Choose timezone
Your profile timezone:
Participants: Concezio, Andrea Sa., Andrea Sc., David, Jan, Pepe.
Jan: we should spell out the constraints in the model (e.g. coming from
the network).
David: interest of model is to extrapolate in 10 years rather than
reproducing current tools.
Important to evaluate trade-offs, e.g. between CPUs and GPUs.
Jan: network requirements will be very important, they have been neglected until now. Need to calculate bandwidths of workloads.
We need a data flow diagram and a graph based model to really understand what is needed.
Important to understand the effect of data and CPU location (i.e. remote vs. local data access).
When writing code we will have to balance having not too dirty code with the ability of having something in a reasonable time.
We need a site model.
Time granularity should be defined, and we should be able to estimate peaks in resource needs. In fact another trade-off is between the
ability to accommodate peaks and the amount of resources.
Andrea Sa.: we should really start writing code (not just discuss), Ken Bloom's framework should be looked into.
We need to calculate CPU, disk, IO (local and remote), which depend on the site model and where data is located.
For example, the impact of local vs remote access should be estimated in a range of scenarios, not just on a fixed one. The ratio between local
and remote should be a parameter.
Constraints at the network level should be in the Site class.
ACTION to Andrea Sa.: study Ken Bloom's framework and report.
ACTION to all: list all the questions that we want to answer.
ACTION to all (or volunteers?): draft a site model.
ACTION to Andrea Sc.: make a proposal to deal with time dependence:
should time be discrete? Or do we want to deal with generic functions of
time? Important to make a choice for the first implementation.
Pepe: for some parameters we definitely need functions (e.g. power
consumption).
Next meeting is scheduled for Friday 2nd of March at 14:00 CET.