Method and system for determining resource allocation to users by granting request based on user associated different limits and resource limit

[0001] 1. Field of Invention

[0002] The present invention pertains to the field of computer systems. More particularly, this invention relates to flexible allocation of a resource in a computer system.

[0003] 2. Art Background

[0004] A computer system typically includes resources that are shared among multiple users. An example of a shared resource is a shared physical memory. Examples of a shared physical memory include main memory, persistent memory including mass storage devices, and information stores, etc. Another example of a shared resource is a communication link. Yet another example of a shared resource is a processor.

[0005] A shared resource usually has a limited capacity or limited capability with respect to the needs of the potential users of the shared resource. For example, a physical memory usually has a limited storage capacity. A communication link typically has a limited bandwidth. A processor usually has a limited instruction execution throughput. As a consequence, computer systems commonly implement methods for allocating the capacity or capability of a shared resource among the users of the shared resource.

[0006] One prior method for allocating a shared resource is to employ static partitioning. For example, static partitioning may be applied to a physical memory having a storage capacity of C bytes by allocating C/n bytes to each of n potential users of the physical memory. Unfortunately, such static partitioning usually limits each user to C/n bytes even when a only a small percentage of the potential users actually use physical memory at any given time. Such partitioning commonly results in severe underutilization of the shared resource.

[0007] Another prior method for allocating a shared resource is to allocate a portion of the shared resource to each requesting user on a first-come-first-served basis. For a physical memory having a storage capacity of C bytes, for example, C/10 bytes may be allocated to each requesting user. Unfortunately, such a method usually exhausts the capacity of the shared resource after the first 10 users, thereby locking subsequent users out of the shared resource.

[0008] A method is disclosed for flexible allocation of a resource. The method involves assigning a soft limit and a hard limit to each of a set of potential users of the resource. The soft limits are selected to guarantee access to the resource by all of the potential users. The hard limits are selected to enable each potential user to exceed the corresponding soft limit on a first-come-first-served basis. A request from a user for allocation of a portion of the resource is handled by granting the request if the request if allowed would not exceed soft limit assigned to the user. The request is denied if the request if allowed would exceed the hard limit assigned to the user. To avoid overtaxing the capacity of the resource, the request is denied even when the hard limit of the user is not exceeded if the request if allowed would cause a total allocation of the resource to exceed a high watermark assigned to the resource.

[0009] Other features and advantages of the present invention will be apparent from the detailed description that follows.

[0014] FIG. 1 shows a computer system 100 that incorporates the present teachings. The computer system 100 includes a resource 10 that is shared among a set of tasks 20–30. Portions of the resource 10 are allocated to the tasks 20–30 by a resource manager 12. The resource manager 12 maintains a set of resource allocation parameters 14 which are used in resource allocation.

[0015] The resource 10 represents any resource having a limited capacity or capability that may be allocated among the tasks 20–30. The resource 10 may be a hardware resource, a software resource, or a combination hardware/software resource. Examples for the resource 10 include physical memory such as main memory, mass storage, persistent stores, information stores including databases, non-volatile memory, processor time, communication links, and input/output devices to name a few examples.

[0016] The tasks 20–30 represent software tasks that may be executed on the computer system 100. Examples for the tasks 20–30 include application programs and related software components and user interface tasks. Each task 20–30 may be associated with a particular user of the computer system 10. More than one of the tasks 20–30 may be associated with the same user. In one embodiment, the resource manager 12 allocates the resource 10 on a per user basis so that all of the tasks associated with a given user are confined to a portion of the resource 10 that is allocated to the given user by the resource manager 12.

[0017] The computer system 100 may be a single processor system, a multiple processor system, multiple networked computer systems, multiple networked devices which include computing capabilities, or any combination of these. The resource manager 12 may be part of an operating system of the computer system 100, may be a component such as a device driver, and/or may function as a server for the resource 10 that handles requests from the tasks 20–30 which function as clients.

[0018] The capacity or capability of the resource 10 may be expressed in terms of units. For example, if the resource 10 is a memory then a unit may be a byte, a block, a line, a kilobyte, a megabyte, etc. In another example, if the resource 10 is a communication link then a unit may be a bit per second, a kilobit per second, or a megabit per second of communication bandwidth, etc. In yet another example, if the resource 10 is a processor then a unit may be a million instructions per second (MIPS) of processor execution time.

[0019] The resource manager 12 receives requests from the tasks 20–30 for allocation of the resource 10. The resource manager 120 allocates portions of the resource 10 to the requesting tasks 20–30 using information provided by the resource allocation parameters 14.

[0020] The resource allocation parameters 14 include a total capacity or capability (T) of the resource 10 expressed in units. The value of T depends on the characteristics of the resource 10 and the selected units. For example, if the resource 10 is a 1000 megabyte memory then T equals 1000 if the units are megabytes.

[0021] The value of T may also take into account a portion of the resource 10 which is allocated to system functions and not available to the tasks 20–30. For example, if the resource 10 is a 1000 megabyte memory, then 50 megabytes may be reserved for system use and unavailable for allocation to the tasks 20–30. This yields a value of T of 950 units in megabytes.

[0022] The resource allocation parameters 14 include a soft limit (S) which applies to each potential user of the resource 10. The soft limit S is a minimum portion of the resource 10 to which each potential user has guaranteed access, thereby preventing potential users from being locked out of the resource 10 at any time.

