Opened 6 years ago

Last modified 4 years ago

#1578 new bug

Memory Leak in MPI_FILE_WRITE/READ_ALL?

Reported by: sverma@… Owned by:
Priority: major Milestone: future
Component: mpich Keywords: core dump, memory leak
Cc: robl

Description (last modified by balaji)

Dear Sir/Madam?,
I was profiling my code with Totalview's Memoryscape and I noticed that if MPI_FILE_WRITE/READ_ALL is called on a file more than once, it leads to a jump in the heap memory usage and the total virtual memory (VM) usage of just one core (core0). This results in my machine running out of VM when processing large datasets.

Is this behaviour expected, and if so could you please suggest a workaround?

Thank you,
Yours faithfully,
Siddhartha Verma.

Change History (5)

comment:1 Changed 6 years ago by thakur

  • Cc robl added

comment:2 Changed 6 years ago by balaji

  • Milestone set to mpich2-1.5
  • Owner set to robl
  • Status changed from new to assigned

comment:3 Changed 5 years ago by robl

I need some more information to be able to reproduce this. I ran a pretty simple collective I/O test against MPICH2's latest SVN where 8 processors wrote elements of a subarray to one part of a file, then wrote to a different part of the file. Valgrind did not find any problems.

I've never used Memoryscape. Is there a way to report the location of the leaked memory? Can you tell me about your test case a bit more?

It's possible the leak was fixed since the version you tried.

comment:4 Changed 5 years ago by balaji

  • Milestone changed from mpich2-1.5 to future

comment:5 Changed 4 years ago by balaji

  • Description modified (diff)
  • Owner robl deleted
  • Status changed from assigned to new
Note: See TracTickets for help on using tickets.