Exam 10: Distributed DBMSs-Advanced Concepts and Object-Oriented DBMSs-Concepts and Design

Correct Answer:

Verified

See Sections 27.2 and 27.5. Expect some justification for the top three selection. Also expect detailed discussion of the advantages, not just bullet points.

Correct Answer:

Verified

• Conventional transaction management systems (TMSs) synchronize simple read and write operations. However, TMSs for advanced database applications must be able to deal with abstract operations. It may even be possible to improve concurrency by utilizing semantic knowledge about the objects and their abstract operations.
• Advanced database applications have different database access patterns than conventional database applications where the access is competitive (e.g., two users accessing the same bank account). Instead, sharing may be more cooperative as in the case, for example, multiple users accessing and working on the same design document. In this case, users accesses need to be synchronized, but users are willing to cooperate rather than compete for access to shared objects.
• Conventional transactions access 'flat' objects (e.g., pages, tuples) whereas transactions for advanced database applications may require synchronized access to composite and complex objects. Synchronization of access to such objects requires synchronization of access to the component objects.
• These applications require the support of long-duration transactions spanning hours, days or even weeks (e.g., working on a design object). Therefore, the transaction mechanism must support the sharing of partial results. Furthermore, to avoid the failure of partial tasks jeopardizing a long activity, it is necessary to distinguish between those activities that are essential for the completion of the transaction and those that are not, and to provide for alternative actions in case the primary activity fails.
• These applications may benefit from active capabilities for timely response to events and changes in the environment. This new database paradigm requires the monitoring of events and the execution of system triggered activities within running transactions.

Correct Answer:

Verified

(a) The basic objective of pointer swizzling is to convert main memory pointers representing interobject references to disk pointers when an object is being written to disk and subsequently to convert disk pointers to main memory pointers when the object is being read back in again. Thus, the program language can access the interobject references as though they are normal pointers, thereby improving performance.
(b) Three dimensions:
(i) eager v. lazy
(ii) direct v. indirect
(iii) copy v. in-place
1. Eager: Guarantees that all the pointers in the main memory are swizzled. When an object is loaded from disk, the object is scanned through and all the pointers in the object are immediately swizzled.
2. Lazy: Only swizzles pointers on demand; i.e., a pointer is not swizzled until the object it refers to is accessed via this particular pointer.

Advantage is that that no pointers are swizzled unnecessarily. On negative side, lazy swizzling must handle two different kinds of pointers at run-time: swizzled and non-swizzled.

3. Direct: Requires that the referenced object be resident in memory. A directly swizzled pointer contains the main memory address of the object it references.

Problem is that if an object is displaced from the system buffer (i.e., is no longer resident) all the directly swizzled objects that reference the displaced object need to be unswizzled.

In the case of eager direct swizzling, one cannot simply unswizzle pointers because eager swizzling guarantees that all pointers in the buffer are swizzled. Instead, these pointers (i.e., their home objects) must be displaced too - possible snowball effect.

4. Indirect: Permits the swizzling of pointers that reference non-resident objects.

Induces an additional overhead when it comes to simple object lookups - leads to an additional level of indirection (due to existence of a descriptor that stores the main memory address of object) and to a residency check, a check of whether the descriptor is valid or not. For direct swizzling, the information that an object is resident is coded in the swizzled pointer.
5. Copy: When faulting objects in, data is copied into the application's local object cache.
6. In-place: When faulting objects in, data is accessed within the data manager's cache.
Copy swizzling may be more efficient as, in worst case, only modified objects have to be swizzled back to their OIDs, whereas with in-place may have to unswizzle an entire page of objects if one object on the page is modified. On other hand, with copy approach, every object must be explicitly copied into the local object cache.
(c) Several alternative strategies (e.g.):
• Leave it to user.
• Map to relational system
• Use a one pointer strategy that leaves it to the run-time system to determine whether object is transient or volatile. Could trap references to persistent objects, for example, by using illegal values such as negative values for pointers to persistent objects.