How to manage iron in the data center using sound
Researchers at the University of St. Louis in the United States have developed the technology of music-defined networking, or MDN. It balances the load on the network by analyzing special sound signals that the data center equipment supplies. We tell how it works.
/ photo Dennis van Zuijlekom CC BY-SA
As noted Cisco analysts, the volume of data processed in data centers, growing at 25% annually. This trend leads to an increase in the number of servers in data centers and the complexity of network topology. It becomes more difficult for sysadmins to manage such systems and monitor their “health”.
For example, in November 2018, due to incorrect configuration of servers in the Facebook data center, the company's services, including WhatsApp and Instagram, worked intermittently for twelve hours. This affected not only ordinary users of applications, but also advertisers who launched campaigns on the eve of Black Friday.
To help data center operators monitor the state of the IT infrastructure, researchers from St. Louis University presented the music-defined networking technology. It independently controls the networks, focusing on the sound signals generated by the equipment in the engine room. The developers managed to apply it to load balancing, detect cooling system failures and detect unauthorized access.
MDN-system is a set of Raspberry Pi with speakers connected to network switches in the data center. Mini-computers make a special sound every time when one of the programmed events occurs on the network (for example, a large number of requests from an IP address).
Low-cost microphones are placed on the territory of the data center, which capture these sounds and transmit information about them to special controllers. And the controllers decide on further action.
At the same time, the system reacts not only to artificially created sounds, but also to changes in the background noise in the engine room. So, she was able to determine the failure in the cooling system of the server. When testing, the controller was able to detect fan failure on one of the machines and turned it off until the fault was fixed.
One of the main advantages of MDN-technology is cheapness. Although the researchers did not disclose the total cost of the system, they noted that each controller cost them about $ 80. According to the authors of the project, their solution is cheaper than other out-of-band monitoring systems . The latter often require the purchase of additional equipment and the construction of redundant communication lines (this is thousands of dollars of expenses).
To solve this problem, in the future, researchers plan to create a system of microphones that will “listen” not just one, but several routers and servers at once.
/ photo by William Zhang CC BY
Another advantage of an MDN system is its flexibility. The technology works with any devices and does not depend on the network topology, but so far it cannot be applied in large data centers. MDN handles the processing of thousands of sounds from different devices. This number is not enough for large-scale data centers, in which the number of servers can be tens of thousands.
In the world and develop other audio solutions for monitoring data center. For example, the University of Pennsylvania created a system that provides information for the data center administrators about the ports used, the source of traffic, and the sender's IP address.
The system has two modes of operation. The first is designed to monitor connections to specific ports on the network, for each of which the program uses a certain “buzzing” sound. This helps administrators quickly detect port scan attacks.
The second mode is needed to determine the sources of packets received by the network. The program makes a sound gongwhose volume increases with each receipt of a block of data from the same sender. The timbre of the sound depends on the geographical location of the user.
A similar technology was created at the University of Technology Ontario. A group of researchers suggested using well-known melodies , the composition of which changes with the growth of traffic or network connections. The idea is that network administrators notice a change in familiar music faster than in a random set of sounds, and begin to more quickly counteract threats.
Another solution was created by researchers from the University of Northumbria in the UK. The authors of the technology proposed to implement a monitoring system based onsounds of nature : the sound of rain, crickets or a running stream. Together, these sounds are perceived as a single “picture” and do not distract employees from their tasks. However, any changes in this “audiolandscape” will be noticeable for IT professionals who can respond to network problems.
According to experts, the sound in the data center is a source of useful information about networks, which is now practically not used.
For example, the potential of using sound management techniques advocated Cameron Turner (Cameron Turner), an expert on big data, who has worked with Microsoft and Stanford University. In his opinion, systems like MDN can be supplemented with machine learning algorithms, which will expand the capabilities of the technology. For example, it helps the data center to more efficiently use electricity.
Another direction in the development of sound technologies is the use of frequencies outside the human hearing range. This will expand the range of signals for the automatic control system, as well as reduce the noise level in the data center and the potential risk to employee health.
Posts from the corporate blog IT-GRAD:
Our telegram channel:
/ photo Dennis van Zuijlekom CC BY-SA
Why did you need MDN technology
As noted Cisco analysts, the volume of data processed in data centers, growing at 25% annually. This trend leads to an increase in the number of servers in data centers and the complexity of network topology. It becomes more difficult for sysadmins to manage such systems and monitor their “health”.
