings of the Sixteenth ACM Symposium on Operating Systems Principles, pp. 92--101, St. Malo, France (October 1997).

....to the same le as well as replication. Furthermore, while several e orts have explored providing failure atomicity for le systems [9, 16, 22, 31, 38] as well as the underlying disk system [8, 21] most le systems limit failure atomicity to their meta data, if they provide it at all. RVM [30, 40] represents a fault tolerant alternative for managing application state by providing transactional guarantees for regions of virtual memory. However, since memory, like les, is inherently unstructured, RVM su ers from similar problems. As data is directly mapped into an application s address ....

D. E. Lowell and P. M. Chen. Free transactions with Rio Vista. In Proceedings of the 16th ACM Symposium on Operating Systems Principles, pages 92-101, Saint-Malo, France, Oct. 1997.

....However, DVCM is unique in its approach to assembling a tightly coupled computing platform out of commodity components by leveraging current trends in the architecture of these components. Previous research has explored lightweight kernels for transaction processing, like LRVM [17] and Rio Vista [12], and systems leveraging remote cluster memory, like Harp [11] and GMS [6] CRMem shares many concepts with these projects. However, transferring state sharing and global synchronization functionality closer to the network has not been previously explored for systems like these. By leveraging ....

D. E. Lowell and P. M. Chen. Free Transactions with Rio Vista. Proceedings of the 16th ACM Symposium on Operating System Principles, pages 92--101, Oct. 1997.

....approach also scales well to small shared memory multiprocessors. 1 Introduction We address the problem of building a reliable transaction server using a cluster of commodity computers, i.e. standard servers and system area networks (SAN) We use the Vista system as the transaction server [5]. Vista is a very high performance transaction system. It relies on recoverable memory [2] to avoid disk I O, thereby achieving its very high throughput. Because Vista does not store its data on disk, but rather keeps it in reliable memory, the data remains safe when the machine fails, but it is ....

....of that API in (single node) Vista, the relevant characteristics and performance of the hardware on which we did the experiments, and the benchmarks used. 2. 1 API We support a very simple API, first introduced by RVM [8] and later implemented by a variety of systems, including Vista [5]. The transaction data is mapped into the virtual address space of the server. The API contains the following routines: begin transaction( set range( commit transaction( abort transaction( begin transaction, commit transaction, and abort transaction implement the customary transaction ....

[Article contains additional citation context not shown here]

D. Lowell and P. Chen. Free transactions with Rio Vista. In Proceedings of the 16th ACM Symposium on Operating Systems Principles, Oct. 1997.

....RAM s slow writes and bulk erases. The Durable Memory RS 6000 uses batteries, replicated processors, memory ECC, and alternate paths to tolerate a wide variety of hardware failures [1] Finally, many papers have examined the performance advantages and uses of reliable memory [17] 6] 12] 2] [45], 46] 7.2 Field Studies of Failures Studies have shown that software is the dominant cause of system outages [23] 24] and several studies have investigated system software errors. Sullivan and Chillarege classify software faults in the MVS operating system; in particular, they analyze faults ....

D.E. Lowell and P.M. Chen, "Free Transactions with Rio Vista," Proc. 1997.

.... Box , in which the OS uses a protected area of non volatile memory to store crucial state needed for fast recovery [Baker92] This basic idea of segregating and protecting crucial hard state to simplify recovery reappears frequently, for example in the derivatives of the Rio system [Chen96] Lowell97] Lowell98] and in recent work on soft state hardstate segregation in Internet service architectures [Fox97] Gribble00] Most of these systems still use increased performance as a motivation for their techniques; recovery benefits are icing on the cake. The database community has long paid ....

D. E. Lowell and P. M. Chen. Free Transactions with Rio Vista. In Proc. of the 16th ACM Symp.on Operating Systems Principles, October 1997, pp. 92-101.

