摘要:
A method and system for improving memory access patterns of software systems on NUMA systems discovers NUMA system resources, where the NUMA system resources comprises a plurality of NUMA nodes; determines a plurality of database threads, processes, and objects for a database configuration; and generates a policy which assigns the plurality of database threads, processes, and objects to the plurality of NUMA nodes, wherein the generating is performed prior to initialization of the plurality of database threads, processes, and objects. The assignment of the database threads, processes, or objects to NUMA nodes is such that the amount of remote memory accesses is reduced. When the database thread, process, or object initializes, the database server queries the policy for its assigned NUMA node(s). The database thread, process, or object is then bound to the assigned NUMA node(s). In this manner, the costs from remote memory accesses are significantly reduced.
摘要:
A system and method for assessing the activity level of a database management system. The number of page fix operations performed by execution units of a database management system can be counted and used to compute a measure for the activity level of the database management system. In one embodiment of the invention, multiple counters are used to count the page fix operations. This and other embodiments of the invention facilitate activity level assessment without incurring significant performance penalty.
摘要:
A computer system dynamically scales back latch requests for system resources. Tasks seeking access to system resources each dynamically determine the probability that the task will gain access to the latch relating to a given system resource. Where the task estimates that its probability is below a defined threshold, the task will suspend itself for a defined sleep time. The task dynamically adjusts the length of the sleep time based on the number of times the task enters the suspended state and on the relative changes in the estimated probability that the task will gain access to the resource.
摘要:
Provided are a method, system, and program for handling access requests to a page while copying an updated page of data to storage. An update to a page in a memory is received and indication is made that the page has an exclusive lock. The update is applied to the page and an operation is initiated to copy the updated page to a physical log in a storage in response to applying the update to the page. Reads and writes are allowed to proceed against the page while the operation to copy the updated page to the physical log in the storage is occurring.
摘要:
For an information retrieval system coupled to a buffer pool maintaining a plurality of pages of recently accessed information for subsequent re-access, a technique for starting the buffer pool is provided. The technique facilitates a quicker start to the buffer pool by deferring allocation of page storing portions, for example, until they are needed. The technique makes the buffer pool available for storing pages while deferring allocation of a page storing portion of the buffer pool and allocates the page storing portion of the buffer pool in response to a demand to store pages to the buffer pool. The technique may be used to re-start a buffer pool with pages stored to a memory coupled to the information retrieval system where the pages where stored upon a buffer pool shutdown. Further, buffer pool readers or prefetchers may be configured to read pages for storing to the buffer pool and allocate the page storing portions of the buffer pool in response to instructions to read particular pages.
摘要:
A method, computer program product, and system for detecting and fencing off a failed entity instance so that failover time in the computing system is reduced. Upon detection of a failed entity, for example a failed process, a signal handler invokes an operating system interface to fence off the failed instance from modifying its persistent state, so that the entity may be re-instantiated prior to termination of the failed instance. This approach reduces failover time and eliminates split-brain problems without compromising access to the failed instance to obtain diagnostic information, core dumps, and the like.
摘要:
An ordering of operations in log records includes: performing update operations on a shared database object by a node; writing log records for the update operations into a local buffer by the node, the log records each including a local virtual timestamp; determining that a log flush to write the log records in the local buffer to a persistent storage is to be performed; in response, sending a request from the node to a log flush sequence server for a log flush sequence number; receiving the log flush sequence number by the node; inserting the log flush sequence number into the log records in the local buffer; and performing the log flush to write the log records in the local buffer to the persistent storage, where the log records written to the persistent storage comprises the local virtual timestamps and the log flush sequence number.
摘要:
The use of a centralized version table allows for efficient object switching. Rather than synchronizing all database agents to recognize a newly created file as containing the most recent version of a given object, database agents requiring access to the given object need only consult the centralized version table to learn file identity information. That is, the database agents consult the centralized version table to determine which of the files associated with a given object contain the most recent version of the given object. Mechanisms associated with the use of the centralized version table also provide for efficient recovery from a failure that has occurred during an object switching transaction.
摘要:
A system and associated method are provided for directing a database management system, to relocate buffer pages that are pinned in a buffer pool of a data processing system. Each of the buffer pages has a respective page descriptor for indicating the location of the buffer page in the buffer pool. Once the pages are relocated, the now free system memory of the buffer pool is resized. Prior to resizing, a selected pinned page is latched by an agent of the system, wherein the selected page of the buffer pages is in a resize region of the buffer pool. The pinned and latched page becomes a fixed page. The agent for the database management system determines a suitable relocation region of the buffer pool for the fixed page. A resizer module copies the contents of the fixed page to the relocation region and changes the respective page descriptor to indicate the relocation region. The resizer module performs the relocation of other pages in the resize region to allow dynamic resizing of the buffer pool while maintaining the presence of pinned versions of the pages in the buffer pool prior to resizing.
摘要:
Techniques are provided for accessing sector data. An embedded storage function is received. One or more data management functions are generated in response to receiving the embedded storage function. The one or more data management functions are invoked to retrieve the sector data from a sector table.