For example, in November 2018, due to incorrect configuration of servers in the Facebook data center, the company's services, including WhatsApp and Instagram, worked intermittently for twelve hours. This affected not only ordinary users of applications, but also advertisers who launched campaigns on the eve of Black Friday.
To help data center operators monitor the state of the IT infrastructure, researchers from St. Louis University presented the music-defined networking technology. It independently controls the networks, focusing on the sound signals generated by the equipment in the engine room. The developers managed to apply it to load balancing, detect cooling system failures and detect unauthorized access.
Principles of MDN
MDN-system is a set of Raspberry Pi with speakers connected to network switches in the data center. Mini-computers make a special sound every time when one of the programmed events occurs on the network (for example, a large number of requests from an IP address).
Low-cost microphones are placed on the territory of the data center, which capture these sounds and transmit information about them to special controllers. And the controllers decide on further action.
During the tests, researchers applied the technology for load balancing. Sounds of different pitch were used: the higher the frequency of the emitted signal, the more packets are waiting for the queue at the switch. As soon as the MDN controller “heard” that one of the devices was overloaded, it redistributed traffic to other routes of the network.
At the same time, the system reacts not only to artificially created sounds, but also to changes in the background noise in the engine room. So, she was able to determine the failure in the cooling system of the server. When testing, the controller was able to detect fan failure on one of the machines and turned it off until the fault was fixed.
Advantages and disadvantages
One of the main advantages of MDN-technology is cheapness. Although the researchers did not disclose the total cost of the system, they noted that each controller cost them about $ 80. According to the authors of the project, their solution is cheaper than other out-of-band monitoring systems . The latter often require the purchase of additional equipment and the construction of redundant communication lines (this is thousands of dollars of expenses).
To solve this problem, in the future, researchers plan to create a system of microphones that will “listen” not just one, but several routers and servers at once.
/ photo by William Zhang CC BY
Another advantage of an MDN system is its flexibility. The technology works with any devices and does not depend on the network topology, but so far it cannot be applied in large data centers. MDN handles the processing of thousands of sounds from different devices. This number is not enough for large-scale data centers, in which the number of servers can be tens of thousands.
Who else is developing "sound" technology for the data center
In the world and develop other audio solutions for monitoring data center. For example, the University of Pennsylvania created a system that provides information for the data center administrators about the ports used, the source of traffic, and the sender's IP address.
The system has two modes of operation. The first is designed to monitor connections to specific ports on the network, for each of which the program uses a certain “buzzing” sound. This helps administrators quickly detect port scan attacks.
The second mode is needed to determine the sources of packets received by the network. The program makes a sound gongwhose volume increases with each receipt of a block of data from the same sender. The timbre of the sound depends on the geographical location of the user.
A similar technology was created at the University of Technology Ontario. A group of researchers suggested using well-known melodies , the composition of which changes with the growth of traffic or network connections. The idea is that network administrators notice a change in familiar music faster than in a random set of sounds, and begin to more quickly counteract threats.
Another solution was created by researchers from the University of Northumbria in the UK. The authors of the technology proposed to implement a monitoring system based onsounds of nature : the sound of rain, crickets or a running stream. Together, these sounds are perceived as a single “picture” and do not distract employees from their tasks. However, any changes in this “audiolandscape” will be noticeable for IT professionals who can respond to network problems.
The future of sound in data centers
According to experts, the sound in the data center is a source of useful information about networks, which is now practically not used.
For example, the potential of using sound management techniques advocated Cameron Turner (Cameron Turner), an expert on big data, who has worked with Microsoft and Stanford University. In his opinion, systems like MDN can be supplemented with machine learning algorithms, which will expand the capabilities of the technology. For example, it helps the data center to more efficiently use electricity.
Another direction in the development of sound technologies is the use of frequencies outside the human hearing range. This will expand the range of signals for the automatic control system, as well as reduce the noise level in the data center and the potential risk to employee health.
Posts from the corporate blog IT-GRAD:
- How to place 100% of the infrastructure in the cloud IaaS-provider and not regret it
- "How are VMware?": A review of new solutions
- Servers for SAP: Basic Platforms
Our telegram channel: