Adaptive memory allocation

Information

  • Patent Grant
  • 7035989
  • Patent Number
    7,035,989
  • Date Filed
    Wednesday, February 16, 2000
    25 years ago
  • Date Issued
    Tuesday, April 25, 2006
    18 years ago
Abstract
This functions maintains two trees: a fast access tree referring to memory blocks of a size most often requested, and a general access tree referring to memory blocks of a size less often requested. After satisfying a request for a memory block, the function adjusts the trees to ensure that the fast access tree refers to memory blocks of the size most often requested. By providing such functionality, the function improves its performance over time through self-adaptation.
Description
FIELD OF THE INVENTION

This invention relates generally to data processing systems, and more particularly, to methods for optimizing the allocation and deallocation of shared memory to programs executing in a data processing system.


BACKGROUND OF THE INVENTION

During execution, programs typically make many dynamic memory access requests. Such requests involve requesting allocation of memory from a system call (e.g., malloc), utilizing the memory, and deallocating the memory using a system call (e.g., free). Dynamic memory allocation is a fundamental and very important part of many computer programs. It is therefore desirable to improve the performance of memory access functions.


SUMMARY OF THE INVENTION

In accordance with methods and systems consistent with the present invention, an improved memory access function (e.g., malloc) is provided that dynamically improves its performance by changing its operation at runtime. The memory access function adapts its operation based on previous memory access requests to be able to provide the fastest response to those kinds of memory access requests that it predicts will be the most common. For purposes of this description “access to system memory” includes both requests for allocation of system memory and release of system memory, or deallocation. Similarly, a “memory request” refers to either an allocation (a request for system memory), or a deallocation (a return of previously requested memory).


In accordance with methods and systems consistent with the present invention, as embodied and broadly described herein, a method is provided in a data processing system for allocating memory. The method receives a memory request for a reference to a block of memory. Then it returns the reference to the block of memory to satisfy the request. Next, it adjusts an operation of the memory access function based on the memory request.


Furthermore, in accordance with methods and system consistent with the present invention, as embodied and broadly described herein, a system for providing access to memory includes a memory which further includes a program including a memory access function that provides access to memory and adjusts its operation according to a memory request for a reference to a block of memory, and a processor for executing the program.





BRIEF DESCRIPTION OF THE DRAWINGS


FIG. 1 depicts a block diagram of a data processing system suitable for practicing methods and systems consistent with the present invention;



FIG. 2 depicts in greater details the general access tree and the fast access tree depicted in FIG. 1;



FIGS. 3A & 3B depict a flow diagram illustrating operations performed to allocate memory in accordance with the principles of the present invention; and



FIG. 4 depicts a flow diagram illustrating operations performed to deallocate memory in accordance with the principles of the present invention.





DETAILED DESCRIPTION

Methods and systems consistent with the present invention provide a function called “smart-alloc” that adapts in real-time the memory allocation process to threads executing in a data processing system. More specifically, smart-alloc observes past memory allocation behavior and adjusts the process of providing memory access such that the most frequently requested memory blocks of a specific size are allocated faster than less frequently requested memory block sizes. Smart-alloc is a language-independent library function which may be linked to one or more programs operating in a multi-processing environment.


Overview


When a program begins execution, smart-alloc secures access to an amount of memory that may be accessed as needed by threads executing within the program. Upon receiving a pointer from a system memory call to an amount of memory requested, smart-alloc divides the memory into blocks of varying sizes, grouping blocks of like size together in a linked-list, referred to as a free memory linked-list. Smart-alloc creates two trees that point to the free memory linked-lists: a fast access tree and a general access tree. As a program executes, the fast access tree points to the most frequently accessed memory block sizes, and the general access tree points to block sizes not pointed to by the fast access tree. Each free memory linked-list will be pointed to by either the general access tree, or the fast access tree, but not both.


