Add file opening, in order to detect sequential access and checksum during i/o #1
Labels
enhancement
New feature or request
speedup
Important performance improvement
wishlist
not a priority, but a long term wish...
Currently, the library is trigerred on file close, not open. The idea being that we want the final version of a file, after all writes have occurred. However, if we only start working on the post when the file is closed, then in order to calculate the checksum, we have to read the entire file again. It would be better if we could:
set a offset in the file to 0, and initialize the checksum, then as long as the writes are sequential, update the checksums as the writes go by, so that when the file is closed, the checksum is already available.
Saving a file read on large files is a huge win. On the current HPC mirroring use case, the client refused to use checksumming because of the overhead of the extra read. With this mod, the overhead should drop to essentially 0 for many cases.
The text was updated successfully, but these errors were encountered: