MEMBER-ORIENTED HYBRID CLOUD OPERATING SYSTEM ARCHITECTURE AND COMMUNICATION METHOD THEREOF

• US 20150067135A1
• Filed: 11/06/2014
• Published: 03/05/2015
• Est. Priority Date: 08/22/2013
• Status: Active Grant

• Current Assignee: Shanghai Inspur Cloud Computing Service Co., Ltd. (Inspur Group Company Limited)
• Original Assignee: Inspur Electronic Information Industry Company Limited (Inspur Group Company Limited)
• Inventors: WANG, Endong, ZHANG, Dong, LIU, Zhengwei, QI, Kaiyuan, LIU, Junpeng, GUO, Feng, LIU, Chenping, GAO, Fei, ZHU, Bo
• Granted Patent: US 9,307,017 B2
• Time in Patent Office: Days
• US Class Current: 709/223
• CPC Class Codes: G06F 9/45558 Hypervisor-specific management and integration aspects
• CPC Class Codes: G06F 9/5072 Grid computing
• CPC Class Codes: G06Q 10/06 Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
• CPC Class Codes: H04L 41/0803 Configuration setting
• CPC Class Codes: H04L 41/5054 Automatic deployment of services triggered by the service manager, e.g. service implementation by automatic configuration of network components
• CPC Class Codes: H04L 63/10 for controlling access to devices or network resources
• CPC Class Codes: H04L 67/10 in which an application is distributed across nodes in the network software deployment G06F8/60; multiprogramming arrangements G06F9/46
• CPC Class Codes: H04L 67/1001 for accessing one among a plurality of replicated servers
• CPC Class Codes: H04L 67/1008 based on parameters of servers, e.g. available memory or workload monitoring of computer activity G06F11/30
• CPC Class Codes: H04L 67/1034 Reaction to server failures by a load balancer
• CPC Class Codes: H04L 67/133 Protocols for remote procedure calls [RPC]
• CPC Class Codes: Y04S 40/00 Systems for electrical power generation, transmission, distribution or end-user application management characterised by the use of communication or information technologies, or communication or information technology specific aspects supporting them

Abstract

A member-oriented hybrid cloud operating system architecture and a communication method thereof are provided. A hybrid architecture is established based on layer, object and message models, and a member-oriented idea is applied to manage constituent members and a processing environment thereof. On this basis, high-efficient routing, read-write separation and load balancing are performed on a member processing cluster, satisfying the requirements of being open and compatible, loosely coupled and extensible of a cloud operating system, and solving the self-management problem, the horizontal scaling problem of members and the high-availability problem of stateful members of the existing cloud operating system.

Claims

1. A member-oriented hybrid cloud operating system, wherein the system establishing a hybrid architecture based on layer, object and message models, and applying a member-oriented idea to manage constituent members and a processing environment thereof, and on this basis, the system performing high-efficient routing, read-write separation and load balancing on a member processing cluster, satisfying the requirements of being open and compatible, loosely coupled and extensible for a cloud operating system, and solving the self-management problem, the horizontal scaling problem of members and the high availability problem of stateful members of an existing cloud operating system,

from the perspective of a layer model, the system being divided into a portal layer, a logic layer, an adaptation layer and an implementation layer from top to bottom, with each layer being relatively independent, the openness being enhanced by respectively defining a standard interface at each layer, and the compatibility being enhanced by adapting different functions at various layers;

from the perspective of an object model, the cloud operating system being composed of the functional modules of a cloud portal, cloud management portal, cloud resource management, monitoring management, metering and billing, service approval, and authorization and authentication, various functional components performing communication by Rest message-based calling, capable of being combined with each other freely, and being distributed and deployed as required, a new module being able to be developed in a value-added manner as required, and the extensibility of a platform being enhanced by realizing interoperation among different modules in the logic layer, the cloud operating system which is installed in a minimized manner comprising only the modules of a cloud portal, cloud management portal and cloud resource management, on this basis, monitoring, billing, approval or other modules being customized and extended as required, performing communication by Rest message-based calling, and being distributed, deployed and developed in a value-added manner as required, and the extensibility of the system being enhanced by realizing interoperation among different modules in the logic layer; and

