System and method for memory management

• US 9,292,359 B2
• Filed: 07/27/2012
• Issued: 03/22/2016
• Est. Priority Date: 07/27/2012
• Status: Active Grant

• Current Assignee: Intel Corporation
• Original Assignee: Intel Corporation
• Inventors: Hudson, Richard, Herhut, Stephan A.
• Primary Examiner(s): ZHE, MENG YAO
• Application Number:
  US13/560,054
• Publication Number:
  US 20140033213A1
• Time in Patent Office: 1,334 Days
• Field of Search: None
• US Class Current: 1/1
• CPC Class Codes: G06F 12/0253 Garbage collection, i.e. reclamation of unreferenced memory
• CPC Class Codes: G06F 2209/5011 Pool
• CPC Class Codes: G06F 9/5016 the resource being the memory
• CPC Class Codes: G06F 9/54 Interprogram communication

Abstract

A system and method for automatic memory management of a shared memory during parallel processing of a web application. The system includes a computing system configured to allow parallel computing of a web application executed within a web browser. The computing system includes shared memory having a set of blocks distributed at least a first thread and at least one spawned thread of a processing function of the web application. The memory is partitioned into a nursery heap, a mature heap and a database having a plurality of private nurseries, wherein the first thread has access to the nursery heap and mature heap and the at least one spawned thread has access to an associated one of the plurality of private nurseries. During parallel computing of the web application, management of the shared memory includes garbage collection of at least each of the plurality of private nurseries.

Claims

1. An apparatus for automatic memory management for parallel computing of a web application, said apparatus comprising:

a processor having at least two computer processing unit (CPU) cores;

an operating system configured to assign one or more threads of a processing function of a web application to corresponding CPU cores of said processor for concurrent execution on said corresponding CPU cores, said one or more threads comprise a first thread and at least one spawned thread;

memory having a set of blocks distributed to said first thread and said at least one spawned thread, said memory being partitioned into a nursery heap, a mature heap and a database having a plurality of private nurseries;

wherein said first thread has access to said nursery heap and mature heap, said first thread being configured to allocate and manipulate one or more objects in said nursery heap and write references to one or more objects stored in said nursery and mature heaps;

wherein said at least one spawned thread has access to an associated one of said plurality of private nurseries, said at least one spawned thread being configured to allocate and manipulate one or more objects within said associated private nursery; and

wherein, if said associated one of said plurality of private nurseries has insufficient memory space for said at least one spawned thread to complete allocation of one or more objects, said at least one spawned thread is configured to;

request access to an empty private nursery from said plurality of private nurseries, if available; and

when said empty private nursery is not available, request access to only an unused portion of memory space of a previously used private nursery from said plurality of private nurseries to allocate one or more objects.

2. The apparatus of claim 1, further comprising a web browser configured to execute said web application, wherein said web application is coded in a scripting language supported by said web browser.

3. The apparatus of claim 2, wherein said scripting language is JavaScript®

.

4. The apparatus of claim 2, further comprising a programming language engine configured to interpret and execute said web application code within a runtime system of said web browser.

5. The apparatus of claim 4, further comprising a deterministic parallel programming module configured to modify said programming language engine and to extend programming language with deterministic data-parallel constructs translated at runtime into a low-level hardware abstractions layer to enable data-parallelism for said web application.

6. The apparatus of claim 1, further comprising a memory management module configured to monitor execution of at least said first thread and said at least one spawned thread and manage life cycles of one or more objects allocated and stored within said sets of blocks of said memory.

7. The apparatus of claim 6, wherein said memory management module comprises a garbage collector module configured to perform garbage collection on at least said plurality of private nurseries during a garbage collection cycle.

8. The apparatus of claim 7, wherein said garbage collector module is configured to initiate said garbage collection cycle based, at least in part, on a state of execution of said at least one spawned thread, said state of execution determinative of whether said at least one spawned thread reaches a safe point.

9. The apparatus of claim 8, wherein said state of execution is selected from the group consisting of running, dormant and complete.

10. The apparatus of claim 9, wherein said at least one spawned thread reaches a safe point when said state of execution is dormant or complete and fails to reach a safe point when said state of execution is running.

11. A system for automatic memory management for parallel computing of a web application, said system comprising:

a media source; and

a client device configured to communicate with said media source over a network and access and execute a web application provided by said media source, said client device having a computing system configured to allow parallel computing of said web application, said computing system comprising;

a processor having at least two computer processing unit (CPU) cores;

an operating system configured to assign one or more threads of a processing function of said web application to corresponding CPU cores of said processor for concurrent execution on said corresponding CPU cores, said one or more threads comprise a first thread and at least one spawned thread;

memory having a set of blocks distributed to said first thread and said at least one spawned thread, said memory being partitioned into a nursery heap, a mature heap and a database having a plurality of private nurseries;

wherein said first thread has access to said nursery heap and mature heap, said first thread being configured to allocate and manipulate one or more objects in said nursery heap and write references to one or more objects stored in said nursery and mature heaps;

wherein said at least one spawned thread has access to an associated one of said plurality of private nurseries, said at least one spawned thread being configured to allocate and manipulate one or more objects within said associated private nursery; and

wherein, if said associated one of said plurality of private nurseries has insufficient memory space for said at least one spawned thread to complete allocation of one or more objects, said at least one spawned thread is configured to;

request access to an empty private nursery from said plurality of private nurseries, if available; and

when said empty private nursery is not available, request access to only an unused portion of memory space of a previously used private nursery from said plurality of private nurseries to allocate one or more objects.

