Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
T: Fachverband Teilchenphysik
T 66: Grid-Computing
T 66.3: Vortrag
Dienstag, 28. März 2017, 17:20–17:35, JUR 372
Data intensive workflows in the Cloud — Gen Kawamura1, Oliver Keeble2, Arnulf Quadt1, and •Gerhard Rzehorz1,2 — 1II. Physikalisches Institut, Georg-August Universität Göttingen — 2IT Department, CERN
Cloud computing in the Infrastructure as a Service case means to rent computing resources from commercial providers. For Monte-Carlo simulations, which record low data in- and output, running on the Cloud is mostly understood. This stems from the fact that no permanent storage is required and the network is not stressed. On the other hand, for data intensive workflows network and storage I/O can be crucial bottlenecks. Therefore, they are usually only computed on Grid sites that either have the required input data in their storage (fast interconnect) - or in a small number of cases the input is downloaded from other sites. This talk will provide answers on how to set up a Cloud to get the best workflow performance, addressing whether to implement a Cloud site with traditional Grid storage, whether to use object storage or even go storage-less altogether. Going further into detail, questions that will be answered are: How big is the influence of the computing to storage distance on the overall event throughput of the site? Can latency effects be mitigated by optimisation techniques? In order to assess this, the Workflow and Infrastructure Model with the output metric ETC = Events/Time/Cost is formed and applied. This output quantifies the different workflow and infrastructure configurations that are tested against each other in an understandable and comparable fashion.