Title：The Cost of Doing Science on the Cloud: The Montage Example

Title：The Cost of Doing Science on the Cloud: The Montage Example
Author：Deelman
Research Topics：
  1.Question 1: Assume that an application has a set of resources available to them but sometimes it needs more resources than it has, so it reaches out to the cloud from time to time to meet the dditional demands. In this case, the application will provision a set of resources from the cloud and bring in the data and stage the results back to a location where the user can access it.  The question is how many processors to provision in order to optimize application performance while minimizing the monetary cost. 
  2.Question 2: Assume that an application has very limited computational resources and wants to rely on the cloud resources to provide the necessary computing power. Also assume that the application provisions a certain amount of resources over a period
of time to sustain the expected computational load.  That set of resources requested is assumed to be larger then the needs of any single computation. Thus the requests can run at their full level of parallelism. Here the cost is measured only as the cost of the resources used by a single request.  We assume that the application would incur the cost of the resources over time and would need to decide how much to charge the user for a given request.  In this case we are only calculating the cost of this request to the application and not the premium the application may decide to charge on top of it.
 
Here we also examine two situations:
Question 2a:  Given that the application has a local data archive and just wants to farm out the computing, the question is, how much each user request will cost?  
 
Question 2b:  Assume that the application relies fully on both the compute and storage services on the cloud.  Here besides the cost of a particular request, we also determine how many requests it would take to make the cost of storing the data on the cloud worthwhile. The issue is that if you have only a few requests, the cost of storing large amounts of data over time can be prohibitively expensive, so for a small number of requests, it may be cheaper to stage data to the cloud on demand.
 
Question 3:  Finally, we answer a totally application-focused questions: 1) how much money it would cost to generate the mosaic of the entire sky?  This sky
mosaic based on 2Mass data can be created by combining 3,900 plates (mosaics) in three frequency bands, each of 4 degrees square; 2) if one calculates a
mosaic, how long does it make financial sense to store it on the cloud rather than recomputed it on demand.