International Business Machines Corporation

Greg Robert Mewhinney

Mysore Sathyanaranyana Srinivas

David Blair Whitworth

Diane Garza Flemming

Bruce Gerard Mealey

James Anthony Pafumi

Andrew Dunshea

Eric Philip Fried

Jos Manuel Accapadi

William A Maron

Mysore Sathyanarayana Srinivas

Suresh Eswara Warrier

Brian Chase Twichell

Jacob Jason Rosales

David Nevarez

Randal Craig Swanberg

Larry Bert Brenner

James William Van Fleet

David William Mehaffy

Bimal Kiran Doshi

A method, system and program are provided for controlling access to a specified cache level in a cache hierarchy in a multiprocessor system by evaluating cache statistics for a specified application at the specified cache level against predefined criteria to prevent the specified application from accessing the specified cache level if the specified application does not meeting the predefined criteria.

Use of cache statistics to ration cache hierarchy access

A method utilizes cluster-awareness to effectively support a live partition mobility (LPM) event and provide recovery from node failure within a Virtual Input/Output (I/O) Server (VIOS) cluster. An LPM utility creates a monitoring thread on a first VIOS on initiation of a corresponding LPM event. The monitoring thread tracks a status of an LPM and records status information in the mobility table of a database. The LPM utility creates other monitoring threads on other VIOSes running on the (same) source server. If the first VIOS VIOS sustains one of multiple failures, the LPM utility provides notification to other functioning nodes/VIOSes. The LPM utility enables a functioning monitoring thread to update the LPM status. In particular, a last monitoring thread may perform cleanup/update operations within the database based on an indication that there are nodes on the first server that are in failed state.

Multiple node/virtual input/output (I/O) server (VIOS) failure recovery in clustered partition mobility

A mechanism is provided for offloading an input/output (I/O) completion operation. Responsive to a second processor identifying that a flag has been set by a first processor requesting assistance in completing an I/O operation, the second processor copies an I/O response from a first I/O response data structure associated with the first processor to a second I/O response data structure associated with the second processor. The second processor deletes the I/O response from the first I/O response data structure, clears the flag, and processes the I/O operation by addressing the I/O response in the second I/O response data structure. Responsive to completing the I/O operation, the second processor deletes the I/O response from the second I/O response data structure.

Offloading input/output (I/O) completion operations

Standalone software performance optimizer systems for hybrid systems include a hybrid system having a plurality of processors, memory operably connected to the processors, an operating system including a dispatcher loaded into the memory, a multithreaded application read into the memory, and a static performance analysis program loaded into the memory; wherein the static performance analysis program instructs at least one processor to perform static performance analysis on each of the threads, the static performance analysis program instructs at least one processor to assign each thread to a CPU class based on the static performance analysis, and the static performance analysis program instructs at least one processor to store each thread's CPU class. An embodiment of the invention may also include the dispatcher optimally mapping threads to processors using thread CPU classes and remapping threads to processors when a runtime performance analysis classifies a thread differently from the static performance analysis.

Standalone software performance optimizer system for hybrid systems

A hybrid CPU system wherein the plurality of processors forming the hybrid system are initially undifferentiated by type or class. Responsive to the sampling of the threads of a received and loaded computer application to be executed, the function of at least one of the processors is changed so that the threads of the sampled application may be most effectively processed/run on the hybrid system.

Optimizing the performance of hybrid CPU systems based upon the thread type of applications to be run on the CPUs

A first collection of threads which represent a collection of tasks to be executed by at least one of a collection of processing units is monitored. In response to detecting a request by a thread among the first collection of threads to access a shared resource locked by a second thread among the collection of threads, the first thread attempts to access a list associated with the shared resource. The list orders at least one thread among the collection of threads by priority of access to the shared resource. In response to determining the list is locked by a third thread among the collection of threads, the first thread is placed into a sleep state to be reawakened in a fixed period of time. In response to determining that at least one of the collection of processing units has entered into an idle state, the first thread is awakened from the sleep state before the fixed period of time has expired. Also, in response to awakening the first thread from the sleep state, the first thread is assigned to at least one of the collection of processing units and the first thread retries its attempt to access the list.