12. The system of claim 11, further comprising a web browser configured to execute said web application, wherein said web application is coded in a scripting language supported by said web browser.

13. The system of claim 12, further comprising:

a programming language engine configured to interpret and execute said web application code within a runtime system of said web browser; and

a deterministic parallel programming module configured to modify said programming language engine and to extend programming language with deterministic data-parallel constructs translated at runtime into a low-level hardware abstractions layer to enable data-parallelism for said web application.

14. The system of claim 11, further comprising:

a memory management module configured to monitor execution of at least said first thread and said at least one spawned thread and manage life cycles of one or more objects allocated and stored within said sets of blocks of said memory; and

a garbage collector module configured to perform garbage collection on at least said plurality of private nurseries during a garbage collection cycle.

15. The system of claim 14, wherein said garbage collector module is configured to initiate said garbage collection cycle based, at least in part, on a state of execution of said at least one spawned thread, said state of execution determinative of whether said at least one spawned thread reaches a safe point, wherein said state of execution is selected from the group consisting of running, dormant and complete.

16. At least one non-transitory computer accessible medium storing instructions which, when executed by a machine, cause the machine to perform operations for automatic memory management for parallel computing of a web application, said operations comprising:

initiating concurrent execution of one or more threads of a processing function of a web application on corresponding CPU cores of a processor, said one or more threads comprise a first thread and at least one spawned thread;

acquiring associated blocks in a shared memory distributed to said first thread and said at least one spawned thread, said blocks comprising a nursery heap, a mature heap and a database having a plurality of private nurseries, wherein said first thread has access to and is configured to allocate and manipulate one or more objects within said nursery heap and mature heap and said at least one spawned thread has access to and is configured to allocate and manipulate one or more objects within an associated one of said plurality of private nurseries;

monitoring execution of at least said first thread and said at least one spawned thread and managing life cycles of one or more objects allocated and stored within said nursery heap, mature heap and each of said plurality of nurseries of said memory;

determining said associated one of said plurality of private nurseries has insufficient memory space for said at least one spawned thread to complete allocation of one or more objects;

accessing to an empty private nursery from said plurality of private nurseries, if available;

when said empty private nursery is not available, identifying a previously used private nursery from said plurality of private nurseries having a portion of memory space available for said at least one spawned thread to allocate one or more objects; and

accessing only an unused portion of available memory space of said previously used private nursery and allocating one or more objects within said unused portion of available memory space.

17. The non-transitory computer accessible medium of claim 16, further comprising:

performing garbage collection on at least said plurality of private nurseries during initiation of a garbage collection cycle.

18. The non-transitory computer accessible medium of claim 17, wherein said garbage collection cycle is based, at least in part, on a state of execution of said at least one spawned thread, said state of execution determinative of whether said at least one spawned thread reaches a safe point, wherein said state of execution is selected from the group consisting of running, dormant and complete.

19. A method for automatic memory management for parallel computing of a web application, said method comprising:

initiating concurrent execution of one or more threads of a processing function of a web application on corresponding CPU cores of a processor, said one or more threads comprising a first thread and at least one spawned thread;

acquiring associated blocks in a shared memory distributed to said first thread and said at least one spawned thread, said blocks comprising a nursery heap, a mature heap and a database having a plurality of private nurseries, wherein said first thread has access to and is configured to allocate and manipulate one or more objects within said nursery heap and mature heap and said at least one spawned thread has access to and is configured to allocate and manipulate one or more objects within an associated one of said plurality of private nurseries;

monitoring execution of at least said first thread and said at least one spawned thread and managing life cycles of one or more objects allocated and stored within said nursery heap, mature heap and each of said plurality of nurseries of said memory;

determining said associated one of said plurality of private nurseries has insufficient memory space for said at least one spawned thread to complete allocation of one or more objects;

accessing to an empty private nursery from said plurality of private nurseries, if available;

when said empty private nursery is not available, identifying a previously used private nursery from said plurality of private nurseries having a portion of memory space available for said at least one spawned thread to allocate one or more objects; and

accessing only an unused portion of available memory space of said previously used private nursery and allocating one or more objects within said unused portion of available memory space.

20. The method of claim 19, further comprising:

performing garbage collection on at least said plurality of private nurseries during initiation of a garbage collection cycle.

21. The method of claim 20, wherein said garbage collection cycle is based, at least in part, on a state of execution of said at least one spawned thread, said state of execution determinative of whether said at least one spawned thread reaches a safe point, wherein said state of execution is selected from the group consisting of running, dormant and complete.