The Shenango system is planned to be used in data centers.
/ photo
According to one of the providers, data centers only 20-40% of available computing power. At high loads, this indicator . This distribution of resources leads to the emergence of so-called "zombie servers". These are machines that sit idle most of the time, wasting electricity. Today 30% of servers in the world , consuming $30 billion a year in electricity.
MIT decided to fight the inefficient use of computing resources.
Engineering team a processor load balancing system called Shenango. Its purpose is to monitor the state of the task buffer and redistribute "stuck" processes (which cannot get processor time) to free machines.
How Shenango works
Shenango is a Linux C library with Rust and C++ bindings. The project code and test applications are published in on GitHub.
The solution is based on the IOKernel algorithm, which runs on a dedicated core of a multiprocessor system. It manages requests to the CPU using a framework , which allows applications to communicate directly with network devices.
The IOKernel decides which cores to hand off a particular task to. The algorithm also decides how many cores are needed. For each process, the main cores (guaranteed) and additional (burstable) are defined - the latter are launched in case of a sharp increase in the number of requests to the CPU.
The IOKernel request queue is organized as . Every five microseconds, the algorithm checks to see if all the tasks assigned to the kernel have been completed. To do this, it compares the current location of the "head" of the buffer with the previous position of its "tail". If it turns out that the tail was already in the queue at the time of the previous check, the system notes the buffer overload and allocates an additional core for the process.
Load balancing prioritizes cores on which the same process has run before and is partially cached, or any idle cores.
Shenango additionally uses the approach . The cores allocated for the operation of one application monitor the number of tasks each other has. If one core finishes its list of tasks before the others, then it "removes" part of the load from its neighbors.
Advantages and disadvantages
On engineers at MIT, Shenango is capable of processing five million requests per second and maintaining an average response time of 37 microseconds. Experts say that in some cases the technology is able to increase the utilization of processors in data centers up to 100%. As a result, data center operators will be able to save on the purchase and maintenance of servers.
Solution Potential and specialists from other universities. According to a professor from the Korean Institute, the MIT system will help reduce delays in web services. For example, it is useful in the work of online stores. On sale days, even a second page load delay to reduce the number of site views by 11%. Rapid load distribution will help serve more customers.
The technology still has drawbacks - it does not support multiprocessor -systems in which the chips are connected to different memory modules and do not "communicate" with each other. In this case, IOKernel can regulate the operation of a particular group of processors, but not all server chips.
/ photo
Similar technologies
Among other load balancing systems for processors, Arachne can be distinguished. It calculates how many cores the application will need at the time of its launch, and distributes processes in accordance with this indicator. According to the authors, the maximum delay of the application in Arachne is about 10 thousand microseconds.
The technology is implemented as a C++ library for Linux, and its source code is available at .
Another balancing tool is ZygOS. Like Shenango, the technology uses the work stealing method to redistribute processes. According to the authors of ZygOS, the average application latency when using the tool is about 150 microseconds, and the maximum is about 450 microseconds. Project code also .
Conclusions
Modern data centers continue to expand, especially the upward trend is noticeable in the hyperscale data center market: now in the world 430 hyperscale data centers, but in the coming years their number may increase by 30%. For this reason, processor load balancing technologies will be in great demand. Systems like Shenango already now large corporations, and in the future the number of such tools will only grow.
Posts from the First Enterprise IaaS Blog:
Source: habr.com