CN104407915A - Node feature mechanism for distributed virtual machine monitor - Google Patents

Abstract

The present invention provides a node feature mechanism oriented to a distributed virtual machine monitor, and the implementation steps are as follows: a machine group composed of four nodes is virtualized into a physical machine through a DVMM system, and an operating system runs on it; node 0 has more memory resources, node 1 has more PCI/E devices, node 2 has more CPU resources, and node 3 has more storage resources; compared with the prior art, the present invention specifically proposes a node feature description mechanism based on DVMM , a new module NCM is added to the DVMM architecture. As a part of DVMM, the module collects and describes characteristic information of computing resources on each node in a well-designed format. In the system software, the algorithm of resource scheduling can be optimized according to these information, and cooperate with DVMM to make resource scheduling more reasonable, so that the whole system has higher performance and efficiency.

Description

Node Feature Mechanism for Distributed Virtual Machine Monitor

technical field

The invention relates to the field of virtualization technology and operating system, in particular to a distributed virtual machine monitor-oriented node feature mechanism.

Background technique

With the gradual development of virtualization technology, its performance and reliability are getting higher and higher, and its potential commercial and academic value is gradually being discovered. In recent years, the concept of cloud computing has become increasingly popular, and many manufacturers at home and abroad have released cloud computing strategies and products, which have brought great opportunities and challenges to virtualization technology.

Currently, virtualization technology has two major directions. One is to run multiple virtual machines on one physical machine, and allocate physical machine resources to multiple virtual machines. The second is to integrate the hardware resources of multiple physical machines and present them to a single operating system. At present, the virtualization technology adopted by various cloud computing vendors is based on the research results of the first direction, that is, running multiple virtual machines on one physical machine, which can ensure the resource isolation between each virtual machine and make the physical Machine resources are fully utilized. The research on the second direction is mainly the proposal of the distributed virtual machine monitor DVMM (Distributed Virtual Machine Monitor) architecture and related research.

Contents of the invention

The purpose of the present invention is to overcome the deficiencies in the prior art and provide a node feature mechanism oriented to distributed virtual machine monitors.

Technical scheme of the present invention is realized in the following manner, and implementation steps are as follows:

1) The machine group composed of four nodes is virtualized into a physical machine through the DVMM system, and the operating system runs on it;

2) Node 0 has more memory resources, node 1 has more PCI/E devices, node 2 has more CPU resources, and node 3 has more storage resources;

3) DVMM is composed of VMM on each node, and DVMM describes system hardware resources and provides information to the operating system;

4) As part of DVMM, NCM collects the characteristic information of each node, forms an NCM table, and provides an interface to access the NCM table;

5) The operating system manages resources and schedules tasks according to the resource table submitted by DVMM;

6) During the initialization process of the operating system, through the interface provided by NCM and DVMM, access the NCM table, obtain and format the NCM information of each node, so as to know the characteristics of the physical node;

7) When the operating system perceives that the task to be scheduled requires a relatively large amount of memory, it schedules it to run on node 0 with more memory, and does not easily schedule it to other nodes;

8) When the operating system perceives that the task to be scheduled is computational, it schedules it to run on node 2 with more CPUs, and does not easily schedule it to other nodes;

9) When the user informs the operating system that the task to be performed requires a large disk storage space, it will be scheduled to run on the No. 3 node with more storage resources, and it is not easy to schedule to other nodes;

10) When there is a task that needs to use a large number of PCI/E devices, it can be scheduled to run on node 1, and it is not easy to schedule to other nodes;

The scheduling strategies described in 7), 8), 9), and 10) make the corresponding tasks run efficiently, and at the same time make the entire system run efficiently; in the absence of NCM, such scheduling cannot be achieved, when these tasks are not suitable for When the nodes are running, it will lead to a large amount of inter-node communication, resulting in low system efficiency. the

The advantages of the present invention are:

Compared with the prior art, the distributed virtual machine monitor-oriented node feature mechanism of the present invention specifically proposes a DVMM-based node feature description mechanism, and adds a new module NCM in the DVMM framework, and the module is used as a DVMM Part of it is to collect and describe the characteristic information of computing resources on each node in a well-designed format, and expose this information to other modules of DVMM and upper-level system software through appropriate mechanisms. In the system software, especially the operating system, the algorithm of resource scheduling can be optimized according to these information, and cooperate with DVMM to make resource scheduling more reasonable, so that the whole system has higher performance and efficiency.

Detailed ways

The node feature mechanism oriented to the distributed virtual machine monitor of the present invention will be described in detail below.

The node feature mechanism facing the distributed virtual machine monitor of the present invention, the implementation steps are as follows:

1) The machine group composed of four nodes is virtualized into a physical machine through the DVMM system, and the operating system runs on it;

2) Node 0 has more memory resources, node 1 has more PCI/E devices, node 2 has more CPU resources, and node 3 has more storage resources;

3) DVMM is composed of VMM on each node, and DVMM describes system hardware resources and provides information to the operating system;

4) As part of DVMM, NCM collects the characteristic information of each node, forms an NCM table, and provides an interface to access the NCM table;

5) The operating system manages resources and schedules tasks according to the resource table submitted by DVMM;

6) During the initialization process of the operating system, through the interface provided by NCM and DVMM, access the NCM table, obtain and format the NCM information of each node, so as to know the characteristics of the physical node;

7) When the operating system perceives that the task to be scheduled requires a relatively large amount of memory, it schedules it to run on node 0 with more memory, and does not easily schedule it to other nodes;

8) When the operating system perceives that the task to be scheduled is computational, it schedules it to run on node 2 with more CPUs, and does not easily schedule it to other nodes;

9) When the user informs the operating system that the task to be performed requires a large disk storage space, it will be scheduled to run on the No. 3 node with more storage resources, and it is not easy to schedule to other nodes;

10) When there is a task that needs to use a large number of PCI/E devices, it can be scheduled to run on node 1, and it is not easy to schedule to other nodes;

The scheduling strategies described in 7), 8), 9), and 10) make the corresponding tasks run efficiently, and at the same time make the entire system run efficiently; in the absence of NCM, such scheduling cannot be achieved, when these tasks are not suitable for When the nodes are running, it will lead to a large amount of inter-node communication, resulting in low system efficiency.

