W. Ng and P. Chen. Integrating reliable memory in databases. In Procs. of the International Conf. on Very Large Databases, pages 76--85, August 1997.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....is physically close to the area intended to be updated. These studies also categorized the field conditions that triggered the errors and the symptoms which brought the error to the attention of the user; however, this information does not have a significant bearing on our work. Ng and Chen [59] compare the reliability of database software using three different interfaces to a persistent cache (1) based on the I O model, 2) based on direct read write (without memory protection) and (3) based on direct read write with memory protection, using the POSTGRES database system. They inject ....

....limit case by caching more and more memory for longer periods of time. Comparing the two studies illustrates how difficult it is to predict errors scientifically. In particular, addressing errors, found to be 25 30 of the errors in [74] contributed to none of the errors found by Ng and Chen in [59]. Thus, even though Ng and Chen weighted their results by Sullivan s numbers, this category did not weigh in their results at all. Almost certainly this results from the difference between a fault injection scenario and field errors. If a certain class of error causes the system to fail reliably ....

[Article contains additional citation context not shown here]

W. Ng and P. Chen. Integrating reliable memory in databases. In Procs. of the International Conf. on Very Large Databases, pages 76--85, August 1997.

....or fatal consequences. In this study, we define corruption in database image and long DBMS unavailability during recovery (see Sec. 5.5 for double failures) as catastrophic events. Fault injection studies performed on DBMSs demonstrate the possibility of safety violations in DBMSs. Ng and Chen [12] showed that the database got corrupted in 2.3 to 2.7 cases even when a reliable memory for caching data was integrated into POSTGRES DBMS. A study conducted by the authors on ClustRa DBMS points out the necessity of introducing some form of consistency check in order to guard the database image ....

W. T. Ng and P. M. Chen. Integrating Reliable Memory in Databases. Proc. of the 23rd VLDB Conference, 1997, 1997.

....injection study performed by Ng and Chen to study the reliability of a non volatile buffer cache in a DBMS, the addi tion of hardware protection only reduced the incidence of data corruption from 2.7 to 2. 3 , indicating that a significant risk exists that physical corruption will go undetected [16]. Since they used a hardware protection scheme following [21] some of this corruption may have occurred during the time the page was unprotected. Such errors may be detected by the codeword schemes in this paper if the area being corrupted, while on the same page, is not specified as part of the ....

....limit on portability, and since the technique requires a number of free registers to perform well, it may not be applicable on architectures without a large register set, such as the Intel x86 architecture. By contrast, our techniques are language and instruction set independent. Ng and Chen [16] study the reliability of three different interfaces to a persistent cache (1) based on the I O model, 2) based on direct read write (without memory protection) and (3) based on direct read write with memory protection, using the POSTGRES database system. They inject a variety of faults ....

W. Ng and P. Chen. Integrating reliable memory in databases. In Procs. of the International Conf. on Very Large Databases, pages 76--85, Aug. 1997.

....tolerance by replication in their systems, e.g. Tandem Remote Duplicate Database Facility, Sybase Replication Server, Oracle7 Symmetric Replication Facility , IBM Remote Site Recovery, and Informix Online Server, there are very few studies highlighting the above mentioned aspects. Ng and Chen [9] integrated reliable memory for caching data into Postgress DBMS in three different ways and used fault injection to evaluate them. Chillarege [2] studied the failure characteristics of the commercial transaction processing system, IBM sIMS, Version 1.3. These studies were performed on centralized ....

W. T. Ng and P. M. Chen. Integrating Reliable Memory in Databases. Proc. of the 23rd VLDB Conference, 1997, 1997.

....access speed, and cost dictate the mechanism to be used for recovery [CL99] Unfortunately, DBMSs have typically found it difficult to make use of hardware improvements because changing the recovery logic would involve rewriting much of the DBMS. For example, although researchers [SR86, NC97] have demonstrated orders of magnitude performance improvements by using battery backed memories instead of disk for storing logs, commercial vendors have not implemented these. Separating the recovery logic makes it easy to change as hardware technology improves. A second goal of our project is ....

....if we use replication directly for recovery, instead of duplicating our efforts in single node recovery and multi node replication. 1.2 Design decisions We use replication and non volatile memory for recovery, instead of writing logs. Non volatile memory is fairly inexpensive nowadays [NC97] and memory can be made safe against OS crashes using virtual memory protection [C 96] Using replication instead of logging, and assuming memory non volatility, makes the ACID memory much simpler to implement. We provide support for concurrency control along with recovery; the unit of ....

W. Ng and P. Chen. Integrating reliable memory in databases. In VLDB, 1997.

....the state of its buffer pool. Following a crash, performance is reduced until that buffer pool has been repopulated, which, for a large buffer pool, may take quite a while. Various workers in the field have explored ways to exploit a non volatile memory, for example eNVy [19] Sprite [3] and Rio [7, 16]. Their approach is to place the file system cache in non volatile memory and to modify the operating system and file system to preserve this memory across reboots. Although the level of protection that can be provided to such a cache may be adequate for most purposes [19] this approach does not ....

W. T. Ng and P. M. Chen. Integrating Reliable Memory in Databases. In Proceedings of the 23rd VLDB, 1997.

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