Upon receiving a request from an executing thread for access to memory, smart-alloc increments a counter associated with the block size that will be used to satisfy the memory request. To satisfy the memory request, smart-alloc first determines whether the fast access tree points to memory blocks of the size needed to satisfy the request. If the fast access tree includes memory blocks of the size needed to satisfy the request, smart-alloc satisfies the request from the fast access tree. Otherwise, smart-alloc satisfies the request from the general access tree. If a request is satisfied from the general access tree, smart-alloc compares the value of the counter associated with the allocated block size to the counter values for block sizes included in the fast access tree to determine whether the general access tree refers to block sizes most frequently requested. If the counter value for the allocated block size (from the general access tree) is greater than any of the counter values of block sizes included in the fast access tree, smart-alloc replaces that general access tree pointer to the block size having the highest counter value with a pointer from the fast access tree, and replaces the pointer to the block size included in the fast access tree having the lowest counter value with a pointer from the general access tree. This process ensures that the fast access tree continues to point to the most frequently requested block sizes.


Smart-alloc also receives requests from executing threads desiring to release references to blocks of memory. After receiving a released reference, smart-alloc adds the reference to the free memory linked-list that includes memory blocks corresponding to the size of the returned reference.


Implementation Details



FIG. 1 depicts a block diagram of a data processing system 100 suitable for practicing methods and implementing systems consistent with the present invention. Data processing system 100 includes computer 102 connected to network 105. Computer 102 includes secondary storage 110, processors 120a . . . n, output device 130, input device 140, and memory 150. Memory 150 includes programs 155, 160, and 165, operating system 170, and shared memory 180. Operating system 170 includes a system shared memory access function, malloc 182. Each of programs 155, 160, and 165 includes smart-alloc 185, a shared memory access function operating in accordance with the principles of the present invention. Each instance of smart-alloc 185 includes two data structures, a fast access tree 186 and a general access tree 187 that are used by smart-alloc 185 when allocating and deallocating memory.


Programs 155, 160, and 165 share access to shared memory 180. Programs 155, 160, and 165 may include a single thread or multiple threads. Although multiple processors are shown, one skilled in the art will appreciate that programs 155, 160, and 165 may execute on a single processor to implement methods and practice systems consistent with the present invention.


Operating system 170 represents the Solaris® operating system from Sun Microsystems, Inc., which specifically facilitates multi-threaded program execution, although any operating system may be used to implement methods and systems consistent with the present invention.


Although aspects of this implementation are depicted as being stored in memory 150, one skilled in the art will appreciate that all or part of systems and methods consistent with the present invention may be stored on or read from other computer readable media, such as secondary storage devices, like hard disks, floppy disks, and CD-ROM; a digital signal received from a network such as the Internet; or other forms of RAM or ROM, either currently known or later developed. Sun, Sun Microsystems, and the Sun logo are trademark or registered trademarks of Sun Microsystems, Inc. in the United States and other countries.



FIG. 2 depicts general access tree 202 and fast access tree 204 pointing to free memory linked-lists 262272. In this example, general access tree 202 includes two main branches: branch 206 including pointers to free memory linked-lists containing references to memory blocks of a size less than or equal to 4K and branch 208 including pointers to free memory linked-lists containing references to memory blocks of a size greater than 4K. Branches 210, 215, and 225 serve as head pointers to free memory linked-lists 262, 266, and 272, which include memory blocks of size 1K, 4K, and 64K, respectively. Fast access tree 204 also includes two main branches, 230 and 240. Fast access tree 204 further includes head pointers 250, 255, and 260 to free memory linked-lists 264, 268, and 270, including memory blocks of size 2K, 8K, and 16K, respectively, indicating that these memory block sizes were most frequently accessed by the threads executing in the associated program. General access tree 202 may include additional branches that include head pointers to additional free memory linked-lists. Fast access tree 204, however, only includes head pointers to a specific number of free memory linked-lists, typically no more than eight. A “head pointer” is a pointer to the beginning of a free memory linked-list. Thus, because the fast access tree does not typically grow as large as the general access tree, accesses to the fast access tree are faster than accesses to the general access tree.



