Energy-efficient management of computer clusters
Energy-efficient management of computer clusters
Background
The University of Oldenburg - like many other institutions - is struggling with a sharp rise in energy costs. Among other things, this is due to the increased use of computers that run around the clock
. One way of counteracting this is to switch off computers when they are not in use.
In addition to the option of logging on and off again every day, more modern computers also offer the option of freezing
the status of the computer, including all applications, and saving it via suspend-to-RAM or suspend-to-disk, so that energy consumption can be drastically reduced. However, the computer is then unavailable for as long as it is in this idle state. However, many computers can also be restarted remotely via the network using a magic packet.
On computer clusters in particular, however, the problem arises that several users may be working remotely on one computer. In this case, the user working directly on this computer should of course not switch it off. This is one of the reasons why many computers run constantly and consume energy unnecessarily.
In the server sector, computer virtualisation is increasingly being used to reduce costs. Many energy-hungry
computers are replaced by a small number of powerful computers. Virtualisation solutions such as VMware, XEN, VirtualPC, VServer and KVM make it possible to assign different applications, including the operating system, to different virtual machines. This offers several advantages: 1) Each application can be run on its own virtual machine so as not to jeopardise other applications in the event of an error. 2) Applications can be installed and tested on a trial basis before they go into productive operation unchanged. 3) In the event of a hardware defect or peak loads, individual virtual machines can be easily migrated to another physical computer.
Virtualisation can also be used on computers that are used directly by several people in dialogue mode: if they leave their workstation computer, their active applications are temporarily stored on another computer so that their workstation computer can then be switched off. When they return to their original (or even another) computer, they can continue working with their applications there. Sun Ray technology and the classic X terminals already allow people to use their applications from different computers, but the applications themselves are executed on less powerful servers. However, it can make a lot of sense to utilise the resources of the local computer, especially for computing-intensive applications.
Job description
The aim of this project is to develop a solution for Linux computer clusters that allows parts of a computer cluster to be separated and switched off during operation. If computing time is required or at the user's request, the switched-off computers should then be restarted and added to the computer cluster again.
The focus of the investigation is on dialogue-oriented applications in particular: for example, all applications that are still running on one computer in the evening are to be migrated to another computer in order to hibernate
there non-interactively.
In addition to developing a viable concept, this will also be implemented as a prototype on the computer cluster of the System Software and Distributed Systems department.