a message-based communication being introduced to support asynchronous calling, and a message communication interface JMS being used to transmit a Rest message, thus making the system architecture further decoupled, and on this basis, a construction-oriented design being applied, and a member management portal being responsible for managing metadata of members and monitoring operation states thereof;

despite being able to realize extension, distribution and deployment as required, an object architecture belonging to an RPC (remote process call) synchronization communication manner, a sending end is able to continue execution after waiting for the response from a receiving end, and the processes of the two parties are closely coupled, and with the magnification and complication of the system, an association relationship among the members being over-complicated, regarding this problem, a message-based communication being introduced on the basis of the object architecture, a message communication interface JMS being used to transmit the Rest information, making the life periods of the sending and receiving ends different, supporting asynchronous calling and making the system structure further decoupled, at least the operations of turning on, shutting down and suspending of a virtual machine between a portal and a cloud resource layer being realized in an asynchronous manner, and the portal being able to return without the need to wait for a response after sending out a command, thus improving user interaction effects;

the cloud operating system based on the above-mentioned hybrid architecture being able to satisfy the requirements of openness, compatibility and extension, and on this basis, based on a construction-oriented design idea, the member management portal being responsible for managing metadata information about the members, supporting at least the operations of registration, deletion, modification and querying, wherein,

a member is a triple, comprising;

a name, a service set, an access address, and a description;

a service is a quadruple, comprising;

a name, a type, a message protocol, a parameter list, a key name, a functional description, and a non-functional description;

in addition to describing and managing the members, a member management module further monitors a processing environment thereof, provides a basic service for ensuring the scalability and availability of the members, and perfects the self-management capability of the cloud operating system, and when a member is registered, the system allocates a username user, a password psw and a unique member id thereto, and then an access process of a member processing cluster comprises;

1) the processing cluster initiating an access request to a system bus with an address being url, the system bus verifying a username, password and id of an access node, and if the verification is passed, establishing a connection, the code being;

connection=ConnectionFactory.createConnection(user,psw,url);

2) establishing a write operation topic, and each processing node of the member subscribing to a write operation from the topic;

write _topic=session,createTopic(id+“

WRITE_TOPIC”

);

write_topic_consumer=session.createConsumer(write_topic);

3) the member realizing write processing in an onMessage method of a writeTopicListener, and registering to the system bus;

write_topic₁₃ consumer.setMessagetistener(writeTopicListener);

4) establishing a read operation queue group according to the number mum of processing nodes of the member, each processing node corresponding to one queue subscription read operation

read_queue=session.createMultiQueue(num,id+“

READ_QUEUE”

);

read_queue_consumer=session.createConsumer(read_queue); and

5) realizing a specific read processing function in an onMessage method of a read QueueListener, and registering to the system bus

read_queue_consumer.setMessageListener(readQueueListener);

on the basis of the above-mentioned read-write separation queue groups, service types of various members are distinguished in the member management module, setting the service types as idempotent or non-idempotent, an idempotent operation belonging to an unstateful operation, the result of each execution in the same state being the same, and a non-idempotent operation belonging to a stateful operation, the result of each execution in the same state being different, a router performing routing according to the service type, the non-idempotent operation being sent to a unique write queue, and the idempotent operation being sent to different read queues according to a load balancing strategy;

a read operation load balancing process comprises

1) calculating a processing capability of a node;

if a CPU frequency, memory capacity and I/O bandwidth of a node i are respectively C_i, M_i and B_i, various resources of a cluster being the sums of various resources of nodes, i.e., C=Σ

C_i, M=Σ

M_i, B=Σ

B_i;

then a CPU weight value of the node i is W_i^CPU=C_i/C, the memory capacity W_i^RAM=M_i/M, and the I/O bandwidth W_i^IO =B_i/B; and

if the required resource proportions of the member service are respectively p^CPU, p^RAM and p^IO, the processing capability of the node i being

W_i=p^CPUW_i^CPU+p^RAMW_i^RAM+p^IOW_i^IO;