FIGS. 3A & 3B depict a flow diagram of a smart-alloc function operating in accordance with the principles of the present invention. First, smart-alloc requests access to an amount of memory from a system memory call included as part of an operating system (step 305). If the amount of memory is available (step 310), the system memory call returns to smart-alloc a pointer to the amount of memory requested (step 315). Upon receiving a pointer to a requested amount of memory, smart-alloc divides the memory into blocks of varying sizes and builds a free memory linked-list for each block size (step 318). The block sizes in the free memory linked-lists begin at 1K and increase in an amount equal to double the size of a previous block. For example, 1K, 2K, 4K, 8K, 16K, etc. Once the free memory linked-lists are created, smart-alloc creates the general access tree which includes head pointers to each of the free memory linked-lists (step 320). Initially, the fast access tree is empty. As a program executes and a series of memory requests are processed, the fast access tree is populated as discussed below.


Upon receiving a memory request from an executing thread (step 325), smart-alloc increments the value of a counter associated with the block size used to satisfy the request (step 330). Thus, if a thread requests 1026 bytes of memory and the free memory linked-lists include 1K and 2K blocks, a reference to the 2K block will be returned to the requesting thread, and the value of the counter associated with the 2K free memory linked-list will be incremented. If this memory request is satisfied from the fast access tree (step 335), processing ends.


Otherwise, if the memory request is satisfied from the general access tree (step 340), then the counter associated with the block size used to satisfy the request is compared with the counter for block sizes pointed to by the fast access tree (step 370). Smart-alloc then determines whether the block size should be added to the free memory linked-list pointed to by the fast access tree step 375 because it includes memory blocks of a size most often requested by the executing program. In this step, smart-alloc compares the value of the counter associated with a block size included in the general access tree with the value of the counter of a block size included in the fast access tree. If the counter value for the block in the general access tree is greater than the counter value for the block in the fast access tree, the pointer from the general access will be replaced with a pointer from the fast access tree. Additionally, the fast access tree pointer that points to the block size having the lowest counter value is replaced with the pointer from the general access tree (step 380). Changing the pointers in this manner ensures that the fast access tree continually points to the most frequently allocated block sizes, thereby increasing the efficiency of memory allocation. Otherwise, if the counter value associated with the block in the general access tree is not greater than the counter value of any of the free memory linked-lists pointed to by the fast access tree, processing ends.


If a memory request cannot be satisfied from either the general or fast access trees, smart-alloc requests additional memory from a system memory call (step 345). If the additional memory is available (step 350), a pointer to the amount of memory is returned to smart-alloc (step 355), the memory is divided into blocks that are added to the free memory linked-lists pointed to by a pointer from the general access tree (step 360), and processing continues to step 335. Otherwise, if the requested amount of memory is not available, the system memory call returns a null pointer to smart-alloc (step 365), and processing ends.



FIG. 4 depicts a flow diagram of operations performed by smart-alloc when a thread releases a reference to a block of memory. Upon receiving a reference to a block of memory from an executing thread (step 410), smart-alloc adds the reference to the appropriate free memory linked-list (step 430) and processing ends.


CONCLUSION

Methods and systems operating in accordance with the principles of the present invention optimize the process of allocating memory to threads executing in programs operating in a data processing system by adapting the allocation process according to the behavior of an executing program. This smart-alloc function is a library function that may be linked to any program executing in a data processing system. As discussed above, smart-alloc operates in the user space and does not require accessing the operating system each time memory is allocated or deallocated. These distributive and adaptive features of smart-alloc allow it to minimize the number of accesses to the operating system, a costly and time consuming event, and ultimately yields increased system performance.


Methods and systems consistent with the present invention are applicable to all single and multi-threaded programs written in all computer programming languages, including Fortran 77, Fortran 90, Java, C, and C++. Although maximum increases in efficiency will be realized in a multi-processor computing environment, methods and systems consistent with the present invention may also provide operational benefits in a single or multi-threaded, single processor environment.


Although the foregoing description has been described with reference to a specific implementation, those skilled in the art will know of various changes in form and detail which may be made without departing from the spirit and scope of the present invention as defined in the appended claims and the full scope of their equivalents.

