"HDF5ERROR in aiori-HDF5.c (line 553): cannot create data set" error when benchmarking HDF5 in Blue Gene Q about ior

[morrone]

I'm using zlib/1.2.7 and hdf5/1.8.10, and compiling for the compute nodes (not front nodes).

You are one up on me; IOR with HDF5 won't even compile for me on BG/Q. :) Did you have to make any modifications to get it to build?

I'm using zlib/1.2.7 and hdf5/1.8.10, and compiling for the compute nodes (not front nodes).

Your screenshot says 1.8.8. Are you sure about that?

PS: As per contract with IBM, I was told by my line manager that we have to use to the version 2.10.3 of IOR... is this version fully compatible with Blue Gene Q? Any known issues?

Sorry, I haven't used that version of IOR in at least two years. Off the top of my head, I know I changed the way IOR gets the host name to identify tasks per node. I think that made IOR's output a little more sane. I cannot remember if anything was more clearly broken than that.

[brunomaga]

You are right, in my test I used 1.8.8 but I've just tested 1.8.10 and it returns the same error.

About the compilation process, see if this helps:

I didn't have to modify the source code, just past the parameters into the ./configure and make steps.
To be clear, I use modules to load the zlib and hdf5 dependencies.
For the compute nodes compilation, I loaded our compiler (wrapper/xl).
Here are the variables set by the modules:

HDF5
setenv HDF5_ROOT {bgsys}/local/hdf5/1.8.10
setenv HDF5_INCDIR {bgsys}/local/hdf5/1.8.10/include
setenv HDF5_LIBDIR {bgsys}/local/hdf5/1.8.10/lib
setenv HDF5_LINK_OPTS -L{bgsys}/local/hdf5/1.8.10/lib -lhdf5

ZLIB
module-whatis loads the zlib environment for BG/Q CNs
setenv ZLIB_ROOT {bgsys}/local/zlib/1.2.7
setenv ZLIB_INCDIR {bgsys}/local/zlib/1.2.7/include
setenv ZLIB_LIBDIR {bgsys}/local/zlib/1.2.7/lib
setenv ZLIB_LINK_OPTS -L{bgsys}/local/zlib/1.2.7/lib -lz

Compiler
module-whatis loads the BG/Q mpi xl wrappers environment
prepend-path PATH {bgsys}/drivers/ppcfloor/comm/xl/bin
prepend-path MANPATH {bgsys}/drivers/ppcfloor/comm/xl/man
setenv CC {bgsys}/drivers/ppcfloor/comm/xl/bin/mpixlc
setenv CXX {bgsys}/drivers/ppcfloor/comm/xl/bin/mpixlcxx
setenv FC {bgsys}/drivers/ppcfloor/comm/xl/bin/mpixlf90

In order to configure it for hdf5 building, I run:
./configure --with-hdf5=yes CFLAGS=-I$HDF5_INCDIR LDFLAGS="-L$HDF5_LIBDIR -L$ZLIB_LIBDIR" LIBS="-lm"

Then I just run make normally, and the make command would come up like this:
mpixlc -I/bgsys/local/hdf5/1.8.10/include -L/bgsys/local/hdf5/1.8.10/lib -L/bgsys/local/zlib/1.2.7/lib -o ior ior.o utilities.o parse_options.o aiori-POSIX.o aiori-MPIIO.o aiori-HDF5.o -lhdf5 -lz -lm

I noticed now that I get two warning in the compilation process:
"aiori-HDF5.c", line 551.30: 1506-098 (E) Missing argument(s).
"aiori-HDF5.c", line 555.34: 1506-098 (E) Missing argument(s).

This is the function where the execution dies, maybe its related to this issue?

Then when running, I get the following output:

HDF5-DIAG: Error detected in HDF5 (1.8.10) MPI-process 0:
#000: H5D.c line 152 in H5Dcreate2(): not link creation property list
major: Invalid arguments to routine
minor: Inappropriate type
.
IOR-3.0.1: MPI Coordinated Test of Parallel I/O

Began: Tue Jul 16 09:47:10 2013
Command line used: ./IOR/src/ior -a HDF5
Machine: CNK bgqio10-ib

Test 0 started: Tue Jul 16 09:47:10 2013
Summary:
api = HDF5-1.8.10 (Parallel)
test filename = testFile
access = single-shared-file
ordering in a file = sequential offsets
ordering inter file= no tasks offsets
clients = 1 (1 per node)
repetitions = 1
xfersize = 262144 bytes
blocksize = 1 MiB
aggregate filesize = 1 MiB

access bw(MiB/s) block(KiB) xfer(KiB) open(s) wr/rd(s) close(s) total(s) iter

---

** error **
ERROR in aiori-HDF5.c (line 553): cannot create data set.
** exiting **

Hope this helps

[morrone]

I noticed now that I get two warning in the compilation process:
"aiori-HDF5.c", line 551.30: 1506-098 (E) Missing argument(s).
"aiori-HDF5.c", line 555.34: 1506-098 (E) Missing argument(s).

This is the function where the execution dies, maybe its related to this issue?

Yes, I would very much suspect that to be the case. The "(E)" denotes: "Error conditions exist that the compiler can correct, but the program might not produce the expected results."

I am a bit surprised that the compiler doesn't halt, actually. I would dig into that error.

[brunomaga]

Ok thank you. I will start looking into that. Let me know if you need help with the BGQ compilation.

[roblatham00]

I too am surprised you could build this far.

"aiori-HDF5.c", line 555.34: 1506-098 (E) Missing argument(s).

means IOR is trying to use old-style HDF5 arguments. There is a backwards compatiblity option, though:

add -DH5_USE_16_API to your cflags.

Maybe IOR should just do that automatically. I shudder to think anyone is actually running HDF5-1.6.x, but even if they were that define would not affect them.

[roblatham00]

I guess some day I'll set up my github life so I can "send a pull request. Until then, I blat patches into the issue tracker:

From bf979035e87250970e022527d44bf745ff20cf87 Mon Sep 17 00:00:00 2001
From: Rob Latham <[email protected]>
Date: Thu, 18 Jul 2013 21:31:54 +0000
Subject: [PATCH] backwards compatiblity macro

Since aiori-HDF5 uses old 1.6-style function calls (create, open, etc),
turn on the backwards compatiblity here (the one place HDF5 calls are
made)
---
 src/aiori-HDF5.c |    3 +++
 1 files changed, 3 insertions(+), 0 deletions(-)

diff --git a/src/aiori-HDF5.c b/src/aiori-HDF5.c
index 2f222f0..7649be0 100644
--- a/src/aiori-HDF5.c
+++ b/src/aiori-HDF5.c
@@ -19,6 +19,9 @@
 #include <stdio.h>              /* only for fprintf() */
 #include <stdlib.h>
 #include <sys/stat.h>
+/* HDF5 routines here still use the old 1.6 style.  Nothing wrong with that but
+ * save users the trouble of  passing this flag through configure */
+#define H5_USE_16_API
 #include <hdf5.h>
 #include <mpi.h>

-- 
1.7.1

[morrone]

Thanks, Rob! I added that change to master.

[brunomaga]

It worked!! Thank you for you help Rob.