search

CODARcode / Example-Heat_Transfer

3 stars 3 forks source link

BP rmethod does not work? #5

Open ianfoster opened 7 years ago

ianfoster commented 7 years ago

Initial comment: The README in stage_write says that BP is a valid rmethod. But it doesn't seem to work (see below).

Follow on comment: I understand that MPI and BP allow for coupling via files. Let's make this example show how that is done.

./run-workflow.sh 4 swift: launching: ./heat_transfer_adios2 size: 6 swift: launching: stage_write/stage_write Input stream = heat.bp Output stream = staged.bp Read method = BP (id=0) Read method parameters = "max_chunk_size=100; app_id =32767; verbose= 3;poll_interval = 100;" Write method = MPI Write method parameters = "" Variable to transform = "" Transform parameters = ""

Waiting to open stream heat.bp... ERROR: MPI open failed for heat.bp: 'File does not exist, error stack: ADIOI_UFS_OPEN(69): File heat.bp does not exist' Process number : 1 x 1 Array size per process at first step: 100 x 100 Step 1: ERROR: MPI open failed for heat.bp: 'File does not exist, error stack: ADIOI_UFS_OPEN(69): File heat.bp does not exist' Writing: filename size(GB) io_time(sec) GB/s

jychoi-hpc commented 7 years ago

Although I created a version for file-based heat_transfer/stage_write coupling in #8, it looks like it can create more confusions to users. File-based coupling needs extra cares for execution orders, which is out of CoDAR's focuses, I think. We can use file-based coupling for debugging purposes to developers.

In this release, I think, we better focus on staging methods like FLEXPATH and DATASPACES. How do you think, @ianfoster ? Anyone, is there any other opinion?

ianfoster commented 7 years ago

This confuses me. Scott says that file based staging is a useful codesign option. IA that not true?

On Jul 19, 2017, at 3:11 PM, Jong Choi notifications@github.com wrote:

Although I created a version for file-based heat_transfer/stage_write coupling in #8, it looks like it can create more confusions to users. File-based coupling needs extra cares for execution orders, which is out of CoDAR's focuses, I think. We can use file-based coupling for debugging purposes to developers.

In this release, I think, we better focus on staging methods like FLEXPATH and DATASPACES. How do you think, @ianfoster ? Anyone, is there any other opinion?

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or mute the thread.

jychoi-hpc commented 7 years ago

I have just talked to Scott and discussed. I have convinced now that file-based coupling is worth supporting in this example. #8 supports file-based coupling now. I will integrate this branch into the main.