Claims
  • 1. A method in a data processing system for allocating memory by a memory allocation function, comprising: receiving a memory request for a reference to a block of memory;returning the reference to the block of memory to satisfy the request;forming a plurality of linked-lists referring to memory blocks of a plurality of sizes, each of the plurality of linked-lists referring to memory blocks of a common size;setting a fast access tree to refer to a first one of the plurality of linked-lists; andsetting a general access tree to refer to a second one and a third one of the plurality of linked-lists, wherein a size of a memory block referred to by the first linked-list is larger than a size of a memory block referred to by the second linked-list and smaller than a size of a memory block referred to by the third linked-list.
  • 2. The method of claim 1, further comprising adjusting a structure of the fast access tree and the general access tree based on the memory request.
  • 3. The method of claim 2, further comprising searching the fast access tree to satisfy the memory request and, if the memory request is not satisfied, searching the general access tree to satisfy the memory request.
  • 4. The method of claim 1, further including ensuring that the fast access tree refers to one of the plurality of linked-lists that is most frequently requested.
  • 5. The method of claim 1, wherein receiving, returning, forming, and setting are performed in a user space of the memory.
  • 6. A system for allocating memory, comprising: means for receiving a memory request for a reference to a block of memory;means for returning the reference to the block of memory to satisfy the request; andmeans for forming a plurality of linked-lists referring to memory blocks of a plurality of sizes, each of the plurality of linked-lists referring to memory blocks of a common size;means for setting a fast access tree to refer to a first one of the plurality of linked-lists;means for setting a general access tree to refer to a second one and a third one of the plurality of linked-lists, wherein a size of a memory block referred to by the first linked-list is larger than a size of a memory block referred to by the second linked-list and smaller than a size of a memory block referred to by the third linked-list.
  • 7. A data processing system for providing access to memory, comprising: a memory including:an access tree structure comprising a fast access tree and a general access tree;a program including a memory access function that provides access to the memory, forms a plurality of linked-lists referring to memory blocks of a plurality of sizes, each of the plurality of linked-lists referring to memory blocks of a common size, sets the fast access tree to refer to a first one of the plurality of linked-lists, and sets the general access tree to refer to a second one and a third one of the plurality of linked-lists, wherein a size of a memory block referred to by the first linked-list is larger than a size of a memory block referred to by the second linked-list and smaller than a size of a memory block referred to by the third linked-list; anda processor for executing the program.
  • 8. The data processing system of claim 7, further including an operating system with a system memory function, and wherein the memory access function provides access to the memory by utilizing the system memory function.
  • 9. The data processing system of claim 7, wherein the memory access function adjusts a structure of the fast access tree and the general access tree based on a memory request.
  • 10. The data processing system of claim 7, wherein the fast access tree refers to one of the plurality of linked-lists that is most frequently accessed.
