InfoQ is running my recent 3-part article series on Azure grid computing with some expanded content.
Part 1 went up today, the other parts will run at 2-week intervals.
2 comments:
Anonymous
said...
I'm a new hire for a team at MSoft who desperately need multi-host load balancing. But the Azure computing model doesn't really understand data locality. I'll give you a call next week to ask about this.
What we want to do is compute a bunch of summary functions on a very large set of data in real time-sql is good at this on small data sets online and large data sets offline. Data warehousing is terrific for this offline.
Azure, like most cloud computing solutions, is dead in the water unless we write the entire infrastructure from scratch. What I really want is an STL for grid/cloud computing: apply a vector of metrics to a very large horizontally partitioned table.
I guess I need to hear more, but I'm not sure why Azure Grid wouldn't do the job for you. It only asks you to write application-specific code for task data loading, task execution, and result data aggregation. Your tasks could access any storage repository they want including a large horizontally partitioned table.
2 comments:
I'm a new hire for a team at MSoft who desperately need multi-host load balancing. But the Azure computing model doesn't really understand data locality. I'll give you a call next week to ask about this.
What we want to do is compute a bunch of summary functions on a very large set of data in real time-sql is good at this on small data sets online and large data sets offline. Data warehousing is terrific for this offline.
Azure, like most cloud computing solutions, is dead in the water unless we write the entire infrastructure from scratch. What I really want is an STL for grid/cloud computing: apply a vector of metrics to a very large horizontally partitioned table.
I guess I need to hear more, but I'm not sure why Azure Grid wouldn't do the job for you. It only asks you to write application-specific code for task data loading, task execution, and result data aggregation. Your tasks could access any storage repository they want including a large horizontally partitioned table.
Post a Comment