Cluster Challenge Application sneak peek
Posted: 21 Aug 2015 | 11:09
Anna Chantzoplaki is working on the project “HPC Cluster Challenge app” with EPCC's Nick Brown and Amy Krause as part of the PRACE Summer of HPC Summer School. This project is developing a public outreach application that will allow participants to design their own computer cluster by choosing components within a fixed budget, configuring them and measuring performance.
Here Anna gives us a sneak preview of her work.
Since we are heading towards the end of our summer school I would like to give you a small preview of my project, the name of which is “ARCHER Challenge”, named after the UK National Supercomputing Service, located a few kilometres away from EPCC.
Game target. It will be a race-against-time kind of game. The target will be to run as many supercomputing jobs (eg simulation of Earth’s climate, extreme weather forecasting, aircraft research, etc) as possible within a specific time frame.
Game rules. The game has four levels, referring to HPC centres: “Local”, “City”, “Regional”, and “National”. The player will start the game at the “Local” level and with a predetermined amount of money. As time passes and more jobs run on the supercomputer, the player will be able to unlock better computer components that will make the supercomputer faster and more efficient, reaching higher levels and acquiring cool HPC badges!
Game description: For reasons of simplicity, there will be four available cabinets and each cabinet will have up to three nodes. The player will choose one cabinet and the button “Add node” will appear, with which they can either select one of the predetermined nodes or make a custom node. In case of custom node, the player will be guided through a wizard to select in each step a CPU, a memory, and an accelerator. The moment the player buys a node, the supercomputing jobs will start running, and they will earn some money every time a job is executed. There will also be an option to upgrade a node (ie buy more memory) or fix it, since in a supercomputing cluster the failure of nodes is a usual and expected thing!
Moreover, when the player reaches a higher level, the “Cabinet Settings” button will unlock. This option gives the player the chance to upgrade the cabinet’s cooling and interconnect systems. Since communication is an important part of a cluster, a better interconnect system results in better performance. And of course, a better cooling system keeps the nodes’ temperature lower leading to less failures.
The picture below gives a taste of how a user will be able to add a new node on the cabinets.