NCM (Node Characterization Mechanism), a node characterization mechanism for distributed virtual machine monitor DVMM (Distributed Virtual Machine Monitor).

Add a new module NCM to the aforementioned DVMM architecture. As a part of DVMM, the module collects and describes the characteristic information of computing resources on each node in a well-designed format, and exposes this information to other modules and upper layers of DVMM through appropriate mechanisms system software. In the system software, especially the operating system, the algorithm of resource scheduling can be optimized according to these information, and cooperate with DVMM to make resource scheduling more reasonable, so that the whole system has higher performance and efficiency.

During the initialization process, NCM detects and identifies the physical resource feature information and topology structure of each node in the system, and forms a node feature description table in a specific format, that is, the NCM table.

The NCM table is divided into two parts. The first part describes the resources owned by each node, and the second part describes which node a single physical resource belongs to.

The first part is composed of description information of each node, and the description format of each node is the same. In the description information of each node, the first item is the type of the node, which can be divided into computing type, main memory type, storage type, network type, PCI/PCIE type, PC and mobile device. For the specific description of each type, see back. The types are not mutually exclusive, and each node can belong to multiple types. The second item is a description of the CPU of the node, including various parameter specifications of each CPU in the node. The third item is a description of the main memory of the node, including various parameter specifications of the main memory. The fourth item is a description of the node's storage resources, including various parameter specifications of the storage device. The fifth item is the description of the PCI/PCIE device of the node, including number, type, purpose, etc. The items in this table can be expanded with reference to the above according to needs and technical development.

The types of nodes can be divided into computing type, main memory type, storage type, network type, PCI/PCIE type, PC and mobile devices, etc. The computing type means that there are more CPUs on the node and has strong computing power; the main memory type means that the node has more memory and has greater memory throughput; the storage type means that the node has more storage resources, larger storage space; network type means that the node has more network cards and other devices, with strong network throughput; PCI/PCIE type means that the node has more PCI/PCIE devices; PC It means that the node is a PC, and the resource configuration is a typical PC configuration; a mobile device means that the node is a mobile device, and the resource configuration is limited. The types of nodes can be expanded according to the needs and technical development referring to the above.

The second part of the NCM table is composed of the classification description of the hardware resources in the cluster, mainly describing the affiliation relationship between a certain resource and each node. The first category is CPU resources, and each item is the corresponding relationship between the number of the physical CPU and the node number of the node where it is located. The second category is main memory resources, and each item is the corresponding relationship between the number of the memory DIMM and the node number of the node where it is located. The third category is hard disk resources, each of which is the corresponding relationship between the number of the hard disk and the node number of the node where it is located. The fourth category is PCI resources, each of which is the corresponding relationship between the number of the PCI/PCIE device and the node number of the node where it is located. The items in this table can be expanded with reference to the above according to needs and technical development.

The NCM table is stored in the firmware and is the same type of resource as the resource table that the firmware presents to the operating system. The firmware accesses the NCM table through the corresponding interface, and provides various interfaces and routines to respond to the operating system's access request to the NCM table. The operating system requests access to the NCM table resource from the firmware through a specific call.

The processing and manufacture of the distributed virtual machine monitor-oriented node feature mechanism of the present invention is very simple and convenient, and can be processed as shown in the manual.

Except for the technical features described in the instructions, all are known technologies by those skilled in the art.

Claims (1)

1. the node diagnostic mechanism of Based on Distributed virtual machine monitor, is characterized in that implementation step is as follows:

The machine group of (1) four node composition, becomes a physical machine by DVMM system virtualization, operation system on it;

(2) node 0 has more memory source, and node 1 has more PCI/E equipment, and node 2 has more cpu resource, and node 3 has more storage resources;

(3) form DVMM, DVMM by the VMM on each node and describe system hardware resources, information is supplied to operating system;

(4) as a part of DVMM, NCM have collected the characteristic information of each node, forms NCM table, provides interface accessing NCM to show;

(5) operating system submits the resource table come up to according to DVMM, management resource and scheduler task;

(6) in operating system initialization process, the interface provided by NCM and DVMM, access NCM table, obtains and formats the NCM information of each node, thus learn the feature of physical node;

(7) demand of task to internal memory will dispatched when operating system perception is larger, just schedules it to No. 0 node having more internal memory and runs, be not dispatched to other nodes easily;

(8) will dispatching when operating system perception of task belongs to calculation type, just schedules it to No. 2 nodes had compared with multi-CPU and runs, be not dispatched to other nodes easily;

(9) will carrying out when user's teaching process system of task needs larger disk storage space, just schedules it to No. 3 nodes having more storage resources and runs, be not dispatched to other nodes easily;

(10) when having task to need to use a large amount of PCI/E equipment, node 1 can be scheduled it to and run, not being dispatched to other nodes easily;

The scheduling strategy that picture (7), (8), (9), (10) describe makes corresponding task Effec-tive Function, also makes whole system Effec-tive Function simultaneously; When there is no NCM, not accomplishing such scheduling, when these tasks are run at unaccommodated node, will a large amount of inter-node communications be caused, and cause ineffective systems.