Using observed thread activity to dynamically tune a virtual machine for responsiveness
One embodiment of the present invention provides a system that uses observed thread activity to dynamically tune a virtual machine for responsiveness and/or throughput. During operation, the system observes the activity of a thread to determine whether the thread needs responsiveness and/or throughput. Next, the system gathers statistics for the thread based on these observations of thread activity, and then uses these statistics to dynamically tune a virtual machine upon which the thread will execute.
1. A method for using observed thread activity to dynamically tune a virtual machine for responsiveness or throughput, comprising:
in a computer system, performing the operations of:
observing the activity of a thread in a set of threads to determine whether the thread needs at least one of responsiveness or throughput, wherein observing the activity of a thread involves identifying a priority for the thread, wherein if the thread is a high-priority thread, the thread is categorized as needing responsiveness;
gathering statistics for the thread from the observations of thread activity to determine a level of responsiveness needed by the thread, wherein gathering the statistics comprises tracking at least one of frequency of context switches of high-priority threads and intervals between the execution of high-priority threads, and maintaining a set of measured values corresponding to the level of responsiveness, the measured values being decayed over time to ensure that the measured values reflect recent behavior; and
dynamically tuning a virtual machine upon which the thread will execute based on the gathered statistics and the level of responsiveness.
2. The method of claim 1 , wherein gathering the statistics for the thread involves:
translating the set of measured values into a set of configuration parameters for the virtual machine.
3. The method of claim 1 ,
wherein dynamically tuning a virtual machine involves dynamically changing or re-configuring virtual machine behavior or priorities; and
wherein dynamically changing or re-configuring virtual machine behavior can involve dynamically changing or re-configuring virtual machine runtime components including:
a garbage-collector;
a just-in-time compiler; and/or
a thread-synchronization system.
4. The method of claim 1 ,
wherein observing the activity of the thread involves instrumenting a system scheduler to track priorities when switching threads; and
wherein tracking priorities assists in determining the responsiveness needed by the thread.
5. The method of claim 1 , wherein gathering the statistics further comprises observing a population of objects being instantiated.
6. The method of claim 1 ,
wherein gathering statistics involves tracking a set of access patterns for the set of threads; and
wherein tracking the set of access patterns includes tracking a frequency of context switches of high-priority threads and an interval between the execution of high-priority threads.
7. The method of claim 1 ,
wherein observing the activity of the thread includes observing inter-thread synchronization mechanisms; and
wherein inter-thread synchronization mechanisms include notify-wait operations and locking mechanisms.
8. A non-transitory computer-readable storage medium storing instructions that when executed by a computer cause the computer to perform a method for using observed thread activity to dynamically tune a virtual machine for responsiveness or throughput, the method comprising:
observing the activity of a thread in a set of threads to determine whether the thread needs at least one of responsiveness or throughput, wherein observing the activity of a thread involves identifying a priority for the thread, wherein if the thread is a high-priority thread, the thread is categorized as needing responsiveness;
gathering statistics for the thread from the observations of thread activity to determine a level of responsiveness needed by the thread, wherein gathering the statistics comprises tracking at least one of frequency of context switches of high-priority threads and intervals between the execution of high-priority threads, and maintaining a set of measured values corresponding to the level of responsiveness, the measured values being decayed over time to ensure that the measured values reflect recent behavior; and
dynamically tuning a virtual machine upon which the thread will execute based on the gathered statistics and the level of responsiveness.
9. The non-transitory computer-readable storage medium of claim 8 , wherein gathering the statistics for the thread involves:
translating the set of measured values into a set of configuration parameters for the virtual machine.
10. The non-transitory computer-readable storage medium of claim 8 ,
wherein dynamically tuning a virtual machine involves dynamically changing or re-configuring virtual machine behavior or priorities; and
wherein dynamically changing or re-configuring virtual machine behavior can involve dynamically changing or reconfiguring virtual machine runtime components including:
a garbage-collector;
a just-in-time compiler; and/or
a thread-synchronization system.
11. The non-transitory computer-readable storage medium of claim 8 , wherein observing the activity of the thread involves instrumenting a system scheduler to track priorities when switching threads; and
wherein tracking priorities assists in determining the responsiveness needed by the thread.
12. The non-transitory computer-readable storage medium of claim 8 ,
wherein gathering statistics involves tracking a set of access patterns for the set of threads; and
wherein tracking the set of access patterns includes tracking a frequency of context switches of high-priority threads and an interval between the execution of high-priority threads.
13. The non-transitory computer-readable storage medium of claim 8 ,
wherein observing the activity of the thread includes observing inter-thread synchronization mechanisms; and
wherein inter-thread synchronization mechanisms include notify-wait operations and locking mechanisms.
14. A computing device that includes a processor that uses observed thread activity to dynamically tune a virtual machine for responsiveness and/or throughput, wherein the processor comprises:
an observation mechanism that observes the activity of a thread in a set of threads to determine whether the thread needs at least one of responsiveness or throughput, wherein the observation mechanism is further configured to identify a priority for the thread, and wherein if the thread is identified as a high-priority thread, the thread is categorized as needing responsiveness;
a gathering mechanism that gathers statistics for the thread from the observations of thread activity to determine a level of responsiveness needed by the thread, wherein gathering the statistics comprises tracking a set of access patterns comprises tracking at least one of frequency of context switches of high-priority threads and intervals between the execution of high-priority threads, and maintaining a set of measured values corresponding to the level of responsiveness, the measured values being decayed over time to ensure that the measured values reflect recent behavior; and
a tuning mechanism configured to dynamically tune a virtual machine upon which the thread will execute based on the gathered statistics and the level of responsiveness.
15. The apparatus of claim 14 ,
wherein the gathering mechanism is further configured to translate the set of measured values into a set of configuration parameters for the virtual machine.
16. The apparatus of claim 14 ,
wherein the tuning mechanism is further configured to dynamically change or re-configure virtual machine behavior or priorities; and
wherein dynamically changing or re-configuring virtual machine behavior can involve dynamically changing or re-configuring virtual machine runtime components including:
a garbage-collector;
a just-in-time compiler; and/or
a thread-synchronization system.
17. The apparatus of claim 14 ,
wherein the gathering mechanism is further configured to track a set of access patterns for the set of threads, and
wherein tracking the set of access patterns includes tracking a frequency of context switches of high-priority threads and an interval between the execution of high-priority threads.
18. The apparatus of claim 14 ,
wherein the observation mechanism is further configured to observe inter-thread synchronization mechanisms; and
wherein inter-thread synchronization mechanisms include notify-wait operations and locking mechanisms.