Due to the diversity of resources and different personalized management requirements, Cloud management is faced with great challenges in complexity and difficulty. For constructing a management system to satisfy a specific requirement, a redevelopment solution based on existing system is usually more practicable than developing the system from scratch. However, the difficulty and workload of redevelopment are also very high. This paper demonstrates Jade Cloud, which is a runtime model based Cloud management system. Based on the runtime models of OpenStack and Hyperic, Jade Cloud can manage both the infrastructure and software resources in a unified manner. Demo link: https://www.youtube.com/watch?v=o2SiTGL689Q
With the rapid development of Cloud computing, it brings unprecedented challenges to management of diverse Cloud resources, which mainly include two aspects. First, there are many different kinds of hardware and software resources in Cloud which have to be managed well. Second, there are different personalized management requirements consisting of specific scenarios and appropriate management styles.
To satisfy the personalized management requirements, Cloud administrators
usually conduct redevelopment based on the existing management systems other than
develop a new system from scratch [
To solve the above issues, we present a runtime model based approach to
managing diverse Cloud resources and have published a paper in Models 2013 [
To further demonstrate the above approach, in this paper, we present Jade Cloud
a runtime model based Cloud platform managing diverse Cloud resources. Jade Cloud
is constructed based on OpenStack [
We only define the architecture-based meta-model specifying the structure of the running system (OpenStack or Hyperic) and the access model specifying how to ma1 *
Project -id
Appliances 1 1 1 -t1enantId *
1
nipulate the system’s elements, then the construction of runtime models of OpenStack
and Hyperic are completed with the help of SM@RT [
Cloud -projectList
*
-id -applianceId -name -ip
ApacheSwConfig -id 1 1 ---asvvteaaritsluiasobnility ---inapapmpelianceId -requestsPerSecond -requestsPerMinute -BytesPerMinute -....
1 1 ---JsavOtvenaarAtisSuliSsaowbnCiolniftiyg ----niaiappd*mpMelyiSanQcLe1Id1 1 -----pasvcrvtepMoaaruycitsUSeluisQsasoaLsbngSMiewelmiCotorynyfSiigze -usedMemory -tablesPerMinute -totalMemory -filesPerMinute -.... -queriesPerMinute -.... We deploy Jade Cloud on 11 physical servers and it can support about 150 appliances. Figure 5 shows a screenshot of the user interface of Jade Cloud. Administrators can manage the resources through updating the system deployment diagram on the web interface. In the diagram, the elements of the appliance stand for virtual machines with software products, which compose the runtime model of the Jade Cloud. Taking the appliance named “APACHE” as an example, we can get and set the hardware information (freeMemory, LoadAverage_5min etc.) and the software information (Bytes_Served_per_Minute, Requests_Served_per_Second etc.). Jade Cloud reuses and reorganizes the management interfaces of OpenStack Compute and Hyperic, instead of modifying the two systems. The runtime model based management system also benefits from many model-centric analyzing or planning methods and mechanisms. We have implemented three kinds of model-based Cloud services on Jade Cloud, which are Micro-reboot, Reliability Analyze and Load Analyze. The Microreboot service can reboot an appliance automatically when the appliance is shutdown unexpectedly. The Reliability Analyze service implements a SBRA-based algorithm to find out the key components which have crucial influence to the system. And the Load Analyze service can find the appliance with the highest workload.
With the help of SM@RT [
For the “create”, “delete”, “set” and “restart” management tasks, the execution time of Java programs is less than the QVT ones. The main reason is that the two sets of programs are based on the same management APIs and there are some extra operations in runtime model based approach, which ensure the synchronization between the runtime model and real system. However, the difference is small and acceptable.
For the “get ‘used-memory’ attribute” management tasks, the execution time of Java programs is longer than the QVT ones, but the data delay of QVT programs is longer than the Java ones. The reasons are two folds. On one hand, the Java programs query the attributes of appliances through directly invoking the management APIs, so the execution time increases linearly with the number of the appliances and the delay is little. On the other hand, the runtime model is equivalent to the snapshot of system metrics and getting the attributes of appliances just needs a reading operation, so the execution time of the QVT programs is shorter. The runtime model is synchronized with the running system with traversing all the metrics of the running system, so the data delay increases linearly with the size of the model. 4
In this paper, we demonstrated Jade Cloud, which is a runtime model based Cloud management system. Based on the runtime models of OpenStack and Hyperic, Jade Cloud can manage both the infrastructure and software resources in a unified manner without modifying existing systems. The work load and performance of Jade Cloud is also evaluated. For future work, we plan to add more model-based management functions and apply the approach in production environment.
ACKNOWLEDGMENT. This work is supported by the National High-Tech Research and Development Program of China (863) under Grant No. 2013AA01A208; the National Natural Science Foundation of China under Grant No. 61300002 and the China Postdoctoral Science Foundation under Grant No. 2013M530011