Dynamically modifying the resources of a virtual server

Abstract

A system and a method dynamically adjusts the quality of service guarantees for virtual servers based upon the resource demands experienced by the virtual servers. virtual server resource denials are monitored to determine if a virtual server is overloaded based upon the resource denials. virtual server resources are modified dynamically to respond to the changing resource requirements of each virtual server. Occasionally, a physical host housing a virtual server may not have additional resources to allocate to a virtual server requiring increased resources. In this instance, a virtual server hosted by the overloaded physical host is transferred to another physical host with sufficient resources.

Description

CROSS REFERENCE TO RELATED APPLICATIONS

In one embodiment, a “jumping-window” technique is used. The jumping-window technique measures the number of resource denials a received in consecutive windows of time length t. A new time interval t starts immediately after the end of the last time interval t. In another embodiment, a “moving-window” technique is used. The moving-window technique measures the number of resource denials a received in a continuously moving window of time length t. In the moving-windows technique, all windows of time length t are measured.

The virtual server resource monitor 110 checks 340 if the metric a(t) calculated is beyond the pre-specified threshold T. This determination is made individually for each type of resource denial signal (340A, 340B, 340C and 340D), and need not be made simultaneously. Each different type of resource denial signal 312 may have a different pre-specified threshold T.

If the metric a(t) representing the average resource denial rate does not exceed the threshold T, the method continues to calculate a(t) 330 so that resource denials are continuously monitored. Using the jumping-window technique, after the next consecutive time interval t passes, the method will again check 340 if a(t)>T. Using the moving-windows technique, a continuous loop of steps 330 and 340 is used to measure each continuously-moving window of time t. In another embodiment, a pre-specified schedule for repeating calculating mean resource denials 330 and checking 340 if the threshold T has been exceeded can be established to limit the amount of processing required by the virtual server resource monitor 110.

However, if the metric a(t) does exceed the threshold T, a “resource overloaded” signal is sent 350 to the virtual server resource modifier 120. Each type of resource denial signal 312 has an associated resource overloaded signal. FIG. 3 shows four examples of resource overloaded signals: disk resource overloaded signal 350A, memory resource overloaded signal 350B, network resource overloaded signal 350C, and CPU resource overloaded signal 350D. It is to be understood that there may be many other types of signals indicating an overloaded resource. The examples shown herein are used purely for illustrative purposes.

FIG. 4 shows a flowchart of an embodiment of a method for determining when to increase or decrease a particular resource allocation within a virtual server. The virtual server resource modifier 120 performs the method shown in FIG. 4. A separate analysis using the method of FIG. 4 is performed for each type of resource being monitored.

The modifier 120 waits 410 to receive a resource overloaded signal 350 from the virtual server resource monitor 110. When a resource overloaded signal 350 is received, the modifier 120 checks 420 to determine whether the signal 350 falls within a pre-specified “hysteresis time window” H. The hysteresis time window H check 420 damps the modifier 120 system to avoid rapid changes in the system state. For example, in a situation in which a virtual server has overloaded its existing memory resource allocation, the virtual server may attempt to access memory repeatedly before the memory resource allocation is increased. Each memory access attempt may generate a memory resource overloaded signal 350B. The modifier 120 only needs to respond to one of these signals. The hysteresis time window H check 420 avoids repetitive responses to resource overloaded messages. Thus, the modifier 120 checks 420 whether the most recently received resource overloaded signal 350 (received at T₁) is close in time (within the hysteresis time window H) to a previously received resource overloaded signal 350 (received at T₀) for a particular resource:
T₁−T₀<H?

If the recent and previous resource overloaded signals have occurred close enough in time to fall within the pre-specified hysteresis time window H, no further action will be taken and the modifier 120 returns and waits 410 to receive another resource overloaded signal 350. If the current resource overloaded message is not received within the hysteresis time window H, the modifier 120 proceeds to increase 430 the virtual server resource allocation.

The resource allocation for a particular overloaded resource is increased 430 by a pre-specified amount i. Amount i may be specified as a certain percentage of the resources of a physical host, or alternatively amount i may be specified as a certain number of resource units. Amount i may also be specified as a certain percentage of each particular virtual server's current resource allocation, e.g. increase a resource by 5% of its current value. After a particular resource has been increased the modifier 120 sets 440 a timer for a pre-specified time period.

When the timer expires, the modifier 120 determines 450 if the recently increased resource is being fully utilized. In one embodiment, a resource is fully utilized if a corresponding resource denial signal has been received within the timer period 440 after the resource was increased.

If the resource is determined 450 to be fully utilized, the modifier 120 returns and waits 410 for an overloaded signal. However, if it is determined that the resource is not being fully utilized, the modifier 120 decreases 460 the resource by a pre-specified amount d. Amount d may be specified as a certain percentage of the resources of a physical host, or amount d may be specified as a certain number of resource units. Amount d may also be specified as a certain percentage of each particular virtual server's current resource allocation, e.g. decrease a resource by 10% of its current value.

In one embodiment, d (the resource decreases amount) is larger than i (the resource increase amount). This allows unused resources to be decreased aggressively, but overloaded resources to be increased cautiously. In another embodiment, d and i are set such that the resource allocation is increased and decreased by equal amounts. For example, assume that the increase in virtual server resources i is specified as a percentage of each virtual server's current resource allocation. The decrease in virtual server resources d is specified as d=1−(1/1+i), which returns the resource allocation to its previous level. Once the resource reaches a fully utilized state, the modifier 120 then returns to waiting 410.

FIG. 5 shows a block diagram of an embodiment of a process for performing resource load balancing among physical hosts, in the context of a working example of overloaded physical host 160A. The physical host load balancer 130 periodically monitors the resource usage of a group of physical hosts 160 (160A, 160B and 160C) and transfers virtual servers to different ones of these physical hosts 160 in order to balance the resource loads between the physical hosts 160. Requests to increase virtual server resource allocations are also sent to the physical host load balancer 130 in order to assist in the balancing of physical host 160 resource loads. This process is next explained by example.

In this example, physical host load balancing module 130 receives a signal 510 from the virtual server resource modifier 120 indicating that virtual server 162B requires an increased resource allocation. This signal is used as an input 520 into the load-balancing calculator 530. The load-balancing calculator 530 also requests and receives as input the current physical host resource loads 535 from the physical host resource monitor 540.

The physical host resource monitor 540 performs periodic physical host resource checks 545 upon the group of physical hosts 160 (160A, 160B and 160C). Resource checks 545 monitor the current virtual server resource guarantees in each quality of service table for each physical host 160.

The load-balancing calculator 530 determines whether a virtual server's request for additional resources 510 will overload the particular physical host currently hosting the virtual server. Using the example shown in FIG. 5, the load-balancing calculator 530 determines whether physical host 160A is capable of supporting the request for additional virtual server 162B resources 510. If the resource request 510 exceeds the available resources of physical host 160A, the load-balancing calculator 530 determines that physical host 160A is overloaded.

In one embodiment, the load-balancing calculator 530 uses an easiest fit heuristic to find the physical host that has the most available resources. Each different type of resource is associated with an ordinal and a weight. The i^th resource R_i has ordinal i and weight w_i. For example, resource R₁ represents disk resources, R₂ represents memory resources, R₃ represents network resources and R₄ represents CPU resources. The weights for each respective resource are determined by the system operator.

Let R_i(V) denote the resource requirement of the virtual server under consideration, e.g. virtual server 162B, including the requested resource increase from signal 510. Let R_i(S_j) denote the resource availability at the j^th physical host. The load-balancing calculator 530 computes the weighted resource availability of physical host j as the sum over i:

$\sum _i{w}_i*\left(R_i\left(S_j\right)-R_i\left(V\right)\right)$

Using the easiest fit heuristic, the load-balancing calculator 530 will select the physical host with the largest weighted resource availability to receive the virtual server 162B (in the example of FIG. 5, physical host 160B). The choice of physical host 160B is subject to the constraint that the selected physical host 160B has sufficient resources to meet the resource demands of virtual server 162B. The load-balancing calculator 530 sends 550 a signal 560 to the dynamic virtual server mover 140 indicating that virtual server 162B is to be transferred to physical host 160B.

It will be understood by one of skill in the art that load-balancing calculator 530 may use other criteria for selecting which virtual server to transfer out of an overloaded physical host. In the embodiment given above, the load balancing calculator 530 transfers the virtual server that has most recently requested additional resources. However, in another embodiment, the load balancing calculator could select, for example, the smallest virtual server within an overloaded physical host for transfer, regardless of which virtual server has recently made a request for increased resources.

FIG. 6 is a flowchart of an embodiment of the process for transferring a virtual server from one physical host to another physical host. The dynamic virtual server mover 140 directs the process of FIG. 6. This process is next explained by example.

In this example, virtual server 162B is transferred from old physical host 160A to new physical host 160B. The mover 140 waits 610 to receive a transfer virtual server signal 560. The mover 140 receives a signal 560 directing the transfer of virtual server 162B from physical host 160A to physical host 160B. The mover 140 directs physical host 160A to store 620 local state information associated with virtual server 162B in the file system 150. As shown in FIG. 1, file system 150 is commonly accessible from physical hosts 160A, 160B and 160C.

Mover 140 next directs physical host 160A to stop 630 local processes associated with the virtual server being moved, e.g. virtual server 162B. Mover 140 directs physical host 160B to access 640 the virtual server 162B state information stored in file system 150. Mover 140 directs physical host 160B to start 650 processes associated with virtual server 162B locally. This enables virtual server 162B to begin running locally in physical host 160B. The user of virtual server 162B is then transferred 660 from physical host 160A to physical host 160B by transferring the virtual server 162B address to the new physical host 160B. As explained previously, the mover 140 may use either “make, then break” timing or “break, then make” timing for the transfer process. Although the invention has been described in considerable detail with reference to certain embodiments, other embodiments are possible. As will be understood by those of skill in the art, the invention may be embodied in other specific forms without departing from the essential characteristics thereof. For example, the dynamic resource configuration module may support different numbers of physical hosts. Additionally, different fitting heuristic methods may be used to select physical hosts for receiving transferred virtual servers during load balancing among the physical hosts. Accordingly, the present invention is intended to embrace all such alternatives, modifications, and variations as fall within the spirit and scope of the appended claims and equivalents.

