Microsoft and MIT develop mathematical model to improve cloud efficiency
Inspired by mathematical finance , the field of applied mathematics that Microsoft and Massachusetts Institute of Technology (MIT) analyze the stock market, jointly improve cloud service traffic efficiency and use cloud computing while maintaining availability Created a mathematical model to increase the rate.
TEAVAR: striking the right utilization-availability balance in WAN traffic engineering
Using Wall Street secrets to reduce the cost of cloud infrastructure
The cloud service is a network business in which data centers set up in cities in various regions are connected to each other via optical fiber lines to provide various computing services to customers. One of the important points of cloud services is “availability”, which is the “percentage of time the system can be used”.
What is the difference between reliability and availability? | Nikkei xTECH (Cross Tech)
If availability is low, you may not be able to use the service when you want to use it. However, even in large-scale computing systems, the system itself may go down due to an unexpected accident such as a power outage or line disconnection, so it is essential to back up the network and data center itself in order to increase availability. However, the backup network and data center are useless during normal times when there is no trouble, so increasing the number of backups reduces the overall availability. And the decline in availability will come at a cost.
Microsoft and MIT's research team uses network data from Microsoft, Google, IBM, AT & T, etc. to map the probability distribution of traffic failures transmitted from the data center, and find the appropriate bandwidth for each link. Developed a mathematical model to predict the occurrence of further failures. This mathematical model applies the existing mathematical model studied in mathematical finance to network traffic based on the assumption that traffic usage is `` investment amount '' in the stock market and network failure is `` loss ''. .
When this theory was tested with a simulator in an environment close to reality, it was confirmed that traffic obstruction was avoided while maximizing the transfer amount. According to the research team, the published model can increase the data transfer volume by about three times compared to the conventional network design without causing data loss.
The paper co-author, Magya Gobadi said, “When the network usage rate increases due to the mathematical model that we have announced, not only will the burden on the provider be reduced and the usage fee for the cloud service will be reduced, but there will be a huge amount of energy from the cloud infrastructure. “We can expect a positive impact on the environment by reducing consumption.”