ICStats said:
Sure but until we get some different figures, I'm going to assume my estimate is in the right ballpark.
I'm talking about a unit of processing. One Xeon core hour = a unit of compute. How it's divided doesn't matter in this argument. Could be 1 hour on 1 server, or 1 minute on 60 servers. They can't use as many as they feel like, only as many as they have. This is something you keep ignoring. When you have 1 million beans and 10 million users, you can't give 10 beans to each user. Even if Orleans technology could give 10 beans per user, you have only 0.1 beans to give. |
You are talking about a Unit which basically is a node in cloud terms. That type of setup is not the infrastructure that MS has built. Instead think of each node and all of its cores as one huge resource pool. Each task sent to the Azure cloud can leverage each individual core as a unit of processing not just the entire node. Each grain from the Orleans platform operate as a single threaded execution on one core using a small number of threads. For the X1 cloud compute you can have multiple X1s operating on one node each using a number of cores to process their individual grain or code.