  • 11. The data processing system of claim 7, wherein a most frequently accessed memory block size is included in the fast access tree.
  • 12. The data processing system of claim 7, wherein a least frequently accessed memory block size is included in the general access tree.
  • 13. The data processing system of claim 7, wherein the memory access function searches the fast access tree to satisfy a memory request and, if the memory request is not satisfied, searches the general access tree to satisfy the memory request.
  • 14. The data processing system of claim 13, wherein each of the plurality of linked-lists has an associated counter indicating a number of times that the associated linked-list has been accessed.
  • 15. A computer-readable medium including instructions for performing a method for allocating memory by a memory allocation function, the method comprising: receiving a memory request for a reference to a block of memory;returning the reference to the block of memory to satisfy the request;forming a plurality of linked-lists referring to memory blocks of a plurality of sizes, each of the plurality of linked-lists referring to memory blocks of a common size;setting a fast access tree to refer to a first one of the plurality of linked-lists;setting a general access tree to refer to a second one and a third one of the plurality of linked-lists, wherein a size of a memory block referred to by the first linked-list is larger than a size of a memory block referred to by the second linked-list and smaller than a size of a memory block referred to by the third linked-list.
  • 16. The computer-readable medium, including instructions for performing the method, of claim 15, the method further comprising adjusting a structure of the fast access tree and the general access tree based on the memory request.
  • 17. The computer-readable medium, including instructions for performing the method, of claim 15, the method further comprising searching the fast access tree to satisfy the memory request and, if the memory request is not satisfied, searching the general access tree to satisfy the memory request.
  • 18. The computer-readable medium, including instructions for performing the method, of claim 15, the method further comprising ensuring that the fast access tree refers to one of the plurality of linked-lists that is most frequently requested.
US Referenced Citations (145)
Number Name Date Kind
4675832 Robinson et al. Jun 1987 A
4685082 Cheung et al. Aug 1987 A
4812996 Stubbs Mar 1989 A
5073851 Masterson et al. Dec 1991 A
5075847 Fromme Dec 1991 A
5079707 Bird et al. Jan 1992 A
5119465 Jack et al. Jun 1992 A
5146593 Brandle et al. Sep 1992 A
5168563 Shenoy et al. Dec 1992 A
5179702 Spix et al. Jan 1993 A
5274813 Itoh Dec 1993 A
5274821 Rouquie Dec 1993 A
5297274 Jackson Mar 1994 A
5301312 Christopher, Jr. et al. Apr 1994 A
5325499 Kummer et al. Jun 1994 A
5325533 McInerney et al. Jun 1994 A
5353401 Iizawa et al. Oct 1994 A
5390314 Swanson Feb 1995 A
5438659 Notess et al. Aug 1995 A
5450542 Lehman et al. Sep 1995 A
5485619 Lai et al. Jan 1996 A
5497458 Finch et al. Mar 1996 A
5499349 Nikhil et al. Mar 1996 A
5500881 Levin et al. Mar 1996 A
5519866 Lawrence et al. May 1996 A
5530816 Holt Jun 1996 A
5535364 Resman et al. Jul 1996 A
5535393 Reeve et al. Jul 1996 A
5539907 Srivastava et al. Jul 1996 A
5553235 Chen et al. Sep 1996 A
5574922 James Nov 1996 A
5613063 Eustace et al. Mar 1997 A
5636374 Rodgers et al. Jun 1997 A
5640550 Coker Jun 1997 A
5673387 Chen et al. Sep 1997 A
5675790 Walls Oct 1997 A
5675802 Allen et al. Oct 1997 A
5689712 Heisch Nov 1997 A
5696937 White et al. Dec 1997 A
5710727 Mitchell et al. Jan 1998 A
