Parallel write with BP5 engine writes corrupted data exactly at index = maximum BufferChunkSize value #4596
Unanswered
Vassilis-Boubis
asked this question in
Q&A
Replies: 1 comment 9 replies
-
Most probably a bug, an integer overflow of a kind somewhere, this is the min value for an int32_t. |
Beta Was this translation helpful? Give feedback.
9 replies
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
-
I am trying to store a 5 GB dataset in parallel with 2 MPI processes using the BP5 engine.
When I do this with 1,4,8, or more processes, everything works as intended.
When using 2 processes, however, rank 0 writes its partition correctly, while rank 1 writes incorrect values starting at byte 2147381248 of its partition. I verified this by inspecting the contents of what was stored through the bpls command-line tool.
When checking the ADIOS 2 documentation, I noticed that the maximum BufferChunkSize is exactly 2147381248 bytes. Could this be related to my corrupted parallel write?
Any help would be appreciated, thank you!
Beta Was this translation helpful? Give feedback.
All reactions