FRAMEWORK AND METHOD FOR MONITORING PERFORMANCE OF VIRTUALIZED SYSTEMS BASED ON HARDWARE BASE TOOL

• US 20140201357A1
• Filed: 05/10/2012
• Published: 07/17/2014
• Est. Priority Date: 05/27/2011
• Status: Active Grant

• Current Assignee: Transoft (shanghai), Inc. (I-Net Corporation)
• Original Assignee: Transoft (shanghai), Inc. (I-Net Corporation)
• Inventors: Tang, Changbin, Cai, Penghao
• Granted Patent: US 9,071,550 B2
• Time in Patent Office: Days
• US Class Current: 709/224
• CPC Class Codes: G06F 11/3433 for load management allocation of a server based on load conditions G06F9/505; load rebalancing G06F9/5083; redistributing the load in a network by a load balancer H04L67/1029
• CPC Class Codes: G06F 11/3466 Performance evaluation by tracing or monitoring
• CPC Class Codes: G06F 2201/815 Virtual middleware or OS functionality using virtual machines to implement generic software techniques for error detection or fault masking G06F11/1484
• CPC Class Codes: G06F 2201/86 Event-based monitoring
• CPC Class Codes: H04L 43/50 Testing arrangements

Abstract

The disclosed invention involves a framework and method based on hardware base tool to monitor the performance of virtualized systems, wherein the said framework comprises at least one master host, and each of the said master host comprises user space components, guest space components, kernel space components and hardware. The said user space components comprise policy manager, workload mediator, monitor library, and host performance monitor. The said host performance monitor is connected to workload mediator, and host performance monitor comprises user space monitor and kernel space monitor. The disclosed invention applies PMU or the similar tools to monitor the performance of virtualized systems. The performance monitoring of the disclosed invention is to monitor CPU, memory, cache, IO, network, processes, etc. of the host of virtualized systems. Meanwhile, the method based on hardware to monitor performance in this disclosed invention resolves the problem to acquire performance data for virtualized systems.

Claims

1. A framework based on hardware base tool monitoring the performance of virtualized systems, wherein the said framework comprises at least one master host, and each master host comprises user space components, guest space components, kernel space components and hardware, wherein,

the user space components comprising a policy manager, a workload mediator, a monitor library and a host performance monitor connected successively, the host performance monitor also connected to the workload mediator, and the host performance monitor comprising a user space monitor and a kernel space monitor;

the guest space components comprising at least one virtual machine (VM) connected to the host performance monitor via a kernel serial port, and the VM comprising a VM performance agent for collecting performance data of VM resources;

the kernel space components comprising performance monitoring base tool application interface (API) and performance monitoring base tool core code module connected successively to the monitor library, as well as a task scheduler, a memory management unit, a network driver, a file system, and a device driver, and the network driver being connected to the workload mediator;

the hardware being for supporting a peripheral performance monitor base tool, the hardware comprising a PMU connected to the performance monitoring base tool core code module, PMD and PMC connected to the PMU respectively, as well as a CPU, a memory, a network card, and a hard disk, and said the network card being connected to the network driver.

2. The framework as claim 1, wherein the said framework further comprises at least one slave host, the master host being connected to the said at least one slave host via a net pipe to form a cluster, multiple master hosts and their respective slave hosts forming multiple clusters, one of the clusters communicating with other clusters in centralized, distributed or mixed communications style, the slave host comprising the host performance monitor, and the host performance monitor of the slave host being connected to the network card in the master host.

3. The framework as claim 1, wherein the virtualized systems comprise an in-system resource monitor and an out-of-system resource monitor, the policy manager, the workload mediator and the host performance monitor of the said master host forming as upper-layer architecture, wherein each policy manager is a set of at least one workload mediator, each workload mediator is a set of at least one host performance monitor, each host performance monitor is a set of at least one said VM performance agent a set of the user space monitor and the kernel space monitor, and the host performance monitor being inherited respectively by the in-system resource monitor and the out-of-system resource monitor.

4. The framework as claim 3, wherein the said policy manager is used to interpret user policy requests, specify the conditions to be reached for CPU, memory, network, IO, and cache, and divide the user policy requests into high, median, and low levels, when the master host of the cluser can not satisfy the user policy requests, the said policy manager enquiring the policy manager of the master host of other clusters simultaneously or in a prioritized order according to the cluster policy enquired, so as to satisfy the user policy requests.

5. The framework as claim 3, wherein,

the workload mediator in the said master host acquires policy requests from the said policy manager, collecting resource performance data of individual slave hosts in the same cluster where the master host resides via the said net pipe, consolidating to get performance of the entire cluster according to the said performance data, and being responsible for executing policy to the resources of individual slave hosts;

and

the performance data collected by the said workload mediator is used to determine the idle/busy state of the said slave host resources, the state transiting from resource idle to resource busy if the said performance data is over a preset threshold, the state transiting from resource busy to resource idle if the said performance data is lower than the said preset threshold, if the state transits from resource busy to resource not-working, there being anomaly, if the state transits from resource not-working to resource busy, the resource returning to normal.

6. The framework as claim 3, wherein the said host performance monitor is used to monitor a user space, a kernel space and a guest space, wherein

monitoring the user space means monitoring performance information of all processes in the user space using the said out-of-system resource monitor;

monitoring the kernel space means activating, configuring, and stopping the said kernel space monitor, as well as monitoring in-system resources via the said in-system resource monitor;

monitoring the guest space means (1) interacting with the said VM performance agent running at individual VM via the said kernel serial port, acquiring data from the said VM, and the said VM performance agent monitoring virtual resources via the said in-system resource monitor;

(2) monitoring VM via the said out-of system resource monitor, the VM being regarded as a process and acquiring performance of the process, and determining if the said VM satisfies performance requests to match the acquired data by the said VM performance agent.

7. The framework as claim 6, wherein the said in-system resource monitor acquires data by soliciting two relative performance data and subtracting them at a certain time interval based on the performance monitoring base tool, the resources monitored by the said in-system resource monitor comprising CPU resources, memory resources, IO resources, cache resources and network resources, wherein,

the CPU resources comprising total time of processes used by CPU to handle user-space state, total time of processes used by CPU to handle kernel-space state, total CPU idle time, total number of hard interrupts handled by CPU, and total number of soft interrupts handled by CPU, which is acquired through OS performance commands;

the memory resources comprising all memory, free memory, shared memory, and buffer memory obtained through OS performance commands;

the IO resources comprising number of reading and writing of one or more processes obtained through OS performance commands;

the cache resources comprising the data of cache, TLB miss and hit obtained through said performance monitoring base tool; and

the network resources comprising network traffic of the network state obtained through OS performance commands.

8. The framework as claim 6, wherein the said out-of system resource monitor acquires data by soliciting two relative performance data and subtracting them at a certain time interval based on the performance monitoring base tool and the spaces monitored by the said out-of-system resource monitor comprise the user space and the guest space, wherein:

monitoring the user space comprising acquiring CPU usage data, cache usage data and TLB miss data in the user space via OS performance commands;

monitoring the guest space comprises acquiring CPU usage data, cache usage data and TLB miss data of the said processes in the user space via OS performance commands, and observing whether the CPU running data of the said VM being over threshold, so as to control the CPU occupancy factor of the VM via other OS commands.

9. The framework as claim 2, wherein the framework further comprises inherited components of the said monitor library, and the monitor library, inherited components of the monitor library, and the said host performance monitor in the said master host form a bottom-layer architecture, wherein the said monitor is the top-layer class while its inherited components is monitor, the said monitor being inherited by the host performance monitor, the host performance monitor being respectively inherited by the said kernel-space monitor, the user-space monitor, and the VM performance agent, each of the said monitor being a set of at least one monitored object, each of the said monitored object being a set of at least one event, and each of the said event representing one atom monitoring event.

10. The framework as claim 9, wherein the said monitor library creates a monitor for all processes to monitor resources, and the monitor is loaded into the monitor library, when the said monitor library process starts, the resources capable of being monitored by the said performance monitoring base tool being decomposed into atom resources, proceeding to monitor the resources, and the resource monitored information being dispatched to individual said monitor, when the said performance monitoring base tool discovers CPU interrupt overflow, the anomaly is informed to the said monitor, and the said monitor library being used to add or delete monitors.

11. The framework as claim 9, wherein the said monitor is used to save configuration, load configuration, add/delete the said monitored object, start/stop the said monitored object, and refresh/save monitoring information.

12. The framework as claim 9, wherein the said monitored object is created according to the configuration information in a configuration file, and the configuration file comprises objects capable of being monitored by the said performance monitoring base tool.

13. The framework as claim 9, wherein,

if the said performance monitoring base tool monitors based on one-time event, the threshold of the performance monitoring base tool=maximum value−

1;

if the said performance monitoring base tool monitors based on n events, the threshold of the performance monitoring base tool=maximum value−

n wherein n is a nature number greater than 1;

if the said performance monitoring base tool monitors based on event probability, the threshold of the performance monitoring base tool=maximum value−

a random number of n events;

if the said performance monitoring base tool monitors based on the event number in a certain time interval, the number of CPU events triggered by the performance monitoring base tool is tallied, and when clock triggers CPU, the total number of events of individual time interval is reported;

if the said performance monitoring base tool monitors based on n types of events, the number of events triggered to CPU in each time interval i (i=1, . . . n) is tallied by the performance monitoring base tool, and the reported result is the total number of same-type events of each time interval multiplying n; and

if the said performance monitoring base tool monitors based on resource utilization, the utilization between time T1 and T2=N/((T2−

T1)×

number of interrupts of CPU), wherein N is the total number of commands.

14. A method based on the framework based on hardware base tool monitoring the performance of virtualized systems as claim 1, wherein the method comprises the following steps:

firstly, an administrator making performance mediation requests to the said policy manager via a graphical interface or an user application via the said performance monitor application interface;

secondly, after the said policy manager interprets the said performance mediation requests, the said workload mediator being instructed to report performance conditions of the resources of individual slave host to the said master host via the said network pipe; and

finally, the said policy manager reporting the performance conditions to the said user application process or the administrator.