US 10,127,088 B2 · Adaptive techniques for improving performance of hardware transactions on multi-socket machines

Abstract

Socket scheduling modes may prevent non-uniform memory access effects from negatively affecting performance of synchronization mechanisms utilizing hardware transactional memory. Each mode may indicate whether a thread may execute a critical section on a particular socket. For example, under transitional lock elision, locks may include a mode indicating whether threads may acquire or elide the lock on a particular socket. Different modes may be used alternately to prevent threads from starving. A thread may only execute a critical section on a particular socket if allowed by the current mode. Otherwise, threads may block until allowed to execute the critical section, such as after the current mode changes. A profiling session may, for a running workload, iterate over all possible modes, measuring statistics pertaining to the execution of critical sections (e.g., the number of lock acquisitions and/or elisions), to determine the best performing modes for the particular workload.

Claims20

Dependent claims

A method, comprising:

performing by one or more computing nodes in a multi-socket system that supports multithreading and that implements a hardware transactional memory:
beginning execution, on the multi-socket system, of a multithreaded application that includes accesses to a shared resource, wherein the shared resource is accessible by a plurality of threads of the multithreaded application;
attempting, by a particular thread of the plurality of threads, to execute a critical section of the multithreaded application on a particular socket of the multi-socket system;
determining, based at least in part, on a current socket scheduling mode, whether the particular thread is allowed to currently execute the critical section on the particular socket, wherein the current socket scheduling mode is one of a plurality of socket scheduling modes, wherein the current socket scheduling mode indicates whether the particular thread may execute the critical section on the particular socket.

1 more element

A system, comprising:

a plurality of processor cores in multiple sockets;
a hardware transactional memory;
and
another memory coupled to the plurality of processor cores;
wherein the other memory stores program instructions that when executed on the one or more processor cores cause the plurality of processor cores to implement a multithreaded application that comprises a plurality of accesses to a shared resource, wherein the shared resource is accessible by a plurality of threads of the multithreaded application;
wherein, when executed on the one or more processor cores, the program instructions cause the plurality of processor cores to:
begin execution, on the multi-socket system, of the multithreaded application, wherein the shared resource is accessible by a plurality of threads of the multithreaded application;
attempt, by a particular thread of the plurality of threads, to execute a critical section of the multithreaded application on a particular socket of the multi-socket system;
determine, based at least in part, on a current socket scheduling mode, whether the particular thread is allowed to execute the critical section on the particular socket, wherein the current socket scheduling mode is one of a plurality of socket scheduling modes, wherein the current socket scheduling mode indicates where the particular thread may execute the critical section on the particular socket.

6 more elements

A non-transitory, computer-readable storage medium storing program instructions that when executed on one or more computers cause the one or more computers to perform:

beginning execution, on the multi-socket system, of a multithreaded application, in a multi-socket system that supports multithreading and that implements a hardware transactional memory, that includes accesses to a shared resource, wherein the shared resource is accessible by a plurality of threads of the multithreaded application;
attempting, by a particular thread of the plurality of threads, to execute a critical section of the multithreaded application on a particular socket of the multi-socket system;
determining, based at least in part, on a current socket scheduling mode, whether the particular thread is allowed to execute the critical section on the particular socket, wherein the current socket scheduling mode is one of a plurality of socket scheduling modes, wherein the current socket scheduling mode indicates where the particular thread may execute the critical section on the particular socket.

Dependencies are inferred from the claim text.

Adaptive techniques for improving performance of hardware transactions on multi-socket machines

Life of this patent

Adaptive techniques for improving performance of hardware transactions on multi-socket machines

Life of this patent

Abstract

Claims20