[0023] The soft limit S is a tunable parameter of the computer system 100. It is preferable that S be set to a high enough value as to enable advantageous use of the resource 10 but not so high as to needlessly tie up the capacity of the resource 10 when only a few of the potential users access the resource 10.

[0024] The soft limit S may be the same for all potential users or may be set on a per user basis or on the basis of classes of users. For example, some classes of users such as those who pay more or those in management positions, etc., may have a higher soft limit than that of ordinary users.

[0025] The resource allocation parameters 14 include a hard limit (H) which enables users to exceed their soft limits under predetermined conditions. A given user is always granted his soft limit and may be granted up to his hard limit if the current utilization of the resource 10 can accommodate the request. The maximum value for the hard limit H is equal to T minus the sum of the soft limits of all potential users. The hard limit H is a tunable parameter of the computer system 100. The hard limit H may be the same for all potential users or may be set on a per user basis or on the basis of classes of users.

[0026] The resource allocation parameters 14 include a high watermark and a low watermark. The high watermark is an upper limit on the total utilization of the resource 10. The difference between the high and low watermarks provides hysteresis that prevents thrashing that would otherwise occur when one of the tasks 20–30 frees a portion of the resource 10 and then reallocates that portion when the resource 10 is near its capacity.

[0027] FIG. 2 illustrates the handling of a request 200 for allocation of the resource 10 by the resource manager 12 in a normal mode in one embodiment. The normal mode of handling a request for allocation is the initial mode before the high watermark of the resource 10 has been exceeded.

[0028] In this example, the request 200 is generated by the task 20 and specifies a requested portion of the resource 10 expressed as n1 units. The request 200 may be an initial request for n1 units of the resource 10 or a subsequent request for additional allocation of n1 units of the resource 10.

[0029] At step 100, the resource manager 12 determines the total allocation of the resource 10 to the user associated with the task 20 that would result if the request 200 is granted. The resource manager 12 records allocations of the resource 10 to users on a per user basis. For example, assume that the task 20 corresponds to user A and that the tasks 21–22 also correspond to user A and have previously been granted n2, and n3 units of the resource 10, respectively. If so, the total allocation for the user A determined at step 100 is equal to n2+n3+n1. If tasks corresponding to the user A have not previously been granted any units of the resource 10 then the total allocation for the user A determined at step 100 is equal to n1.

[0030] At step 102, the resource manager 12 determines whether the total allocation obtained at step 100 exceeds the soft limit for the user associated with the task 20. If the total allocation obtained at step 100, which includes the request 200 for n1 units, would not exceed the user's soft limit then the request 200 is granted at step 104. Otherwise, the user's hard limit is tested at step 106.

[0031] At step 106, the resource manager 12 determines whether the total allocation obtained at step 100 exceeds the hard limit for the user associated with the task 20. If the total allocation obtained at step 100, which includes the new request 200 for n1 units, would exceed the user's hard limit then the request 200 is denied at step 108. Otherwise, the high watermark is tested at step 110.

[0032] At step 110, the resource manager 12 determines whether the total allocation obtained at step 100 would cause the grand total allocation of the resource 10 to all users to exceed the high watermark of the resource 10. If the granting of the request 200 would not cause the grand total allocation to exceed the high watermark then the request 200 is granted at step 114.

[0033] If the granting of the request 200 would cause the grand total allocation of the resource 10 to exceed the high watermark then at step 112 the request 200 is denied. In addition, at step 116 the resource manager 12 enters a reduction mode for handling requests. In the reduction mode, the resource manager 12 always allows requests the reduce the consumption of the resource 10.

[0034] FIG. 3 illustrates the handling of a request 220 for allocation of the resource 10 by the resource manager 12 in the reduction mode in one embodiment. The reduction mode of handling a request for allocation provides hysteresis that prevents thrashing that would otherwise occur when one of the tasks 20–30 frees a portion of the resource 10 and then reallocates that portion when the resource 10 is near its capacity.

[0035] In this example, the request 220 is generated by the task 30 and specifies a requested portion of the resource 10 expressed as n10 units. The request 200 may be an initial request for the resource 10 by a user associated with the task 30 or a subsequent request for additional allocation of n10 units of the resource 10.

[0036] At step 130, the resource manager 12 determines the total allocation of the resource 10 to the user associated with the task 30 that would result if the request 220 is granted.

[0037] At step 132, the resource manager 12 determines whether the total allocation obtained at step 130 exceeds the soft limit for the user associated with the task 30. If the total allocation obtained at step 130, which includes the request 220 for n10 units, would not exceed the user's soft limit then the request 220 is granted at step 134. Otherwise, the hard limit is tested at step 136.

[0038] At step 136, the resource manager 12 determines whether the total allocation obtained at step 130 exceeds the hard limit for the user associated with the task 30. If the total allocation obtained at step 130, which includes the request 220 for n10 units, would exceed the user's hard limit then the request 220 is denied at step 138. Otherwise, the low watermark is tested at step 140.

[0039] At step 140, the resource manager 12 determines whether the total allocation of the resource 10 is below its low watermark. If the total allocation is not below the low watermark then the request 220 is denied at step 146.

[0040] If the total allocation is below the low watermark then the request 220 is granted at step 142. In addition, at step 144 the resource manager 12 returns to the normal mode for handling requests.

[0041] The foregoing detailed description of the present invention is provided for the purposes of illustration and is not intended to be exhaustive or to limit the invention to the precise embodiment disclosed. Accordingly, the scope of the present invention is defined by the appended claims.