Overload Management Technology Based on Service Degradation


Overload Management Technology Based on Service Degradation

Participants

NameCompany/Institutee-mail
Ziyou WangPKUwangzy06 at sei.pku.edu.cn
Zhou JiangPKUzhoujiang at sei.pku.edu.cn

Actions

As the load of internet software often changes in an unpredictable way over time, such applications are often faced with overload. Overload is at least one resource requirement of the server, such as bandwidth, CPU, memory, etc., beyond the existing capacity of the system [Bianca 06]. For example, in a sudden short period of time, a large amount of customer requests can lead to system overload. When overload occurs, the software's response time will quickly grow to an unacceptable level; the same time as running out of computing resources, software does not control the behavior will couse even downtime and denial of service and other serious errors. For e-commerce applications, such a serious error will cause huge economic losses. Therefore, for a large number of internet softwares, prevention of overload is an important goal. So the service is still guaranteed, even when the number of requests exceed the system capacity several times.

Overload prevention refers to the system in the event of overload conditions can still offer services, even when the access request is serveal times over the capacity of the system [Guitart 10]. Therefore, for most internet softwares, the overload prevention is an important demand, and overload management techniques are achieved through a certain method to prevent overload. Currently, the most commonly used overload management technology, under normal circumstances in the application in excess of the excess computing resources to their needs, and sometimes even request the required resources is expected to peak times. However, this method still does not respond to requests for the case of the peak increases rapidly. On the other hand, such an approach dose not only cause a large number of long-term idle computing resources and waste, and the purchase, operation and maintenance of such a large-scale server clusters is also very expensive. 

In this case, we hope that gives a systematic approach for overload management, specific research objectives include:

  1. through monitoring network configuration software, the basic component elements of the system, sort out the different functions of the software parts occupancy of computing resources, real-time discovery of the causes overload and location; 
  2. to provide an automated service degradation mechanisms, components size on overload management, to reduce the workload of maintenance personnel and to avoid introducing artificial operational errors; 
  3. to provide good support for cluster applications, ensure that the method has good scalability; 
  4. construct a middleware platform, based on the results of the monitoring and analysis for the application of the need for provision of service degradation to achieve levels support. Based on these objectives, we propose a service-based automatic downgrade overload management.