Difference between revisions of "Shareholder model"
(→Shareholder share) |
|||
| Line 1: | Line 1: | ||
== Introduction == | == Introduction == | ||
| − | The clusters operated and maintained by the ID SIS HPC group are based on a shareholder model and are therefore financed by its users. On each HPC cluster, there is a share financed by the IT Services is open to all members of ETH at no cost (public share), while professors who financed nodes in an HPC cluster receive a share of CPU time proportional to their investment. | + | The clusters operated and maintained by the ID SIS HPC group are based on a '''shareholder model''' and are therefore financed by its users. On each HPC cluster, there is a share financed by the IT Services is open to all members of ETH at no cost ('''public share'''), while professors who financed nodes in an HPC cluster receive a share of CPU time proportional to their investment. |
NOTE: Shares on different clusters are treated separately, i.e., Brutus shareholders who did not finance nodes in Euler do not automatically have shareholder privileges on Euler or any other HPC cluster. | NOTE: Shares on different clusters are treated separately, i.e., Brutus shareholders who did not finance nodes in Euler do not automatically have shareholder privileges on Euler or any other HPC cluster. | ||
| Line 10: | Line 10: | ||
== Public share == | == Public share == | ||
| − | Resources in the public share are in contrary to shareholder resources not guaranteed. If the clusters are busy, jobs might have a longer waiting time in the queue. There are also certain limits in terms of maximum number of cores and maximal amount of memory that guest users can use at the same time. These limits might are subject to changes. | + | Resources in the public share are in contrary to shareholder resources '''not guaranteed'''. If the clusters are busy, jobs might have a longer waiting time in the queue. There are also certain limits in terms of maximum number of cores and maximal amount of memory that guest users can use at the same time. These limits might are subject to changes. |
== Becoming a shareholder == | == Becoming a shareholder == | ||
Professors, institutes or departments who would like to become shareholders of an HPC cluster have the possibility to invest at any time. ETH is not allowed to publish the prices on the web but Olivier Byrde will be happy to communicate them to prospective shareholders on demand. | Professors, institutes or departments who would like to become shareholders of an HPC cluster have the possibility to invest at any time. ETH is not allowed to publish the prices on the web but Olivier Byrde will be happy to communicate them to prospective shareholders on demand. | ||
Revision as of 14:42, 10 August 2016
Introduction
The clusters operated and maintained by the ID SIS HPC group are based on a shareholder model and are therefore financed by its users. On each HPC cluster, there is a share financed by the IT Services is open to all members of ETH at no cost (public share), while professors who financed nodes in an HPC cluster receive a share of CPU time proportional to their investment.
NOTE: Shares on different clusters are treated separately, i.e., Brutus shareholders who did not finance nodes in Euler do not automatically have shareholder privileges on Euler or any other HPC cluster.
Shareholders receive on average a guaranteed share of resources, proportional to their investment. Instead of physical resources (compute nodes), the shareholders buy an equivalent in CPU time.
Resources in the public share are in contrary to shareholder resources not guaranteed. If the clusters are busy, jobs might have a longer waiting time in the queue. There are also certain limits in terms of maximum number of cores and maximal amount of memory that guest users can use at the same time. These limits might are subject to changes.
Professors, institutes or departments who would like to become shareholders of an HPC cluster have the possibility to invest at any time. ETH is not allowed to publish the prices on the web but Olivier Byrde will be happy to communicate them to prospective shareholders on demand.
