Scott Wittenburg

fyi @zackgalbreath

fyi @pnorbert and @eisenhauer This is my strawman for data-size based aggregation. Over the next couple days I plan to comment in various locations in the changeset to ask questions...

With my current implementation, I have found a broken case when 3 of 5 ranks get an aggregator cache hit, while 2 of 5 do not. I'll try to describe...

> A quick question. Are you using SelectiveAggregationMetadata? Ah yes, great question, I assume you asked because you noticed the re-ordering between timesteps. I just confirmed that in this case,...

The aggregator rank in each mpi chain receives the final data position from the last rank in the chain. Is that because there is an assumption there that the aggregator...

By the way, on the `master` branch, when I `#define DUMPDATALOCINFO`, this is what it outputs for the first time step: Extra debug output: ``` 315: Flush count is :0...

> OK, try substituting in this code into the matching bits of WriteMetadataFileIndex() Thanks Greg, that works and jives with both the extra debug printing I was doing, as well...

> However, things obviously aren't right even with just a single timestep, Well, if I run the test with just a single timestep, everything looks fine and works. But when...

Ok thanks for that information, I'll start looking in `BP5Reader::ReadData()` as well, since I think it will be helpful for me to have at least some familiarity with the read...

> subfile 1 has zeroes up through Offset 12288, and that doesn't look right Huh, that is interesting. There is initialization stuff that gets done on every timestep now (used...