Claims

1. A network system for dynamically modifying the computer resources allocated to a virtual server, the network system comprising a plurality of physical hosts, the virtual server operating in a first physical host, the computer resources allocated to the virtual server being specified as a quality of service guarantee, the network system comprising:

a virtual server resource monitor communicatively coupled to the first physical host and configured to monitor resource denials and to send a virtual server overloaded signal in response to the resource denials;

a virtual server resource modifier communicatively coupled to the first physical host and configured to receive the virtual server overloaded signal and, in response to the virtual server overloaded signal, to modify a resource allocation for the virtual server and to send a virtual server resource modification signal;

a load balancing module communicatively coupled to the plurality of physical hosts and configured to receive the virtual server resource modification signal and to determine whether the first physical host is overloaded and, in response to a determination that the first physical host is overloaded, to send a physical host transfer signal that indicates a second physical host; and

a dynamic virtual server mover communicatively coupled to the plurality of physical hosts and configured to receive the physical host transfer signal and, in response to the physical host transfer signal, to transfer the virtual server from the first physical host to the second physical host.

2. The network system of claim 1, further comprising a file system communicatively coupled to the plurality of physical hosts and configured to store virtual server system files.

3. The network system of claim 2, wherein the dynamic virtual server mover is further configured to direct the first physical host to store, in the file system, a set of system files for the virtual server and to direct the second physical host to access, from the file system, the set of system files for the virtual server, thereby transferring the virtual server from the first physical host to the second physical host.

4. A computer program product to be executed in a computer for dynamically modifying the computer resources allocated to a virtual server operating in a first physical host in a network system, the network system comprising a plurality of physical hosts, the computer resources allocated to the virtual server being specified as a quality of service guarantee, the computer program product comprising:

program code for creating a virtual server resource monitor communicatively coupled to the first physical host and configured to monitor resource denials and, in response to the resource denials, to send a virtual server overloaded signal;

program code for creating a virtual server resource modifier communicatively coupled to the first physical host and configured to receive the virtual server overloaded signal and, in response to the virtual server overloaded signal, to modify a resource allocation for the virtual server and to send a virtual server resource modification signal;

program code for creating a load balancing module communicatively coupled to the plurality of physical hosts and configured to receive the virtual server resource modification signal and to determine whether the first physical host is overloaded and, in response to a determination that the first physical host is overloaded, to send a physical host transfer signal that indicates a second physical host; and

program code for creating a dynamic virtual server mover communicatively coupled to the plurality of physical hosts and configured to receive the physical host transfer signal and, in response to the physical host transfer signal, to transfer the virtual server from the first physical host to the second physical host.

5. A method performed by a computing device, having a processor and memory, for modifying the computer resources allocated to a virtual server operating in a first physical host of multiple physical hosts, comprising:

receiving an indication that a first physical host is overloaded, wherein the indication is based on a determination that a virtual server is overloaded and wherein the determination that a virtual server is overloaded is based on one or more resource unavailable messages resulting from denied requests to modify a resource allocation;

determining that a second physical host can accommodate the requested modified resource allocation; and

generating a physical host transfer signal that indicates the second physical host and transferring the virtual server from the first physical host to the second physical host.

6. A computer-readable storage device storing instructions that, when executed by a computing device, cause the computing device to perform operations configured to modify computer resources allocated to a virtual server operating in a first physical host of multiple physical hosts, the operations comprising:

receiving an indication that a first physical host is overloaded, wherein the indication is based on a determination that a virtual server is overloaded and wherein the determination that a virtual server is overloaded is based on one or more resource unavailable messages resulting from denied requests to modify a resource allocation;

determining that a second physical host can accommodate the requested modified resource allocation; and

generating a physical host transfer signal that indicates the second physical host and transferring the virtual server from the first physical host to the second physical host if the first physical host is overloaded.

7. A system for modifying the computer resources allocated to a virtual server operating in a first physical host of multiple physical hosts, the system comprising:

one or more processors and one or more memories;

a component configured to receive an indication that a first physical host is overloaded, wherein the indication is based on a determination that a virtual server is overloaded and wherein the determination that a virtual server is overloaded is based on one or more resource unavailable messages resulting from denied requests to modify a resource allocation;

a component configured to determine that a second physical host can accommodate the requested modified resource allocation; and

a component configured to generate a physical host transfer signal that indicates a second physical host and to transfer the virtual server from the first physical host to the second physical host if the first physical host is overloaded.