 ICACT20210154 Slide.01
[Big Slide]
|
Chrome  Click!! |
.jpg) |
Hi everybody, my name is Yi Li. For ICACT 2021, I have cooperated with four partners from Northeastern University. Dr. Jun Guo, Mrs. Chen Liu, Mr. Zixuan Zhao, Dr. Bin Zhang. Our presentation today is about a research on evaluation method of virtual machine quantity in cold and hot operation mode based on reliability guarantee |
| ICACT20210154 Slide.02
[Big Slide]
|
Chrome Click!! |
.jpg) |
Reliability is an essential guarantee for the normal operation of the system. A minor mistake will arouse a major failure of the software system and lead to resource damage and business interruption. To keep high reliability of the Could system, in terms of cloud infrastructure, redundant backup is often used to provide one or more back-up virtual machines for related virtual machines (VM). This requires that the resource management system can adjust the dynamic resources when the resource configuration of Cloud service does not match the user concurrency, to keep the SLA penalty cost low while providing services that meet user expectations constantly. There are also some deficiencies in the current study: |
| ICACT20210154 Slide.03
[Big Slide]
|
Chrome Click!! |
.jpg) |
Traditional redundant backup replaces the failed VM with back-up VM directly to meet the system reliability requirements, without considering the energy consumption problem. |
| ICACT20210154 Slide.04
[Big Slide]
|
Chrome Click!! |
.jpg) |
The existing methods use webull function to reflect the failure rate while ignoring the relevance between service concurrency and service failure, as well as the impact of software aging on service failure. Moreover, the reliability calculation is more complicated in existing methods. The complexity of both time and space is high. |
| ICACT20210154 Slide.05
[Big Slide]
|
Chrome Click!! |
.jpg) |
All these problems will result in low reliability of the current system. Therefore, this paper raises a reliability evaluation of cold and hot backup system based on multiple-valued decision diagram algorithm. When designing the evaluation interval, this paper proposes a method to evaluate the system reliability from T to T + 1 period by separating the prediction time interval, which can effectively improve the accuracy of the evaluation method. According to current service concurrency, the quantity of VM in operation mode is adjusted. Under this premise, the number of cold and hot modes should be reasonably allocated to ensure the reliability of the whole system on the basis of meeting the reliability requirement R. |
| ICACT20210154 Slide.06
[Big Slide]
|
Chrome Click!! |
.jpg) |
First of all, HSMM is established to train the parameters of the model by using the historical detection data of virtual machine, and make it reflect the running characteristics of virtual machine correctly; by monitoring the running state information of VM in real-time, a time series reflecting the state change of VM is formed. According to this time series, the future running state of VM and the failure rate in the next moment can be predicted. |
| ICACT20210154 Slide.07
[Big Slide]
|
Chrome Click!! |
.jpg) |
The running state of each VM in the Cloud service system plays a decisive role in the reliability of the whole cloud system. Based on the predicted failure rate of VM in operation mode and hot backup mode, MDD is built for operating mode virtual machine and hot backup mode virtual organization. Then MDD of the backup system is established. Finally, through the evaluation of the constructed MDD, the reliability of the backup system can be acquired. |
| ICACT20210154 Slide.08
[Big Slide]
|
Chrome Click!! |
.jpg) |
This paper designs an evaluation method of VM in hot and cold operation mode based on reliability guarantee. The specific flow chart is shown in Figure |
| ICACT20210154 Slide.09
[Big Slide]
|
Chrome Click!! |
.jpg) |
This figure is evaluation process of VM quantity in cold and hot operation mode based on reliability guarantee |
| ICACT20210154 Slide.10
[Big Slide]
|
Chrome Click!! |
.jpg) |
To ensure the reliability of cloud system within the range of user requirements, according to the performance requirements and the current service concurrency, this paper adjusts the VM quantity in operation mode and adjusts the VM quantity in cold and hot backup mode on the basis of meeting the reliability requirement R.
