System.nanoTime(): Is It Truly Useless?
The concern raised about System.nanoTime() stems from the perception that on multi-core systems, time measurements can produce unexpected results due to variations in CPU-specific time counters. It is theorized that a thread scheduled to a different processor with a lower counter could yield a negative time difference when calculating time spent.
Addressing the Counter Discrepancy
However, contrary to the assumption, System.nanoTime() ensures consistency across processors. Modern operating systems and hardware abstractions (e.g., HAL) have addressed the issue raised in the original post.
Windows:
On Windows systems, newer versions of the Sun JDK (pre-2011) employ Windows' QueryPerformanceCounter rather than TSC. QueryPerformanceCounter guarantees synchronized time across processors, eliminating the counter discrepancy concern.
Linux:
Linux kernels 2.6.18 onwards address this issue by detecting unsynced timebases and falling back to a reliable clock source. This safeguard ensures consistent time keeping across multiple processors.
Other Platforms:
Information is limited regarding the behavior of System.nanoTime() on other platforms (e.g., SPARC, Solaris, IBM JVMs). However, it is reasonable to assume that modern operating systems and JVMs have implemented similar protections to maintain time consistency.
Conclusion:
While the original concerns raised may have been valid in certain historical contexts, System.nanoTime() is not obsolete. Modern operating systems and JVMs have implemented safeguards to ensure reliable time measurements across multiple processors. Therefore, System.nanoTime() remains a valuable tool for measuring elapsed time in Java applications.
The above is the detailed content of Is System.nanoTime() Still Relevant in Modern Multi-Core Systems?. For more information, please follow other related articles on the PHP Chinese website!