....three applications with almost no modi cation to the applications source code. Debit Credit and Order Entry are based on TPC B and TPC C, which are two widely used transaction processing benchmarks. We ran the two benchmarks on top of a lightweight, main memory transaction library, Rio Vista [27]. The Rio Vista is modi ed slightly by inserting two function calls, ignore region and register recovery func , to improve the performance (See Section 3.4) We chose the largest possible database sizes which t in main memory. Postgres [35] is an object relational database developed at ....

....DU AU 0.01 0.1 1 10 checkpoint interval (seconds) 0 20 40 60 80 100 Postgres Figure 5: Relative overhead for varying intervals with three transaction based applications on the SHRIMP cluster. The Debit Credit and Order Entry benchmarks run on top of the Rio Vista main memory database [27]. a ord to checkpoint every 0.05 seconds. As checkpoints are taken less frequently, the relative overhead drops because fewer page faults occur. The deliberate update protocol imposes higher overhead than the AU protocol on the Debit Credit and Order Entry applications. For example, the protocol ....

David E. Lowell and Peter M. Chen. Free Transactions With Rio Vista. In SOSP'97.

....These pieces of software were authored by other researchers at the University of Michigan. Discount Checking (DC) Lowell99] provides fast, user level, full process checkpoints. DC uses reliable main memory provided by Rio [Chen96] and lightweight and fast transactions provided by Vista [Lowell97] to provide fast checkpointing. We use Discount Checking as an example of a generic recovery mechanism in our evaluation of the fail stop property of software. Discount Checking uses transactions to implement checkpointing. The entire process state is mapped into persistent memory and ....

David E. Lowell and Peter M. Chen. Free Transactions with Rio Vista. In Proceedings of the 1997 Symposium on Operating Systems Principles, pages 92--101, October 1997.

No context found.

D. E. Lowell and P. M. Chen. Free transactions with Rio Vista. In Proceedings of the 16th ACM Symposium on Operating Systems Principles, October 1997.

No context found.

Lowell, D. E. and Chen, P. M. (1997). Free Transactions with Rio Vista. In Proc. of the 16th ACM Symposium on Operating Systems Principles, pages 92--101.

No context found.

Lowell, D.E., Chen, P.M. Free Transactions With Rio Vista. In Proceedings of the 16th ACM Symposium on Operating Systems Principles. St. Malo, France, October, 1997.

No context found.

D. Lowell and P. Chen. Free Transactions with Rio Vista. In the 16th symposium on Operating Systems Principles, Oct. 1997.

No context found.

D. Lowell and P.M. Chen. Free transactions with Rio Vista. In Proceedings of the 16th ACM Symposium on Operating Systems Principles, October 1997.

No context found.

Lowell, D. E., and Chen, P. M. Free Transactions with Rio Vista. In Proc. of the 16th ACM Symposium on Operating Systems Principles (October 1997), pp. 92-- 101.

No context found.

D. E. Lowell and P. M. Chen, Free transactions with Rio Vista, in Proc. 16-th Symposium on Operating Systems Principles, October 1997, pp. 92--101.

No context found.

David E. Lowell and Peter M. Chen, "Free Transactions with Rio Vista", Proceedings of the 1997.

No context found.

D. E. Lowell and P. M. Chen. Free transactions with Rio Vista. In Proceedings of the 16th Symposium on Operating Systems Principles (SOSP-97), volume 31,5 of Operating Systems Review, pages 92-101, New York, Oct.5-8 1997. ACM Press.

No context found.

David E. Lowell and Peter M. Chen. Free transactions with Rio Vista. In Proceedings of the 16th Symposium on Operating Systems Principles (SOSP-97), volume 31,5 of Operating Systems Review, pages 92--101, New York, October5--8 1997. ACM Press.

No context found.

David E. Lowell and Peter M. Chen. Free transactions with Rio Vista. In Proceedings of the 17th Symposium on Operating System Principles, October 1997.

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