2) calculating a load of each node according to a read-write operation weight value;

if a read-write operation overhead ratio of a read queue L_r to a write queue L_w is a, a load of the node i being L_i=L^R_i+aL^W and

a load state of each node being S_i=L_i/W_i and

3) selecting a node with the lightest load for routing;

the write operation being performed by means of pipeline, so as to improve the data write-in efficiency, the process thereof comprises;

a node 1 firstly writing in data, after writing in one data fragment of 64 KB, while continuing to receive data, forwarding the written-in 64 K data to a node 2, and receiving and forwarding data in the same manner from the node 2 to a node n until the node n writing in the last data fragment not exceeding 64 KB; and

on the basis of the above-mentioned communication manner, a node monitoring module further sending joining, quitting, invalidation and recovery events of the member processing node to the member management module, wherein a node joining event refers to adding a processing node to the member;

a node quitting event refers to revoking a processing node from the member;

a node invalidation event, refers to one processing node of the member being unavailable; and

a node revival event refers to one unavailable node of the member recovering availability.

2. A communication method for a member-oriented hybrid cloud operating system, comprising a highly-available member cluster communication and a horizontal scaling member cluster communication, wherein:

the highly-available member cluster communication comprises;

a member management module establishing a read-write separation queue group for each member cluster, the member management module further distinguishing service types of members into idempotent and non-idempotent, a router performing routing according to the service types, a non-idempotent operation being sent to a unique write queue, an idempotent operation being sent to different read queues according to a load balancing strategy, and on this basis, a node monitoring module sending joining, quitting, invalidation and recovery events of a processing node to the member management module, and the member management module further adjusting a queue structure according to a node change event, this method improving the communication performance of a stateful member cluster by means of read-write separation and load balancing, and adjusting the queue structure according to a node change to ensure the high availability of a member, and the member management module further adjusting the queue structure according to the node change event sent by the monitoring module, the process thereof comprising;

1) when a node joins, establishing a read operation queue for the node in a queue group, and the node subscribing to a read operation from the queue, and subscribing to a write topic from the write topic;

2) when the node quits, deleting a read queue corresponding to the node, and closing the subscription to the write topic thereof;

3) when the node invalidates, stopping the sending of a read. request to the read queue of the node, and reserving a write operation for the node in the write topic; and

4) when the node revives, synchronizing the write operation in the write topic, and recovering to send the read operation request to the read queue corresponding to the node;

an unstateful cluster load balancing method allocating operations to various nodes in a completely balanced way, which will result in inconsistent processing results in a stateful condition, and by establishing a read-write separation queue and balancing the load regarding node capabilities, the above-mentioned method is capbale of avoiding this problem and improving the communication performance of the stateful. member cluster, and by adjusting the queue structure according to a node change event, the high availability of the stateful cluster is ensured; and

the horizontal scaling member cluster communication comprises;

the member management module establishing a polygon queue structure for each service of a member according to the number of nodes, performing binary lookup routing according to a Hash value, the algorithm efficiency being determined by the number of Hash packets which is much smaller than a keyword scale, thus improving the routing efficiency, and in the case where the number of nodes changes, only some node data states needing to be adjusted, improving the dynamic scaling efficiency, and on the basis of the above-mentioned communication method, in an initialization stage, Hash intervals needing to be processed by various nodes being divided according to node processing capabilities, forming a routing table, realizing the distribution of data states according to proportions, and achieving load balancing;

the member management module establishing a polygon queue group for each service of a member according to the number of nodes, each queue corresponding to one Hash value interval or packet, and performing binary lookup routing according to the Hash value, the routing algorithm comprising;

1) calculating a Hash value h of a key;

2) initializing a lower limit i of a packet as 1, and an upper limit j of the packet as the total number of packets in;

3) repeating;

4) calculating a middle packet t=(i+j)/2, and checking whether the packet t contains h;

5) if h is smaller than the lower limit of the current packet, updating the upper limit j=t−

1;

6) if h is greater than the upper limit of the current packet, updating the lower limit i=t+1; and

7) otherwise, returning to a node where an interval is located for routing; and

the above-mentioned algorithm being equivalent to performing binary lookup by taking a packet as a node, the complexity being O(log₂n), the algorithm efficiency being determined by the number of packets, and since the number of packets is much smaller than the keyword scale, the routing efficiency being improved, in addition, since a packet strategy is applied, only the keyword scale of some packets and data states of corresponding nodes need to be adjusted in the case where the number of nodes changes, improving the dynamic scaling efficiency.