Associating working sets and threads
Associating working sets and threads is disclosed. An indication of a stalling event is received. In response to receiving the indication of the stalling event, a state of a processor associated with the stalling event is saved. At least one of an identifier of a guest thread running in the processor and a guest physical address referenced by the processor is obtained from the saved processor state.
1. A computer system, comprising:
a plurality of interconnected physical nodes, wherein each physical node includes one or more physical processors each having one or more cores, and wherein each core instantiates one or more virtual processors that execute program code;
wherein in response to an indication of a stalling event that stalled a virtual processor, a state of the virtual processor associated with the stalling event is saved;
wherein at least one of an identifier of a guest thread running in the virtual processor and a guest physical address referenced by the guest thread running in the virtual processor is obtained from the virtual processor state saved in response to the indication of the stalling event; and
wherein accessing of the guest physical address by the identified guest thread is recorded in a history of page accesses by the identified guest thread.
2. The computer system of claim 1 , wherein the identifier of the guest thread running in the virtual processor is obtained from an FS-Base0 register in the saved virtual processor state.
3. The computer system of claim 1 , wherein the guest physical address is obtained from a CR3 register in the saved virtual processor state.
4. The computer system of claim 1 , wherein the stalling event is associated with access of a non-local page.
5. The computer system of claim 1 , wherein the stalling event is associated with a thread context switch.
6. The computer system of claim 1 , wherein the recorded history of page accesses by the identified guest thread is stored in a bloom filter.
7. The computer system of claim 1 , wherein a measure of warmth for the identified guest thread is determined based at least in part on the recorded history, wherein the measure of warmth is indicative of a number of pages in the recorded history that are local to a node.
8. The computer system of claim 1 , wherein a measure of utilization for the identified guest thread is determined based at least in part on the recorded history, wherein the measure of utilization is indicative of an amount of overlap in page access behavior of the identified guest thread between two time periods.
9. The computer system of claim 1 , wherein a measure of efficiency for the identified guest thread when running on a physical node in the plurality of interconnected physical nodes is determined.
10. A method, comprising:
in response to an indication of a stalling event that stalled a virtual processor, saving a state of the virtual processor associated with the stalling event, wherein the virtual processor is instantiated by a core of a physical processor included in a physical node in a plurality of interconnected physical nodes, and wherein the virtual processor executes program code;
obtaining, from the virtual processor state saved in response to the indication of the stalling event, at least one of an identifier of a guest thread running in the virtual processor and a guest physical address referenced by the guest thread running in the virtual processor; and
recording accessing of the guest physical address by the identified guest thread in a history of page accesses by the identified guest thread.
11. The method of claim 10 , wherein the identifier of the guest thread running in the virtual processor is obtained from an FS-Base0 register in the saved virtual processor state.
12. The method of claim 10 , wherein the guest physical address is obtained from a CR3 register in the saved virtual processor state.
13. The method of claim 10 , wherein the stalling event is associated with access of a non-local page.
14. The method of claim 10 , wherein the stalling event is associated with a thread context switch.
15. The method of claim 10 , wherein the recorded history of page accesses by the identified guest thread is stored in a bloom filter.
16. The method of claim 10 , further comprising determining a measure of warmth for the identified guest thread based at least in part on the recorded history, wherein the measure of warmth is indicative of a number of pages in the recorded history that are local to a node.
17. The method of claim 10 , further comprising determining a measure of utilization for the identified guest thread based at least in part on the recorded history, wherein the measure of utilization is indicative of an amount of overlap in page access behavior of the identified guest thread between two time periods.
18. The method of claim 10 , further comprising determining a measure of efficiency for the identified guest thread when running on the physical node.