System and method of arbitrating access of threads to shared resources within a data processing system

In an embodiment, a kernel performs autonomic input/output tracing and performance tuning. A first table is provided in a device driver framework and a second table in a kernel of a computer. An input/output device monitoring tool is provided in the device driver framework. A plurality of instructions in the kernel compares each value in the first table with each value in the second table. Responsive to a match of a value in the first table and a value in the second table, the kernel automatically runs a command line to perform a system trace, a component trace, or a tuning task. The first table is populated with a plurality of values calculated from a plurality of data in a plurality of device memories and in the controller memory and the second table is populated in accordance with a second plurality of inputs to the command line interface.

Autonomic I/O tracing and performance tuning

A system and computer program product utilizes cluster-awareness to effectively support a live partition mobility (LPM) event and provide recovery from node failure within a Virtual Input/Output (I/O) Server (VIOS) cluster. An LPM utility creates a monitoring thread on a first VIOS on initiation of a corresponding LPM event. The monitoring thread tracks a status of an LPM and records status information in the mobility table of a database. The LPM utility creates other monitoring threads on other VIOSes running on the (same) source server. If the first VIOS sustains one of multiple failures, the LPM utility provides notification to other functioning nodes/VIOSes. The LPM utility enables a functioning monitoring thread to update the LPM status. In particular, a last monitoring thread may perform cleanup/update operations within the database based on an indication that there are nodes on the first server that are in failed state.

A symmetric multi-processor SMP system includes an SMP processor and operating system OS software that performs automatic SMP lock tracing analysis on an executing application program. System administrators, users or other entities initiate an automatic SMP lock tracing analysis. A particular thread of the executing application program requests and obtains a lock for a memory address pointer. A subsequent thread requests the same memory address pointer lock prior to the particular thread release of that lock. The subsequent thread begins to spin waiting for the release of that address pointer lock. When the subsequent thread reaches a predetermined maximum amount of wait time, MAXSPIN, a lock testing tool in the kernel of the OS detects the MAXSPIN condition. The OS performs a test to determine if the subsequent thread and address pointer lock meet the list of criteria set during initiation of the automatic lock trace method. The OS initiates an SMP lock trace capture automatically if all criteria or the arguments of the lock trace method are met. System administrators, software programmers, users or other entities interpret the results of the SMP lock tracing method that the OS stores in a trace table to determine performance improvements for the executing application program.

Symmetric multi-processor lock tracing

A method, computer program product, and a data processing system for maintaining objects in a lookup cache is provided. A primary list is populated with a first plurality of objects. The primary list is an unordered list of the first plurality of objects. A secondary list is populated with a second plurality of objects. The secondary list is an ordered list of the second plurality of objects. Periodically, at least one object of the first plurality of objects is demoted to the secondary list, and at least one object of the second plurality of objects is promoted to the primary list.

Optimized least recently used lookup cache

A method and system for providing access to a shared resource utilizing selective locking are disclosed. According to one embodiment, a method is provided comprising receiving a request to perform a resource access operation on a shared resource, invoking a first routine to perform the resource access operation, detecting a data processing system exception generated in response to invoking the first routine, and invoking a second routine to perform the resource access operation in response to such detecting. In the described embodiment, the first routine comprises a dereference instruction to dereference a pointer to memory associated with the shared resource, the second routine comprises a lock acquisition instruction to acquire a global lock associated with the shared resource prior to a performance of the resource access operation and a lock release instruction to release the global lock once resource access operation has been performed.

Synchronizing access to a shared resource utilizing selective locking

A method, computer program product, and system are provided for prefetching data into a cache memory. As a program is executed an object identifier is obtained of a first object of the program. A lookup operation is performed on a data structure to determine if the object identifier is present in the data structure. Responsive to the object identifier being present in the data structure, a referenced object identifier is retrieved that is referenced by the object identifier. Then, the data associated with the referenced object identifier is prefetched from main memory into the cache memory.

Programmable data prefetching

A computer implemented method, apparatus, and computer usable code for managing cache data. A partition identifier is associated with a cache entry in a cache, wherein the partition identifier identifies a last partition accessing the cache entry. The partition identifier associated with the cache entry is compared with a previous partition identifier located in a processor register in response to the cache entry being moved into a lower level cache relative to the cache. The cache entry is marked if the partition identifier associated with the cache entry matches the previous partition identifier located in the processor register to form a marked cache entry, wherein the marked cache entry is aged at a slower rate relative to an unmarked cache entry.

Growing community of inventors

Austin, TX, United States of America

Greg Robert Mewhinney