5724262 Ghahramani Mar 1998 A
5734822 Houha et al. Mar 1998 A
5737605 Cunningham et al. Apr 1998 A
5740431 Rail Apr 1998 A
5740433 Carr et al. Apr 1998 A
5742793 Sturges et al. Apr 1998 A
5745897 Perkins et al. Apr 1998 A
5748892 Richardson May 1998 A
5748961 Hanna et al. May 1998 A
5754820 Yamagami May 1998 A
5761426 Ishizaki et al. Jun 1998 A
5774724 Heisch Jun 1998 A
5784698 Brady et al. Jul 1998 A
5787480 Scales et al. Jul 1998 A
5805795 Whitten Sep 1998 A
5812799 Zuravleff et al. Sep 1998 A
5835705 Larsen et al. Nov 1998 A
5850554 Carver Dec 1998 A
5860024 Kyle et al. Jan 1999 A
5864867 Krusche et al. Jan 1999 A
5867649 Larson Feb 1999 A
5867735 Zuravleff et al. Feb 1999 A
5872977 Thompson Feb 1999 A
5890171 Blumer et al. Mar 1999 A
5905488 Demers et al. May 1999 A
5905856 Ottensooser May 1999 A
5913223 Sheppard et al. Jun 1999 A
5920895 Perazzoli, Jr. et al. Jul 1999 A
5963975 Boyle et al. Oct 1999 A
5968114 Wentka et al. Oct 1999 A
5970510 Sher et al. Oct 1999 A
5974510 Cheng et al. Oct 1999 A
5974536 Richardson Oct 1999 A
5978892 Noel et al. Nov 1999 A
5987479 Oliver Nov 1999 A
5991708 Levine et al. Nov 1999 A
5991893 Snider Nov 1999 A
6006031 Andrews et al. Dec 1999 A
6009514 Henzinger et al. Dec 1999 A
6014517 Shagam et al. Jan 2000 A
6016474 Kim et al. Jan 2000 A
6018793 Rao Jan 2000 A
6023583 Honda Feb 2000 A
6044438 Olnowich Mar 2000 A
6049798 Bishop et al. Apr 2000 A
6049855 Jeddeloh Apr 2000 A
6052708 Flynn et al. Apr 2000 A
6052763 Maruyama Apr 2000 A
6055368 Kunioka Apr 2000 A
6065019 Ault et al. May 2000 A
6066181 DeMaster May 2000 A
6072951 Donovan et al. Jun 2000 A
6077312 Bates et al. Jun 2000 A
6081868 Brooks Jun 2000 A
6085029 Kolawa et al. Jul 2000 A
6088771 Steely, Jr. et al. Jul 2000 A
6098169 Ranganathan Aug 2000 A
6101325 Flaat Aug 2000 A
6101525 Hecker Aug 2000 A
6108343 Cruickshank et al. Aug 2000 A
6119198 Fromm Sep 2000 A
6125430 Noel et al. Sep 2000 A
6141692 Loewenstein et al. Oct 2000 A
6145054 Mehrotra et al. Nov 2000 A
6167565 Kanamori Dec 2000 A
6173327 De Borst et al. Jan 2001 B1
6173368 Kreuger et al. Jan 2001 B1
6205537 Albonesi Mar 2001 B1
6223134 Rust et al. Apr 2001 B1
6253252 Schofield Jun 2001 B1
6263485 Schofield Jul 2001 B1
6269457 Lane Jul 2001 B1
6282702 Ungar Aug 2001 B1
6286130 Poulsen et al. Sep 2001 B1
6295600 Parady Sep 2001 B1
6304951 Mealey et al. Oct 2001 B1
6311320 Jibbe Oct 2001 B1
6314429 Simser Nov 2001 B1
6317871 Andrews et al. Nov 2001 B1
6341338 Dennie Jan 2002 B1
6351845 Hinker et al. Feb 2002 B1
6353829 Koblenz et al. Mar 2002 B1
6353869 Ofer et al. Mar 2002 B1
6366994 Kalyur Apr 2002 B1
6369725 Busaba Apr 2002 B1
6430657 Mittal et al. Aug 2002 B1
6434714 Lewis et al. Aug 2002 B1
6438745 Kanamaru et al. Aug 2002 B1
6442162 O'Neill et al. Aug 2002 B1
6473833 Arimilli et al. Oct 2002 B1
6480818 Alverson et al. Nov 2002 B1
6496902 Faanes et al. Dec 2002 B1
6502136 Higuchi et al. Dec 2002 B1
6523090 Tremblay Feb 2003 B1
6542919 Wendorf et al. Apr 2003 B1
6574725 Kranich et al. Jun 2003 B1
6629214 Arimilli et al. Sep 2003 B1
6647546 Hinker et al. Nov 2003 B1
6684296 Hayter et al. Jan 2004 B1
20010003831 Boland Jun 2001 A1
20010051974 Saad Dec 2001 A1
20020046201 Hembry Apr 2002 A1
20020073360 Lewis et al. Jun 2002 A1
20020078010 Ehrman et al. Jun 2002 A1
20030061395 Kingsbury et al. Mar 2003 A1
Foreign Referenced Citations (16)
Number Date Country
199 34 515 Jan 2000 DE
0 390 339 Oct 1990 EP
0 703 534 Mar 1996 EP
0 817 044 Jan 1998 EP
0 965 921 Dec 1999 EP
1 024 432 Aug 2000 EP
1 026 592 Aug 2000 EP
1 081 585 Mar 2001 EP
2 793 908 Nov 2000 FR
2 324 942 Nov 1998 GB
2 343 029 Apr 2000 GB
2 357 873 Jul 2001 GB
03-282731 Dec 1991 JP
07-056716 Mar 1995 JP
07-156716 Mar 1995 JP
WO 9910812 Mar 1999 WO