First of all, the parameters need to be initialized. R_δ refers to the upper limit of reliability qualification, R_ε refers to the floor level of reliability qualification; p_ε refers to the limitation of failure efficiency; k refers to the quantity of VM in operation mode m1; M_o [k] refers to the failure rate matrix at the next moment of VM k in the collection of OM; VM_"cuvo " ={VM_1,VM_2,L,VM_k} represents the collection of the current system in operating mode; VM_"cuve " ={VM_cl,VM_C2,L,VM_"cil " } represents the collection of the current system in hot backup mode. |
| ICACT20210154 Slide.11
[Big Slide]
|
Chrome Click!! |
.jpg) |
In line 2, it calculates and sorts the VM in operation mode. In lines 3 to 4, it means to shut down a VM in hot mode if the reliability of the current system goes beyond the limit. In lines 5 to 11, if the current system reliability is in the normal range, it shall consider whether the VM failure rate in a single operation mode is too high. If it is, then a VM in hot mode shall be turned into operation mode and the VM in operation mode shall be turned into the cold mode to update the collection elements of various modes. In lines 12 to 29, it refers that the reliability of current system is smaller than the lower limiting value. Line 14 is to save the current set. Line 16 is to add a VM in the hot mode to the operation mode. Line 17 is to turn the VM in operation mode with the highest failure rate into the cold mode. Line 18 is to put a VM in cold mode in the collection of the hot mode. In lines 19 to 23, system reliability is reevaluated. If it meets the requirements, the collection elements are updated and the loop ends. In lines 25 to 28, it refers to no increase in VM quantity in hot mode. Because the transfer between modes cannot meet the reliability requirements, it is necessary to add a hot mode virtual machine, restore the collection, and add a cold mode virtual machine to the hot mode to update the collection. |
| ICACT20210154 Slide.12
[Big Slide]
|
Chrome Click!! |
.jpg) |
A set of default experimental parameter values are shown in Table 1. To better indicate the characteristics of various methods, the loadrunner was adopted to compress the backup system in three periods, including 0~5000 requests per second under simulated pressurization for 0~4h, 5000~10000 requests per second under simulated pressurization for 5~8h, and 1000~10000 requests per second under simulated pressurization for8~12h. By taking the adjustment method as an independent variable, three experiments are carried out in each period. For the first time, comparison method 1 is used; for the second time, comparison method 2 is used; for the third time, the proposed method is used. Finally, the performance of each method is compared by the average request failure rate and the average response time. |
| ICACT20210154 Slide.13
[Big Slide]
|
Chrome Click!! |
.jpg) |
Three groups of comparative experiments are designed and completed, including comparison method 1 (all operating mode), comparison method 2 (traditional backup mode), and the proposed method (the adjusting method of VM quantity in cold and hot operation mode based on reliability guarantee). The experimental results are as follows |
| ICACT20210154 Slide.14
[Big Slide]
|
Chrome Click!! |
.jpg) |
In table 2, the average request failure rate and the average response time under three adjusting methods are recorded. Because the proposed method adjusts VM quantity in the operation mode according to the concurrency amount at intervals and fully considers the failure of each working VM, the MDD is built to obtain the system reliability. The proposed method adjusts the operating mode virtual machine by reliability judgment so the average performance of the experimental results is the best with 0.016% of the average request failure rate and 0.108s of the average response time; the average response time of comparison method 1 is the longest for all VM resources are in operation mode; the average request failure rate and response time of comparison method 2 are lower than the proposed method. The failure rate of the VM is predicted in the proposed method. When the VM failure rate in operation mode is high, the VM is converted to the cold backup mode and the operation mode is replaced by the hot backup mode, which ensures the higher reliability of VM in operation mode. Comparison method 2 only replaces the operating system when the operation mode fails and does not evaluate the reliability of the system. Therefore, the reliability of comparison method 2 is weaker than that of the proposed method. |
| ICACT20210154 Slide.15
[Big Slide]
|
Chrome Click!! |
.jpg) |
It can be seen from the figures that various kind of indexes are generally increasing within 12 hours by using the three methods. With the increase of running time, the system reliability decreases. The reliability of the proposed method has little change, which proves the correctness of the method from another respect. |
| ICACT20210154 Slide.16
[Big Slide]
|
Chrome Click!! |
.jpg) |
In this paper, the evaluation method of VM quantity in cold and hot operation mode based on reliability guarantee is put forward to dynamically adjust the VM with operational mode failure. Through the three parallel experiments, the effectiveness of this method is verified. It not only guarantees the reliability of the system but also simplifies the calculating complexity. While considering the reliability of a Cloud computing service system, this paper mainly takes the average request rate into account. In future work, the interruption time and other index parameters shall be considered to be more suitable for the actual environment of Cloud computing. |
| ICACT20210154 Slide.17
[Big Slide]
|
Chrome Click!! |
.jpg) |
Thanks for your attention! |
| ICACT20210154 Slide.18
[Big Slide]
|
Chrome Click!! |
.jpg) |
|
